diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-100/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-100/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-100/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-100/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-100/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..7ce4b072fcbe5b636867e13937c56fedee3b68b4
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-100/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.02728394274790723,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-100/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-100/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-100/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-100/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-100/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-100/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-100/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-100/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..ecc036e87349a639d0327c993d0322670cb2d0ba
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-100/trainer_state.json
@@ -0,0 +1,139 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.24140012070006034,
+  "eval_steps": 20,
+  "global_step": 100,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.024795526266098,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.945375919342041,
+      "learning_rate": 1.6698127428345936e-05,
+      "loss": 1.7825420379638672,
+      "mean_token_accuracy": 0.6315708436071873,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.29921497588747,
+      "eval_loss": 1.169131875038147,
+      "eval_mean_token_accuracy": 0.7185706342204233,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 91.7898,
+      "eval_samples_per_second": 15.47,
+      "eval_steps_per_second": 1.939,
+      "step": 20
+    },
+    {
+      "entropy": 1.0346344470977784,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.7916979789733887,
+      "learning_rate": 3.427510366871008e-05,
+      "loss": 0.9480395317077637,
+      "mean_token_accuracy": 0.7543429024517536,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 0.9673695430326997,
+      "eval_loss": 0.8588430285453796,
+      "eval_mean_token_accuracy": 0.7737233002534073,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 89.8219,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 40
+    },
+    {
+      "entropy": 0.8977566614747048,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7972187995910645,
+      "learning_rate": 5.185207990907422e-05,
+      "loss": 0.8029604911804199,
+      "mean_token_accuracy": 0.7818016350269318,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8436836982041263,
+      "eval_loss": 0.7893033623695374,
+      "eval_mean_token_accuracy": 0.7858914984076211,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 89.9349,
+      "eval_samples_per_second": 15.789,
+      "eval_steps_per_second": 1.979,
+      "step": 60
+    },
+    {
+      "entropy": 0.8650321319699288,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.2128729820251465,
+      "learning_rate": 6.942905614943836e-05,
+      "loss": 0.77417893409729,
+      "mean_token_accuracy": 0.7875479467213153,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8209857180547179,
+      "eval_loss": 0.7572280168533325,
+      "eval_mean_token_accuracy": 0.7925970383574453,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 89.9459,
+      "eval_samples_per_second": 15.787,
+      "eval_steps_per_second": 1.979,
+      "step": 80
+    },
+    {
+      "entropy": 0.8096660938113928,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.1254287958145142,
+      "learning_rate": 8.700603238980251e-05,
+      "loss": 0.7170331001281738,
+      "mean_token_accuracy": 0.7994555257260799,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.792780208788561,
+      "eval_loss": 0.7442984580993652,
+      "eval_mean_token_accuracy": 0.7954704573984896,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 90.294,
+      "eval_samples_per_second": 15.726,
+      "eval_steps_per_second": 1.971,
+      "step": 100
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.0127318221037568e+16,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1000/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1000/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1000/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1000/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1000/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..7ce4b072fcbe5b636867e13937c56fedee3b68b4
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1000/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.02728394274790723,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1000/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1000/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1000/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1000/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1000/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1000/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1000/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1000/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..e10c9c9fe4bbaf8b65e834d596218c06404ac1d3
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1000/trainer_state.json
@@ -0,0 +1,1084 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 2.4103802051901027,
+  "eval_steps": 20,
+  "global_step": 1000,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.024795526266098,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.945375919342041,
+      "learning_rate": 1.6698127428345936e-05,
+      "loss": 1.7825420379638672,
+      "mean_token_accuracy": 0.6315708436071873,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.29921497588747,
+      "eval_loss": 1.169131875038147,
+      "eval_mean_token_accuracy": 0.7185706342204233,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 91.7898,
+      "eval_samples_per_second": 15.47,
+      "eval_steps_per_second": 1.939,
+      "step": 20
+    },
+    {
+      "entropy": 1.0346344470977784,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.7916979789733887,
+      "learning_rate": 3.427510366871008e-05,
+      "loss": 0.9480395317077637,
+      "mean_token_accuracy": 0.7543429024517536,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 0.9673695430326997,
+      "eval_loss": 0.8588430285453796,
+      "eval_mean_token_accuracy": 0.7737233002534073,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 89.8219,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 40
+    },
+    {
+      "entropy": 0.8977566614747048,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7972187995910645,
+      "learning_rate": 5.185207990907422e-05,
+      "loss": 0.8029604911804199,
+      "mean_token_accuracy": 0.7818016350269318,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8436836982041263,
+      "eval_loss": 0.7893033623695374,
+      "eval_mean_token_accuracy": 0.7858914984076211,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 89.9349,
+      "eval_samples_per_second": 15.789,
+      "eval_steps_per_second": 1.979,
+      "step": 60
+    },
+    {
+      "entropy": 0.8650321319699288,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.2128729820251465,
+      "learning_rate": 6.942905614943836e-05,
+      "loss": 0.77417893409729,
+      "mean_token_accuracy": 0.7875479467213153,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8209857180547179,
+      "eval_loss": 0.7572280168533325,
+      "eval_mean_token_accuracy": 0.7925970383574453,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 89.9459,
+      "eval_samples_per_second": 15.787,
+      "eval_steps_per_second": 1.979,
+      "step": 80
+    },
+    {
+      "entropy": 0.8096660938113928,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.1254287958145142,
+      "learning_rate": 8.700603238980251e-05,
+      "loss": 0.7170331001281738,
+      "mean_token_accuracy": 0.7994555257260799,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.792780208788561,
+      "eval_loss": 0.7442984580993652,
+      "eval_mean_token_accuracy": 0.7954704573984896,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 90.294,
+      "eval_samples_per_second": 15.726,
+      "eval_steps_per_second": 1.971,
+      "step": 100
+    },
+    {
+      "entropy": 0.7979829199612141,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2140507698059082,
+      "learning_rate": 0.00010458300863016666,
+      "loss": 0.7141861915588379,
+      "mean_token_accuracy": 0.8000014744699001,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.7938692486018277,
+      "eval_loss": 0.7279797196388245,
+      "eval_mean_token_accuracy": 0.7989798734027348,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 89.7189,
+      "eval_samples_per_second": 15.827,
+      "eval_steps_per_second": 1.984,
+      "step": 120
+    },
+    {
+      "entropy": 0.7916347607970238,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.0506458282470703,
+      "learning_rate": 0.0001221599848705308,
+      "loss": 0.7117055416107178,
+      "mean_token_accuracy": 0.7970046654343605,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.8052781063519167,
+      "eval_loss": 0.722576916217804,
+      "eval_mean_token_accuracy": 0.7982530523551984,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 90.0376,
+      "eval_samples_per_second": 15.771,
+      "eval_steps_per_second": 1.977,
+      "step": 140
+    },
+    {
+      "entropy": 0.7880666613578796,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8409207463264465,
+      "learning_rate": 0.00013973696111089492,
+      "loss": 0.7052523136138916,
+      "mean_token_accuracy": 0.8000123649835587,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7787443134891853,
+      "eval_loss": 0.7137772440910339,
+      "eval_mean_token_accuracy": 0.8014209287220173,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 90.2562,
+      "eval_samples_per_second": 15.733,
+      "eval_steps_per_second": 1.972,
+      "step": 160
+    },
+    {
+      "entropy": 0.7696007996797561,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.0182477235794067,
+      "learning_rate": 0.00015731393735125907,
+      "loss": 0.7008797645568847,
+      "mean_token_accuracy": 0.8007259473204613,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.752259682068664,
+      "eval_loss": 0.7171286344528198,
+      "eval_mean_token_accuracy": 0.80052122574174,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 90.0797,
+      "eval_samples_per_second": 15.764,
+      "eval_steps_per_second": 1.976,
+      "step": 180
+    },
+    {
+      "entropy": 0.7828126326203346,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9553632736206055,
+      "learning_rate": 0.0001748909135916232,
+      "loss": 0.7046853542327881,
+      "mean_token_accuracy": 0.7988562889397144,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7389638169427936,
+      "eval_loss": 0.7187935709953308,
+      "eval_mean_token_accuracy": 0.7995899999409579,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 89.9223,
+      "eval_samples_per_second": 15.791,
+      "eval_steps_per_second": 1.979,
+      "step": 200
+    },
+    {
+      "entropy": 0.7909286297857762,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 1.134459376335144,
+      "learning_rate": 0.00019246788983198735,
+      "loss": 0.6989312171936035,
+      "mean_token_accuracy": 0.7991565138101577,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7496965199374082,
+      "eval_loss": 0.7106321454048157,
+      "eval_mean_token_accuracy": 0.8046183442131857,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 89.6899,
+      "eval_samples_per_second": 15.832,
+      "eval_steps_per_second": 1.985,
+      "step": 220
+    },
+    {
+      "entropy": 0.7651963323354721,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.9557585716247559,
+      "learning_rate": 0.0002100448660723515,
+      "loss": 0.6984559059143066,
+      "mean_token_accuracy": 0.8019823372364044,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7615563079212488,
+      "eval_loss": 0.7098539471626282,
+      "eval_mean_token_accuracy": 0.8047233508544022,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 90.3752,
+      "eval_samples_per_second": 15.712,
+      "eval_steps_per_second": 1.97,
+      "step": 240
+    },
+    {
+      "entropy": 0.787507726252079,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.9279311299324036,
+      "learning_rate": 0.00022762184231271568,
+      "loss": 0.708138370513916,
+      "mean_token_accuracy": 0.8013308703899383,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7856640896100676,
+      "eval_loss": 0.7191570401191711,
+      "eval_mean_token_accuracy": 0.8024677249153008,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 93.7336,
+      "eval_samples_per_second": 15.149,
+      "eval_steps_per_second": 1.899,
+      "step": 260
+    },
+    {
+      "entropy": 0.768489234894514,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.7992776036262512,
+      "learning_rate": 0.0002451988185530798,
+      "loss": 0.7035849571228028,
+      "mean_token_accuracy": 0.8031105332076549,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7682498740346244,
+      "eval_loss": 0.7149022221565247,
+      "eval_mean_token_accuracy": 0.7973497995499814,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 89.7309,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 280
+    },
+    {
+      "entropy": 0.7720955617725849,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.1931514739990234,
+      "learning_rate": 0.0002627757947934439,
+      "loss": 0.6965402603149414,
+      "mean_token_accuracy": 0.8040348663926125,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7214708304807042,
+      "eval_loss": 0.7350823879241943,
+      "eval_mean_token_accuracy": 0.7942911925610532,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 89.6451,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 300
+    },
+    {
+      "entropy": 0.7930095426738262,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8994467258453369,
+      "learning_rate": 0.0002803527710338081,
+      "loss": 0.7331492900848389,
+      "mean_token_accuracy": 0.796185302734375,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.8142535488926963,
+      "eval_loss": 0.7237834334373474,
+      "eval_mean_token_accuracy": 0.799495273761535,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 90.6759,
+      "eval_samples_per_second": 15.66,
+      "eval_steps_per_second": 1.963,
+      "step": 320
+    },
+    {
+      "entropy": 0.7825102441012859,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 1.8224239349365234,
+      "learning_rate": 0.00029792974727417223,
+      "loss": 0.7331175804138184,
+      "mean_token_accuracy": 0.7977667152881622,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7327135443017724,
+      "eval_loss": 0.7388784289360046,
+      "eval_mean_token_accuracy": 0.8007041252730938,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 89.7692,
+      "eval_samples_per_second": 15.818,
+      "eval_steps_per_second": 1.983,
+      "step": 340
+    },
+    {
+      "entropy": 0.819331557303667,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 1.0407779216766357,
+      "learning_rate": 0.0003155067235145364,
+      "loss": 0.7456695556640625,
+      "mean_token_accuracy": 0.7936980701982975,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.8065585420372781,
+      "eval_loss": 0.735519528388977,
+      "eval_mean_token_accuracy": 0.7974206296245704,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 89.4336,
+      "eval_samples_per_second": 15.878,
+      "eval_steps_per_second": 1.99,
+      "step": 360
+    },
+    {
+      "entropy": 0.8114653021097183,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.1209490299224854,
+      "learning_rate": 0.0003330836997549005,
+      "loss": 0.7340899467468261,
+      "mean_token_accuracy": 0.7914272703230381,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.8025527285056168,
+      "eval_loss": 0.7295576930046082,
+      "eval_mean_token_accuracy": 0.7990576628218876,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 89.9579,
+      "eval_samples_per_second": 15.785,
+      "eval_steps_per_second": 1.979,
+      "step": 380
+    },
+    {
+      "entropy": 0.8042497783899307,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 1.2437798976898193,
+      "learning_rate": 0.0003506606759952647,
+      "loss": 0.7381744384765625,
+      "mean_token_accuracy": 0.7967362694442273,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7717011336530193,
+      "eval_loss": 0.7436173558235168,
+      "eval_mean_token_accuracy": 0.7975849482450592,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 89.8411,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 400
+    },
+    {
+      "entropy": 0.7987750010056929,
+      "epoch": 1.012070006035003,
+      "grad_norm": 1.2390918731689453,
+      "learning_rate": 0.000364721224843344,
+      "loss": 0.7317790031433106,
+      "mean_token_accuracy": 0.7986087551364651,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7333170604170038,
+      "eval_loss": 0.7456948757171631,
+      "eval_mean_token_accuracy": 0.7964897577682238,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 89.7288,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 420
+    },
+    {
+      "entropy": 0.7372976846992969,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.1552740335464478,
+      "learning_rate": 0.00036468510102269575,
+      "loss": 0.6734249114990234,
+      "mean_token_accuracy": 0.805834824591875,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.7130355526891987,
+      "eval_loss": 0.7555333375930786,
+      "eval_mean_token_accuracy": 0.7961652781186479,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 89.8713,
+      "eval_samples_per_second": 15.8,
+      "eval_steps_per_second": 1.981,
+      "step": 440
+    },
+    {
+      "entropy": 0.7465668715536594,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 1.1991469860076904,
+      "learning_rate": 0.0003645973816745066,
+      "loss": 0.6871879577636719,
+      "mean_token_accuracy": 0.8042450234293937,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7772979542110743,
+      "eval_loss": 0.7561826705932617,
+      "eval_mean_token_accuracy": 0.7953876427720102,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 89.8395,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 460
+    },
+    {
+      "entropy": 0.7477855801582336,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 1.325486660003662,
+      "learning_rate": 0.00036445809162231435,
+      "loss": 0.6836830139160156,
+      "mean_token_accuracy": 0.8021619468927383,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.7098110676481483,
+      "eval_loss": 0.758007287979126,
+      "eval_mean_token_accuracy": 0.7909953189030122,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 89.6101,
+      "eval_samples_per_second": 15.846,
+      "eval_steps_per_second": 1.986,
+      "step": 480
+    },
+    {
+      "entropy": 0.7632691070437432,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 1.0122262239456177,
+      "learning_rate": 0.0003642672702835562,
+      "loss": 0.7071213245391845,
+      "mean_token_accuracy": 0.8000680930912495,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7645382720432924,
+      "eval_loss": 0.7492606043815613,
+      "eval_mean_token_accuracy": 0.7976858676149604,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 89.6092,
+      "eval_samples_per_second": 15.847,
+      "eval_steps_per_second": 1.986,
+      "step": 500
+    },
+    {
+      "entropy": 0.7623726457357407,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.367774486541748,
+      "learning_rate": 0.00036402497165841384,
+      "loss": 0.6965175151824952,
+      "mean_token_accuracy": 0.8016501650214195,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7868083715438843,
+      "eval_loss": 0.7481449842453003,
+      "eval_mean_token_accuracy": 0.799330582779445,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 89.8241,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 520
+    },
+    {
+      "entropy": 0.7535207405686378,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 1.213614821434021,
+      "learning_rate": 0.00036373126431453207,
+      "loss": 0.7078310489654541,
+      "mean_token_accuracy": 0.7998907208442688,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7829091950748743,
+      "eval_loss": 0.7416072487831116,
+      "eval_mean_token_accuracy": 0.7969146250339036,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 89.4607,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 540
+    },
+    {
+      "entropy": 0.7527363449335098,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 1.0929982662200928,
+      "learning_rate": 0.00036338623136761495,
+      "loss": 0.7058369636535644,
+      "mean_token_accuracy": 0.7997087433934211,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.728419938114252,
+      "eval_loss": 0.744873046875,
+      "eval_mean_token_accuracy": 0.7966634296299366,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 89.6597,
+      "eval_samples_per_second": 15.838,
+      "eval_steps_per_second": 1.985,
+      "step": 560
+    },
+    {
+      "entropy": 0.7406099259853363,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.409364104270935,
+      "learning_rate": 0.00036298997045790515,
+      "loss": 0.6902852058410645,
+      "mean_token_accuracy": 0.805386807769537,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.6991632758231645,
+      "eval_loss": 0.7480319738388062,
+      "eval_mean_token_accuracy": 0.7970068006033308,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 89.8329,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 580
+    },
+    {
+      "entropy": 0.7459449704736472,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.7053111791610718,
+      "learning_rate": 0.00036254259372255277,
+      "loss": 0.7044640064239502,
+      "mean_token_accuracy": 0.8005018755793571,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7454566660891758,
+      "eval_loss": 0.7551484704017639,
+      "eval_mean_token_accuracy": 0.7897093731365846,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 89.8313,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 600
+    },
+    {
+      "entropy": 0.7621391348540782,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 1.174443244934082,
+      "learning_rate": 0.000362044227763882,
+      "loss": 0.7111573696136475,
+      "mean_token_accuracy": 0.798752411454916,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7664387949397055,
+      "eval_loss": 0.7365804314613342,
+      "eval_mean_token_accuracy": 0.7963101200843126,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 89.9623,
+      "eval_samples_per_second": 15.784,
+      "eval_steps_per_second": 1.979,
+      "step": 620
+    },
+    {
+      "entropy": 0.7424944408237935,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.2898184061050415,
+      "learning_rate": 0.000361495013613564,
+      "loss": 0.6913110733032226,
+      "mean_token_accuracy": 0.8009015507996082,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7255479529332579,
+      "eval_loss": 0.7345991730690002,
+      "eval_mean_token_accuracy": 0.8013592945056015,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 89.817,
+      "eval_samples_per_second": 15.81,
+      "eval_steps_per_second": 1.982,
+      "step": 640
+    },
+    {
+      "entropy": 0.7303367637097835,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.1310476064682007,
+      "learning_rate": 0.00036089510669270666,
+      "loss": 0.6875617027282714,
+      "mean_token_accuracy": 0.8035397171974182,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7101308248016271,
+      "eval_loss": 0.7394917011260986,
+      "eval_mean_token_accuracy": 0.7990792942850777,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 89.7431,
+      "eval_samples_per_second": 15.823,
+      "eval_steps_per_second": 1.983,
+      "step": 660
+    },
+    {
+      "entropy": 0.7387678384780884,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.9645389914512634,
+      "learning_rate": 0.0003602446767678725,
+      "loss": 0.6911204338073731,
+      "mean_token_accuracy": 0.8056333385407924,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7281073737010527,
+      "eval_loss": 0.7376708984375,
+      "eval_mean_token_accuracy": 0.7997356415464637,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 90.0338,
+      "eval_samples_per_second": 15.772,
+      "eval_steps_per_second": 1.977,
+      "step": 680
+    },
+    {
+      "entropy": 0.7449554048478604,
+      "epoch": 1.687990343995172,
+      "grad_norm": 1.9069629907608032,
+      "learning_rate": 0.0003595439079030364,
+      "loss": 0.7042286396026611,
+      "mean_token_accuracy": 0.8018639378249646,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7022633817088738,
+      "eval_loss": 0.7320939898490906,
+      "eval_mean_token_accuracy": 0.8006051541044471,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 89.8948,
+      "eval_samples_per_second": 15.796,
+      "eval_steps_per_second": 1.98,
+      "step": 700
+    },
+    {
+      "entropy": 0.723249051719904,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.8333877325057983,
+      "learning_rate": 0.00035879299840749777,
+      "loss": 0.7043714046478271,
+      "mean_token_accuracy": 0.8028345607221127,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.672436946898364,
+      "eval_loss": 0.713233470916748,
+      "eval_mean_token_accuracy": 0.8047760577684038,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 89.6394,
+      "eval_samples_per_second": 15.841,
+      "eval_steps_per_second": 1.986,
+      "step": 720
+    },
+    {
+      "entropy": 0.730014206469059,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.4580037593841553,
+      "learning_rate": 0.00035799216077976137,
+      "loss": 0.6985176563262939,
+      "mean_token_accuracy": 0.8015547141432762,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.6908873795123582,
+      "eval_loss": 0.7198202013969421,
+      "eval_mean_token_accuracy": 0.8024131682481659,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 89.3771,
+      "eval_samples_per_second": 15.888,
+      "eval_steps_per_second": 1.992,
+      "step": 740
+    },
+    {
+      "entropy": 0.7219605796039105,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.51682710647583,
+      "learning_rate": 0.000357141621647403,
+      "loss": 0.6844424724578857,
+      "mean_token_accuracy": 0.8032813094556331,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7113401109582922,
+      "eval_loss": 0.7195309400558472,
+      "eval_mean_token_accuracy": 0.8019187035185568,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 89.4658,
+      "eval_samples_per_second": 15.872,
+      "eval_steps_per_second": 1.99,
+      "step": 760
+    },
+    {
+      "entropy": 0.729826743900776,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 1.0838000774383545,
+      "learning_rate": 0.0003562416217029361,
+      "loss": 0.685858678817749,
+      "mean_token_accuracy": 0.8050567395985126,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.649370267484965,
+      "eval_loss": 0.7177029252052307,
+      "eval_mean_token_accuracy": 0.8050174043419656,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 89.2287,
+      "eval_samples_per_second": 15.914,
+      "eval_steps_per_second": 1.995,
+      "step": 780
+    },
+    {
+      "entropy": 0.6799364153295755,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 1.1364924907684326,
+      "learning_rate": 0.00035529241563569903,
+      "loss": 0.6714428901672364,
+      "mean_token_accuracy": 0.8087341658771038,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6603799047095052,
+      "eval_loss": 0.7201809287071228,
+      "eval_mean_token_accuracy": 0.802863868769635,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 89.2718,
+      "eval_samples_per_second": 15.906,
+      "eval_steps_per_second": 1.994,
+      "step": 800
+    },
+    {
+      "entropy": 0.7078610375523567,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 1.2191942930221558,
+      "learning_rate": 0.0003542942720597808,
+      "loss": 0.6996590614318847,
+      "mean_token_accuracy": 0.8049512051045895,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.7240545264120852,
+      "eval_loss": 0.7301309108734131,
+      "eval_mean_token_accuracy": 0.8015973166133581,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 89.3438,
+      "eval_samples_per_second": 15.894,
+      "eval_steps_per_second": 1.992,
+      "step": 820
+    },
+    {
+      "entropy": 0.6557391556826505,
+      "epoch": 2.024140012070006,
+      "grad_norm": 1.36203134059906,
+      "learning_rate": 0.0003532474734380064,
+      "loss": 0.6327113628387451,
+      "mean_token_accuracy": 0.8142829847026181,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6573808648613062,
+      "eval_loss": 0.7334365844726562,
+      "eval_mean_token_accuracy": 0.8043633893634496,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 89.3359,
+      "eval_samples_per_second": 15.895,
+      "eval_steps_per_second": 1.992,
+      "step": 840
+    },
+    {
+      "entropy": 0.6409128502011299,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 1.5804955959320068,
+      "learning_rate": 0.0003521523160020035,
+      "loss": 0.6023559093475341,
+      "mean_token_accuracy": 0.8222251623868942,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6983234788594621,
+      "eval_loss": 0.7232135534286499,
+      "eval_mean_token_accuracy": 0.8010654268639811,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 89.2615,
+      "eval_samples_per_second": 15.908,
+      "eval_steps_per_second": 1.994,
+      "step": 860
+    },
+    {
+      "entropy": 0.6151149723678827,
+      "epoch": 2.12070006035003,
+      "grad_norm": 1.1975436210632324,
+      "learning_rate": 0.00035100910966837193,
+      "loss": 0.5803564548492431,
+      "mean_token_accuracy": 0.8256325736641884,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6636555573243773,
+      "eval_loss": 0.7205662727355957,
+      "eval_mean_token_accuracy": 0.8036768369460374,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 89.2957,
+      "eval_samples_per_second": 15.902,
+      "eval_steps_per_second": 1.993,
+      "step": 880
+    },
+    {
+      "entropy": 0.6201727617532015,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9907131195068359,
+      "learning_rate": 0.0003498181779509813,
+      "loss": 0.6101872444152832,
+      "mean_token_accuracy": 0.8196512959897518,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.6473788086617931,
+      "eval_loss": 0.7062397003173828,
+      "eval_mean_token_accuracy": 0.8078362184963869,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 88.6198,
+      "eval_samples_per_second": 16.024,
+      "eval_steps_per_second": 2.009,
+      "step": 900
+    },
+    {
+      "entropy": 0.6097332935780286,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.9055633544921875,
+      "learning_rate": 0.00034857985786942026,
+      "loss": 0.5956094264984131,
+      "mean_token_accuracy": 0.8240575887262821,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.612629868341296,
+      "eval_loss": 0.7106770277023315,
+      "eval_mean_token_accuracy": 0.805324277181304,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 89.4383,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 920
+    },
+    {
+      "entropy": 0.6182701248675585,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.9513736367225647,
+      "learning_rate": 0.00034729449985362404,
+      "loss": 0.6019269466400147,
+      "mean_token_accuracy": 0.8187106013298034,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6437647456533453,
+      "eval_loss": 0.7173502445220947,
+      "eval_mean_token_accuracy": 0.8056439573175451,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.9568,
+      "eval_samples_per_second": 15.963,
+      "eval_steps_per_second": 2.001,
+      "step": 940
+    },
+    {
+      "entropy": 0.6508570514619351,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.9606207013130188,
+      "learning_rate": 0.0003459624676447067,
+      "loss": 0.6296409606933594,
+      "mean_token_accuracy": 0.8167732007801533,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6288247135248077,
+      "eval_loss": 0.7160675525665283,
+      "eval_mean_token_accuracy": 0.8066010689467527,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 89.0044,
+      "eval_samples_per_second": 15.954,
+      "eval_steps_per_second": 2.0,
+      "step": 960
+    },
+    {
+      "entropy": 0.6380081418901682,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.9677181839942932,
+      "learning_rate": 0.000344584138192027,
+      "loss": 0.6210546493530273,
+      "mean_token_accuracy": 0.8186135642230511,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6407805031604981,
+      "eval_loss": 0.7055043578147888,
+      "eval_mean_token_accuracy": 0.8070103157772107,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 89.5606,
+      "eval_samples_per_second": 15.855,
+      "eval_steps_per_second": 1.987,
+      "step": 980
+    },
+    {
+      "entropy": 0.6310251638293266,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 2.1046407222747803,
+      "learning_rate": 0.000343159901546516,
+      "loss": 0.6165830135345459,
+      "mean_token_accuracy": 0.8207855716347694,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6461507470420237,
+      "eval_loss": 0.7292136549949646,
+      "eval_mean_token_accuracy": 0.8030418261383356,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 89.315,
+      "eval_samples_per_second": 15.899,
+      "eval_steps_per_second": 1.993,
+      "step": 1000
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.0031683194745037e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1020/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1020/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1020/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1020/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1020/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..7ce4b072fcbe5b636867e13937c56fedee3b68b4
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1020/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.02728394274790723,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1020/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1020/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1020/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1020/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1020/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1020/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1020/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1020/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..f329bbea390a413dc66bf985259ab8dab4b275b2
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1020/trainer_state.json
@@ -0,0 +1,1105 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 2.4586602293301145,
+  "eval_steps": 20,
+  "global_step": 1020,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.024795526266098,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.945375919342041,
+      "learning_rate": 1.6698127428345936e-05,
+      "loss": 1.7825420379638672,
+      "mean_token_accuracy": 0.6315708436071873,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.29921497588747,
+      "eval_loss": 1.169131875038147,
+      "eval_mean_token_accuracy": 0.7185706342204233,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 91.7898,
+      "eval_samples_per_second": 15.47,
+      "eval_steps_per_second": 1.939,
+      "step": 20
+    },
+    {
+      "entropy": 1.0346344470977784,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.7916979789733887,
+      "learning_rate": 3.427510366871008e-05,
+      "loss": 0.9480395317077637,
+      "mean_token_accuracy": 0.7543429024517536,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 0.9673695430326997,
+      "eval_loss": 0.8588430285453796,
+      "eval_mean_token_accuracy": 0.7737233002534073,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 89.8219,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 40
+    },
+    {
+      "entropy": 0.8977566614747048,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7972187995910645,
+      "learning_rate": 5.185207990907422e-05,
+      "loss": 0.8029604911804199,
+      "mean_token_accuracy": 0.7818016350269318,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8436836982041263,
+      "eval_loss": 0.7893033623695374,
+      "eval_mean_token_accuracy": 0.7858914984076211,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 89.9349,
+      "eval_samples_per_second": 15.789,
+      "eval_steps_per_second": 1.979,
+      "step": 60
+    },
+    {
+      "entropy": 0.8650321319699288,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.2128729820251465,
+      "learning_rate": 6.942905614943836e-05,
+      "loss": 0.77417893409729,
+      "mean_token_accuracy": 0.7875479467213153,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8209857180547179,
+      "eval_loss": 0.7572280168533325,
+      "eval_mean_token_accuracy": 0.7925970383574453,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 89.9459,
+      "eval_samples_per_second": 15.787,
+      "eval_steps_per_second": 1.979,
+      "step": 80
+    },
+    {
+      "entropy": 0.8096660938113928,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.1254287958145142,
+      "learning_rate": 8.700603238980251e-05,
+      "loss": 0.7170331001281738,
+      "mean_token_accuracy": 0.7994555257260799,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.792780208788561,
+      "eval_loss": 0.7442984580993652,
+      "eval_mean_token_accuracy": 0.7954704573984896,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 90.294,
+      "eval_samples_per_second": 15.726,
+      "eval_steps_per_second": 1.971,
+      "step": 100
+    },
+    {
+      "entropy": 0.7979829199612141,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2140507698059082,
+      "learning_rate": 0.00010458300863016666,
+      "loss": 0.7141861915588379,
+      "mean_token_accuracy": 0.8000014744699001,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.7938692486018277,
+      "eval_loss": 0.7279797196388245,
+      "eval_mean_token_accuracy": 0.7989798734027348,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 89.7189,
+      "eval_samples_per_second": 15.827,
+      "eval_steps_per_second": 1.984,
+      "step": 120
+    },
+    {
+      "entropy": 0.7916347607970238,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.0506458282470703,
+      "learning_rate": 0.0001221599848705308,
+      "loss": 0.7117055416107178,
+      "mean_token_accuracy": 0.7970046654343605,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.8052781063519167,
+      "eval_loss": 0.722576916217804,
+      "eval_mean_token_accuracy": 0.7982530523551984,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 90.0376,
+      "eval_samples_per_second": 15.771,
+      "eval_steps_per_second": 1.977,
+      "step": 140
+    },
+    {
+      "entropy": 0.7880666613578796,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8409207463264465,
+      "learning_rate": 0.00013973696111089492,
+      "loss": 0.7052523136138916,
+      "mean_token_accuracy": 0.8000123649835587,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7787443134891853,
+      "eval_loss": 0.7137772440910339,
+      "eval_mean_token_accuracy": 0.8014209287220173,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 90.2562,
+      "eval_samples_per_second": 15.733,
+      "eval_steps_per_second": 1.972,
+      "step": 160
+    },
+    {
+      "entropy": 0.7696007996797561,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.0182477235794067,
+      "learning_rate": 0.00015731393735125907,
+      "loss": 0.7008797645568847,
+      "mean_token_accuracy": 0.8007259473204613,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.752259682068664,
+      "eval_loss": 0.7171286344528198,
+      "eval_mean_token_accuracy": 0.80052122574174,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 90.0797,
+      "eval_samples_per_second": 15.764,
+      "eval_steps_per_second": 1.976,
+      "step": 180
+    },
+    {
+      "entropy": 0.7828126326203346,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9553632736206055,
+      "learning_rate": 0.0001748909135916232,
+      "loss": 0.7046853542327881,
+      "mean_token_accuracy": 0.7988562889397144,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7389638169427936,
+      "eval_loss": 0.7187935709953308,
+      "eval_mean_token_accuracy": 0.7995899999409579,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 89.9223,
+      "eval_samples_per_second": 15.791,
+      "eval_steps_per_second": 1.979,
+      "step": 200
+    },
+    {
+      "entropy": 0.7909286297857762,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 1.134459376335144,
+      "learning_rate": 0.00019246788983198735,
+      "loss": 0.6989312171936035,
+      "mean_token_accuracy": 0.7991565138101577,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7496965199374082,
+      "eval_loss": 0.7106321454048157,
+      "eval_mean_token_accuracy": 0.8046183442131857,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 89.6899,
+      "eval_samples_per_second": 15.832,
+      "eval_steps_per_second": 1.985,
+      "step": 220
+    },
+    {
+      "entropy": 0.7651963323354721,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.9557585716247559,
+      "learning_rate": 0.0002100448660723515,
+      "loss": 0.6984559059143066,
+      "mean_token_accuracy": 0.8019823372364044,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7615563079212488,
+      "eval_loss": 0.7098539471626282,
+      "eval_mean_token_accuracy": 0.8047233508544022,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 90.3752,
+      "eval_samples_per_second": 15.712,
+      "eval_steps_per_second": 1.97,
+      "step": 240
+    },
+    {
+      "entropy": 0.787507726252079,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.9279311299324036,
+      "learning_rate": 0.00022762184231271568,
+      "loss": 0.708138370513916,
+      "mean_token_accuracy": 0.8013308703899383,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7856640896100676,
+      "eval_loss": 0.7191570401191711,
+      "eval_mean_token_accuracy": 0.8024677249153008,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 93.7336,
+      "eval_samples_per_second": 15.149,
+      "eval_steps_per_second": 1.899,
+      "step": 260
+    },
+    {
+      "entropy": 0.768489234894514,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.7992776036262512,
+      "learning_rate": 0.0002451988185530798,
+      "loss": 0.7035849571228028,
+      "mean_token_accuracy": 0.8031105332076549,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7682498740346244,
+      "eval_loss": 0.7149022221565247,
+      "eval_mean_token_accuracy": 0.7973497995499814,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 89.7309,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 280
+    },
+    {
+      "entropy": 0.7720955617725849,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.1931514739990234,
+      "learning_rate": 0.0002627757947934439,
+      "loss": 0.6965402603149414,
+      "mean_token_accuracy": 0.8040348663926125,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7214708304807042,
+      "eval_loss": 0.7350823879241943,
+      "eval_mean_token_accuracy": 0.7942911925610532,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 89.6451,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 300
+    },
+    {
+      "entropy": 0.7930095426738262,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8994467258453369,
+      "learning_rate": 0.0002803527710338081,
+      "loss": 0.7331492900848389,
+      "mean_token_accuracy": 0.796185302734375,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.8142535488926963,
+      "eval_loss": 0.7237834334373474,
+      "eval_mean_token_accuracy": 0.799495273761535,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 90.6759,
+      "eval_samples_per_second": 15.66,
+      "eval_steps_per_second": 1.963,
+      "step": 320
+    },
+    {
+      "entropy": 0.7825102441012859,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 1.8224239349365234,
+      "learning_rate": 0.00029792974727417223,
+      "loss": 0.7331175804138184,
+      "mean_token_accuracy": 0.7977667152881622,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7327135443017724,
+      "eval_loss": 0.7388784289360046,
+      "eval_mean_token_accuracy": 0.8007041252730938,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 89.7692,
+      "eval_samples_per_second": 15.818,
+      "eval_steps_per_second": 1.983,
+      "step": 340
+    },
+    {
+      "entropy": 0.819331557303667,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 1.0407779216766357,
+      "learning_rate": 0.0003155067235145364,
+      "loss": 0.7456695556640625,
+      "mean_token_accuracy": 0.7936980701982975,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.8065585420372781,
+      "eval_loss": 0.735519528388977,
+      "eval_mean_token_accuracy": 0.7974206296245704,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 89.4336,
+      "eval_samples_per_second": 15.878,
+      "eval_steps_per_second": 1.99,
+      "step": 360
+    },
+    {
+      "entropy": 0.8114653021097183,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.1209490299224854,
+      "learning_rate": 0.0003330836997549005,
+      "loss": 0.7340899467468261,
+      "mean_token_accuracy": 0.7914272703230381,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.8025527285056168,
+      "eval_loss": 0.7295576930046082,
+      "eval_mean_token_accuracy": 0.7990576628218876,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 89.9579,
+      "eval_samples_per_second": 15.785,
+      "eval_steps_per_second": 1.979,
+      "step": 380
+    },
+    {
+      "entropy": 0.8042497783899307,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 1.2437798976898193,
+      "learning_rate": 0.0003506606759952647,
+      "loss": 0.7381744384765625,
+      "mean_token_accuracy": 0.7967362694442273,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7717011336530193,
+      "eval_loss": 0.7436173558235168,
+      "eval_mean_token_accuracy": 0.7975849482450592,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 89.8411,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 400
+    },
+    {
+      "entropy": 0.7987750010056929,
+      "epoch": 1.012070006035003,
+      "grad_norm": 1.2390918731689453,
+      "learning_rate": 0.000364721224843344,
+      "loss": 0.7317790031433106,
+      "mean_token_accuracy": 0.7986087551364651,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7333170604170038,
+      "eval_loss": 0.7456948757171631,
+      "eval_mean_token_accuracy": 0.7964897577682238,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 89.7288,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 420
+    },
+    {
+      "entropy": 0.7372976846992969,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.1552740335464478,
+      "learning_rate": 0.00036468510102269575,
+      "loss": 0.6734249114990234,
+      "mean_token_accuracy": 0.805834824591875,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.7130355526891987,
+      "eval_loss": 0.7555333375930786,
+      "eval_mean_token_accuracy": 0.7961652781186479,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 89.8713,
+      "eval_samples_per_second": 15.8,
+      "eval_steps_per_second": 1.981,
+      "step": 440
+    },
+    {
+      "entropy": 0.7465668715536594,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 1.1991469860076904,
+      "learning_rate": 0.0003645973816745066,
+      "loss": 0.6871879577636719,
+      "mean_token_accuracy": 0.8042450234293937,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7772979542110743,
+      "eval_loss": 0.7561826705932617,
+      "eval_mean_token_accuracy": 0.7953876427720102,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 89.8395,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 460
+    },
+    {
+      "entropy": 0.7477855801582336,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 1.325486660003662,
+      "learning_rate": 0.00036445809162231435,
+      "loss": 0.6836830139160156,
+      "mean_token_accuracy": 0.8021619468927383,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.7098110676481483,
+      "eval_loss": 0.758007287979126,
+      "eval_mean_token_accuracy": 0.7909953189030122,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 89.6101,
+      "eval_samples_per_second": 15.846,
+      "eval_steps_per_second": 1.986,
+      "step": 480
+    },
+    {
+      "entropy": 0.7632691070437432,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 1.0122262239456177,
+      "learning_rate": 0.0003642672702835562,
+      "loss": 0.7071213245391845,
+      "mean_token_accuracy": 0.8000680930912495,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7645382720432924,
+      "eval_loss": 0.7492606043815613,
+      "eval_mean_token_accuracy": 0.7976858676149604,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 89.6092,
+      "eval_samples_per_second": 15.847,
+      "eval_steps_per_second": 1.986,
+      "step": 500
+    },
+    {
+      "entropy": 0.7623726457357407,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.367774486541748,
+      "learning_rate": 0.00036402497165841384,
+      "loss": 0.6965175151824952,
+      "mean_token_accuracy": 0.8016501650214195,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7868083715438843,
+      "eval_loss": 0.7481449842453003,
+      "eval_mean_token_accuracy": 0.799330582779445,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 89.8241,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 520
+    },
+    {
+      "entropy": 0.7535207405686378,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 1.213614821434021,
+      "learning_rate": 0.00036373126431453207,
+      "loss": 0.7078310489654541,
+      "mean_token_accuracy": 0.7998907208442688,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7829091950748743,
+      "eval_loss": 0.7416072487831116,
+      "eval_mean_token_accuracy": 0.7969146250339036,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 89.4607,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 540
+    },
+    {
+      "entropy": 0.7527363449335098,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 1.0929982662200928,
+      "learning_rate": 0.00036338623136761495,
+      "loss": 0.7058369636535644,
+      "mean_token_accuracy": 0.7997087433934211,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.728419938114252,
+      "eval_loss": 0.744873046875,
+      "eval_mean_token_accuracy": 0.7966634296299366,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 89.6597,
+      "eval_samples_per_second": 15.838,
+      "eval_steps_per_second": 1.985,
+      "step": 560
+    },
+    {
+      "entropy": 0.7406099259853363,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.409364104270935,
+      "learning_rate": 0.00036298997045790515,
+      "loss": 0.6902852058410645,
+      "mean_token_accuracy": 0.805386807769537,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.6991632758231645,
+      "eval_loss": 0.7480319738388062,
+      "eval_mean_token_accuracy": 0.7970068006033308,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 89.8329,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 580
+    },
+    {
+      "entropy": 0.7459449704736472,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.7053111791610718,
+      "learning_rate": 0.00036254259372255277,
+      "loss": 0.7044640064239502,
+      "mean_token_accuracy": 0.8005018755793571,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7454566660891758,
+      "eval_loss": 0.7551484704017639,
+      "eval_mean_token_accuracy": 0.7897093731365846,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 89.8313,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 600
+    },
+    {
+      "entropy": 0.7621391348540782,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 1.174443244934082,
+      "learning_rate": 0.000362044227763882,
+      "loss": 0.7111573696136475,
+      "mean_token_accuracy": 0.798752411454916,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7664387949397055,
+      "eval_loss": 0.7365804314613342,
+      "eval_mean_token_accuracy": 0.7963101200843126,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 89.9623,
+      "eval_samples_per_second": 15.784,
+      "eval_steps_per_second": 1.979,
+      "step": 620
+    },
+    {
+      "entropy": 0.7424944408237935,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.2898184061050415,
+      "learning_rate": 0.000361495013613564,
+      "loss": 0.6913110733032226,
+      "mean_token_accuracy": 0.8009015507996082,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7255479529332579,
+      "eval_loss": 0.7345991730690002,
+      "eval_mean_token_accuracy": 0.8013592945056015,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 89.817,
+      "eval_samples_per_second": 15.81,
+      "eval_steps_per_second": 1.982,
+      "step": 640
+    },
+    {
+      "entropy": 0.7303367637097835,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.1310476064682007,
+      "learning_rate": 0.00036089510669270666,
+      "loss": 0.6875617027282714,
+      "mean_token_accuracy": 0.8035397171974182,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7101308248016271,
+      "eval_loss": 0.7394917011260986,
+      "eval_mean_token_accuracy": 0.7990792942850777,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 89.7431,
+      "eval_samples_per_second": 15.823,
+      "eval_steps_per_second": 1.983,
+      "step": 660
+    },
+    {
+      "entropy": 0.7387678384780884,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.9645389914512634,
+      "learning_rate": 0.0003602446767678725,
+      "loss": 0.6911204338073731,
+      "mean_token_accuracy": 0.8056333385407924,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7281073737010527,
+      "eval_loss": 0.7376708984375,
+      "eval_mean_token_accuracy": 0.7997356415464637,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 90.0338,
+      "eval_samples_per_second": 15.772,
+      "eval_steps_per_second": 1.977,
+      "step": 680
+    },
+    {
+      "entropy": 0.7449554048478604,
+      "epoch": 1.687990343995172,
+      "grad_norm": 1.9069629907608032,
+      "learning_rate": 0.0003595439079030364,
+      "loss": 0.7042286396026611,
+      "mean_token_accuracy": 0.8018639378249646,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7022633817088738,
+      "eval_loss": 0.7320939898490906,
+      "eval_mean_token_accuracy": 0.8006051541044471,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 89.8948,
+      "eval_samples_per_second": 15.796,
+      "eval_steps_per_second": 1.98,
+      "step": 700
+    },
+    {
+      "entropy": 0.723249051719904,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.8333877325057983,
+      "learning_rate": 0.00035879299840749777,
+      "loss": 0.7043714046478271,
+      "mean_token_accuracy": 0.8028345607221127,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.672436946898364,
+      "eval_loss": 0.713233470916748,
+      "eval_mean_token_accuracy": 0.8047760577684038,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 89.6394,
+      "eval_samples_per_second": 15.841,
+      "eval_steps_per_second": 1.986,
+      "step": 720
+    },
+    {
+      "entropy": 0.730014206469059,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.4580037593841553,
+      "learning_rate": 0.00035799216077976137,
+      "loss": 0.6985176563262939,
+      "mean_token_accuracy": 0.8015547141432762,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.6908873795123582,
+      "eval_loss": 0.7198202013969421,
+      "eval_mean_token_accuracy": 0.8024131682481659,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 89.3771,
+      "eval_samples_per_second": 15.888,
+      "eval_steps_per_second": 1.992,
+      "step": 740
+    },
+    {
+      "entropy": 0.7219605796039105,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.51682710647583,
+      "learning_rate": 0.000357141621647403,
+      "loss": 0.6844424724578857,
+      "mean_token_accuracy": 0.8032813094556331,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7113401109582922,
+      "eval_loss": 0.7195309400558472,
+      "eval_mean_token_accuracy": 0.8019187035185568,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 89.4658,
+      "eval_samples_per_second": 15.872,
+      "eval_steps_per_second": 1.99,
+      "step": 760
+    },
+    {
+      "entropy": 0.729826743900776,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 1.0838000774383545,
+      "learning_rate": 0.0003562416217029361,
+      "loss": 0.685858678817749,
+      "mean_token_accuracy": 0.8050567395985126,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.649370267484965,
+      "eval_loss": 0.7177029252052307,
+      "eval_mean_token_accuracy": 0.8050174043419656,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 89.2287,
+      "eval_samples_per_second": 15.914,
+      "eval_steps_per_second": 1.995,
+      "step": 780
+    },
+    {
+      "entropy": 0.6799364153295755,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 1.1364924907684326,
+      "learning_rate": 0.00035529241563569903,
+      "loss": 0.6714428901672364,
+      "mean_token_accuracy": 0.8087341658771038,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6603799047095052,
+      "eval_loss": 0.7201809287071228,
+      "eval_mean_token_accuracy": 0.802863868769635,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 89.2718,
+      "eval_samples_per_second": 15.906,
+      "eval_steps_per_second": 1.994,
+      "step": 800
+    },
+    {
+      "entropy": 0.7078610375523567,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 1.2191942930221558,
+      "learning_rate": 0.0003542942720597808,
+      "loss": 0.6996590614318847,
+      "mean_token_accuracy": 0.8049512051045895,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.7240545264120852,
+      "eval_loss": 0.7301309108734131,
+      "eval_mean_token_accuracy": 0.8015973166133581,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 89.3438,
+      "eval_samples_per_second": 15.894,
+      "eval_steps_per_second": 1.992,
+      "step": 820
+    },
+    {
+      "entropy": 0.6557391556826505,
+      "epoch": 2.024140012070006,
+      "grad_norm": 1.36203134059906,
+      "learning_rate": 0.0003532474734380064,
+      "loss": 0.6327113628387451,
+      "mean_token_accuracy": 0.8142829847026181,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6573808648613062,
+      "eval_loss": 0.7334365844726562,
+      "eval_mean_token_accuracy": 0.8043633893634496,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 89.3359,
+      "eval_samples_per_second": 15.895,
+      "eval_steps_per_second": 1.992,
+      "step": 840
+    },
+    {
+      "entropy": 0.6409128502011299,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 1.5804955959320068,
+      "learning_rate": 0.0003521523160020035,
+      "loss": 0.6023559093475341,
+      "mean_token_accuracy": 0.8222251623868942,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6983234788594621,
+      "eval_loss": 0.7232135534286499,
+      "eval_mean_token_accuracy": 0.8010654268639811,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 89.2615,
+      "eval_samples_per_second": 15.908,
+      "eval_steps_per_second": 1.994,
+      "step": 860
+    },
+    {
+      "entropy": 0.6151149723678827,
+      "epoch": 2.12070006035003,
+      "grad_norm": 1.1975436210632324,
+      "learning_rate": 0.00035100910966837193,
+      "loss": 0.5803564548492431,
+      "mean_token_accuracy": 0.8256325736641884,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6636555573243773,
+      "eval_loss": 0.7205662727355957,
+      "eval_mean_token_accuracy": 0.8036768369460374,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 89.2957,
+      "eval_samples_per_second": 15.902,
+      "eval_steps_per_second": 1.993,
+      "step": 880
+    },
+    {
+      "entropy": 0.6201727617532015,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9907131195068359,
+      "learning_rate": 0.0003498181779509813,
+      "loss": 0.6101872444152832,
+      "mean_token_accuracy": 0.8196512959897518,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.6473788086617931,
+      "eval_loss": 0.7062397003173828,
+      "eval_mean_token_accuracy": 0.8078362184963869,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 88.6198,
+      "eval_samples_per_second": 16.024,
+      "eval_steps_per_second": 2.009,
+      "step": 900
+    },
+    {
+      "entropy": 0.6097332935780286,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.9055633544921875,
+      "learning_rate": 0.00034857985786942026,
+      "loss": 0.5956094264984131,
+      "mean_token_accuracy": 0.8240575887262821,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.612629868341296,
+      "eval_loss": 0.7106770277023315,
+      "eval_mean_token_accuracy": 0.805324277181304,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 89.4383,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 920
+    },
+    {
+      "entropy": 0.6182701248675585,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.9513736367225647,
+      "learning_rate": 0.00034729449985362404,
+      "loss": 0.6019269466400147,
+      "mean_token_accuracy": 0.8187106013298034,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6437647456533453,
+      "eval_loss": 0.7173502445220947,
+      "eval_mean_token_accuracy": 0.8056439573175451,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.9568,
+      "eval_samples_per_second": 15.963,
+      "eval_steps_per_second": 2.001,
+      "step": 940
+    },
+    {
+      "entropy": 0.6508570514619351,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.9606207013130188,
+      "learning_rate": 0.0003459624676447067,
+      "loss": 0.6296409606933594,
+      "mean_token_accuracy": 0.8167732007801533,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6288247135248077,
+      "eval_loss": 0.7160675525665283,
+      "eval_mean_token_accuracy": 0.8066010689467527,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 89.0044,
+      "eval_samples_per_second": 15.954,
+      "eval_steps_per_second": 2.0,
+      "step": 960
+    },
+    {
+      "entropy": 0.6380081418901682,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.9677181839942932,
+      "learning_rate": 0.000344584138192027,
+      "loss": 0.6210546493530273,
+      "mean_token_accuracy": 0.8186135642230511,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6407805031604981,
+      "eval_loss": 0.7055043578147888,
+      "eval_mean_token_accuracy": 0.8070103157772107,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 89.5606,
+      "eval_samples_per_second": 15.855,
+      "eval_steps_per_second": 1.987,
+      "step": 980
+    },
+    {
+      "entropy": 0.6310251638293266,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 2.1046407222747803,
+      "learning_rate": 0.000343159901546516,
+      "loss": 0.6165830135345459,
+      "mean_token_accuracy": 0.8207855716347694,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6461507470420237,
+      "eval_loss": 0.7292136549949646,
+      "eval_mean_token_accuracy": 0.8030418261383356,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 89.315,
+      "eval_samples_per_second": 15.899,
+      "eval_steps_per_second": 1.993,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6256617344915867,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 1.1768349409103394,
+      "learning_rate": 0.0003416901607502972,
+      "loss": 0.6142709255218506,
+      "mean_token_accuracy": 0.8220669947564602,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6150523993406403,
+      "eval_loss": 0.7019563317298889,
+      "eval_mean_token_accuracy": 0.8094323099998946,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 89.419,
+      "eval_samples_per_second": 15.88,
+      "eval_steps_per_second": 1.991,
+      "step": 1020
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.0239715575080755e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1040/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1040/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1040/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1040/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1040/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..7ce4b072fcbe5b636867e13937c56fedee3b68b4
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1040/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.02728394274790723,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1040/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1040/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1040/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1040/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1040/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1040/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1040/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1040/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..8d1c79dc08fd89585fd8cc3a71aef55bc9f9595e
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1040/trainer_state.json
@@ -0,0 +1,1126 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 2.506940253470127,
+  "eval_steps": 20,
+  "global_step": 1040,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.024795526266098,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.945375919342041,
+      "learning_rate": 1.6698127428345936e-05,
+      "loss": 1.7825420379638672,
+      "mean_token_accuracy": 0.6315708436071873,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.29921497588747,
+      "eval_loss": 1.169131875038147,
+      "eval_mean_token_accuracy": 0.7185706342204233,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 91.7898,
+      "eval_samples_per_second": 15.47,
+      "eval_steps_per_second": 1.939,
+      "step": 20
+    },
+    {
+      "entropy": 1.0346344470977784,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.7916979789733887,
+      "learning_rate": 3.427510366871008e-05,
+      "loss": 0.9480395317077637,
+      "mean_token_accuracy": 0.7543429024517536,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 0.9673695430326997,
+      "eval_loss": 0.8588430285453796,
+      "eval_mean_token_accuracy": 0.7737233002534073,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 89.8219,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 40
+    },
+    {
+      "entropy": 0.8977566614747048,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7972187995910645,
+      "learning_rate": 5.185207990907422e-05,
+      "loss": 0.8029604911804199,
+      "mean_token_accuracy": 0.7818016350269318,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8436836982041263,
+      "eval_loss": 0.7893033623695374,
+      "eval_mean_token_accuracy": 0.7858914984076211,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 89.9349,
+      "eval_samples_per_second": 15.789,
+      "eval_steps_per_second": 1.979,
+      "step": 60
+    },
+    {
+      "entropy": 0.8650321319699288,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.2128729820251465,
+      "learning_rate": 6.942905614943836e-05,
+      "loss": 0.77417893409729,
+      "mean_token_accuracy": 0.7875479467213153,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8209857180547179,
+      "eval_loss": 0.7572280168533325,
+      "eval_mean_token_accuracy": 0.7925970383574453,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 89.9459,
+      "eval_samples_per_second": 15.787,
+      "eval_steps_per_second": 1.979,
+      "step": 80
+    },
+    {
+      "entropy": 0.8096660938113928,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.1254287958145142,
+      "learning_rate": 8.700603238980251e-05,
+      "loss": 0.7170331001281738,
+      "mean_token_accuracy": 0.7994555257260799,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.792780208788561,
+      "eval_loss": 0.7442984580993652,
+      "eval_mean_token_accuracy": 0.7954704573984896,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 90.294,
+      "eval_samples_per_second": 15.726,
+      "eval_steps_per_second": 1.971,
+      "step": 100
+    },
+    {
+      "entropy": 0.7979829199612141,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2140507698059082,
+      "learning_rate": 0.00010458300863016666,
+      "loss": 0.7141861915588379,
+      "mean_token_accuracy": 0.8000014744699001,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.7938692486018277,
+      "eval_loss": 0.7279797196388245,
+      "eval_mean_token_accuracy": 0.7989798734027348,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 89.7189,
+      "eval_samples_per_second": 15.827,
+      "eval_steps_per_second": 1.984,
+      "step": 120
+    },
+    {
+      "entropy": 0.7916347607970238,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.0506458282470703,
+      "learning_rate": 0.0001221599848705308,
+      "loss": 0.7117055416107178,
+      "mean_token_accuracy": 0.7970046654343605,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.8052781063519167,
+      "eval_loss": 0.722576916217804,
+      "eval_mean_token_accuracy": 0.7982530523551984,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 90.0376,
+      "eval_samples_per_second": 15.771,
+      "eval_steps_per_second": 1.977,
+      "step": 140
+    },
+    {
+      "entropy": 0.7880666613578796,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8409207463264465,
+      "learning_rate": 0.00013973696111089492,
+      "loss": 0.7052523136138916,
+      "mean_token_accuracy": 0.8000123649835587,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7787443134891853,
+      "eval_loss": 0.7137772440910339,
+      "eval_mean_token_accuracy": 0.8014209287220173,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 90.2562,
+      "eval_samples_per_second": 15.733,
+      "eval_steps_per_second": 1.972,
+      "step": 160
+    },
+    {
+      "entropy": 0.7696007996797561,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.0182477235794067,
+      "learning_rate": 0.00015731393735125907,
+      "loss": 0.7008797645568847,
+      "mean_token_accuracy": 0.8007259473204613,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.752259682068664,
+      "eval_loss": 0.7171286344528198,
+      "eval_mean_token_accuracy": 0.80052122574174,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 90.0797,
+      "eval_samples_per_second": 15.764,
+      "eval_steps_per_second": 1.976,
+      "step": 180
+    },
+    {
+      "entropy": 0.7828126326203346,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9553632736206055,
+      "learning_rate": 0.0001748909135916232,
+      "loss": 0.7046853542327881,
+      "mean_token_accuracy": 0.7988562889397144,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7389638169427936,
+      "eval_loss": 0.7187935709953308,
+      "eval_mean_token_accuracy": 0.7995899999409579,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 89.9223,
+      "eval_samples_per_second": 15.791,
+      "eval_steps_per_second": 1.979,
+      "step": 200
+    },
+    {
+      "entropy": 0.7909286297857762,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 1.134459376335144,
+      "learning_rate": 0.00019246788983198735,
+      "loss": 0.6989312171936035,
+      "mean_token_accuracy": 0.7991565138101577,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7496965199374082,
+      "eval_loss": 0.7106321454048157,
+      "eval_mean_token_accuracy": 0.8046183442131857,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 89.6899,
+      "eval_samples_per_second": 15.832,
+      "eval_steps_per_second": 1.985,
+      "step": 220
+    },
+    {
+      "entropy": 0.7651963323354721,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.9557585716247559,
+      "learning_rate": 0.0002100448660723515,
+      "loss": 0.6984559059143066,
+      "mean_token_accuracy": 0.8019823372364044,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7615563079212488,
+      "eval_loss": 0.7098539471626282,
+      "eval_mean_token_accuracy": 0.8047233508544022,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 90.3752,
+      "eval_samples_per_second": 15.712,
+      "eval_steps_per_second": 1.97,
+      "step": 240
+    },
+    {
+      "entropy": 0.787507726252079,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.9279311299324036,
+      "learning_rate": 0.00022762184231271568,
+      "loss": 0.708138370513916,
+      "mean_token_accuracy": 0.8013308703899383,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7856640896100676,
+      "eval_loss": 0.7191570401191711,
+      "eval_mean_token_accuracy": 0.8024677249153008,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 93.7336,
+      "eval_samples_per_second": 15.149,
+      "eval_steps_per_second": 1.899,
+      "step": 260
+    },
+    {
+      "entropy": 0.768489234894514,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.7992776036262512,
+      "learning_rate": 0.0002451988185530798,
+      "loss": 0.7035849571228028,
+      "mean_token_accuracy": 0.8031105332076549,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7682498740346244,
+      "eval_loss": 0.7149022221565247,
+      "eval_mean_token_accuracy": 0.7973497995499814,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 89.7309,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 280
+    },
+    {
+      "entropy": 0.7720955617725849,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.1931514739990234,
+      "learning_rate": 0.0002627757947934439,
+      "loss": 0.6965402603149414,
+      "mean_token_accuracy": 0.8040348663926125,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7214708304807042,
+      "eval_loss": 0.7350823879241943,
+      "eval_mean_token_accuracy": 0.7942911925610532,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 89.6451,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 300
+    },
+    {
+      "entropy": 0.7930095426738262,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8994467258453369,
+      "learning_rate": 0.0002803527710338081,
+      "loss": 0.7331492900848389,
+      "mean_token_accuracy": 0.796185302734375,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.8142535488926963,
+      "eval_loss": 0.7237834334373474,
+      "eval_mean_token_accuracy": 0.799495273761535,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 90.6759,
+      "eval_samples_per_second": 15.66,
+      "eval_steps_per_second": 1.963,
+      "step": 320
+    },
+    {
+      "entropy": 0.7825102441012859,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 1.8224239349365234,
+      "learning_rate": 0.00029792974727417223,
+      "loss": 0.7331175804138184,
+      "mean_token_accuracy": 0.7977667152881622,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7327135443017724,
+      "eval_loss": 0.7388784289360046,
+      "eval_mean_token_accuracy": 0.8007041252730938,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 89.7692,
+      "eval_samples_per_second": 15.818,
+      "eval_steps_per_second": 1.983,
+      "step": 340
+    },
+    {
+      "entropy": 0.819331557303667,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 1.0407779216766357,
+      "learning_rate": 0.0003155067235145364,
+      "loss": 0.7456695556640625,
+      "mean_token_accuracy": 0.7936980701982975,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.8065585420372781,
+      "eval_loss": 0.735519528388977,
+      "eval_mean_token_accuracy": 0.7974206296245704,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 89.4336,
+      "eval_samples_per_second": 15.878,
+      "eval_steps_per_second": 1.99,
+      "step": 360
+    },
+    {
+      "entropy": 0.8114653021097183,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.1209490299224854,
+      "learning_rate": 0.0003330836997549005,
+      "loss": 0.7340899467468261,
+      "mean_token_accuracy": 0.7914272703230381,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.8025527285056168,
+      "eval_loss": 0.7295576930046082,
+      "eval_mean_token_accuracy": 0.7990576628218876,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 89.9579,
+      "eval_samples_per_second": 15.785,
+      "eval_steps_per_second": 1.979,
+      "step": 380
+    },
+    {
+      "entropy": 0.8042497783899307,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 1.2437798976898193,
+      "learning_rate": 0.0003506606759952647,
+      "loss": 0.7381744384765625,
+      "mean_token_accuracy": 0.7967362694442273,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7717011336530193,
+      "eval_loss": 0.7436173558235168,
+      "eval_mean_token_accuracy": 0.7975849482450592,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 89.8411,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 400
+    },
+    {
+      "entropy": 0.7987750010056929,
+      "epoch": 1.012070006035003,
+      "grad_norm": 1.2390918731689453,
+      "learning_rate": 0.000364721224843344,
+      "loss": 0.7317790031433106,
+      "mean_token_accuracy": 0.7986087551364651,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7333170604170038,
+      "eval_loss": 0.7456948757171631,
+      "eval_mean_token_accuracy": 0.7964897577682238,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 89.7288,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 420
+    },
+    {
+      "entropy": 0.7372976846992969,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.1552740335464478,
+      "learning_rate": 0.00036468510102269575,
+      "loss": 0.6734249114990234,
+      "mean_token_accuracy": 0.805834824591875,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.7130355526891987,
+      "eval_loss": 0.7555333375930786,
+      "eval_mean_token_accuracy": 0.7961652781186479,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 89.8713,
+      "eval_samples_per_second": 15.8,
+      "eval_steps_per_second": 1.981,
+      "step": 440
+    },
+    {
+      "entropy": 0.7465668715536594,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 1.1991469860076904,
+      "learning_rate": 0.0003645973816745066,
+      "loss": 0.6871879577636719,
+      "mean_token_accuracy": 0.8042450234293937,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7772979542110743,
+      "eval_loss": 0.7561826705932617,
+      "eval_mean_token_accuracy": 0.7953876427720102,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 89.8395,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 460
+    },
+    {
+      "entropy": 0.7477855801582336,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 1.325486660003662,
+      "learning_rate": 0.00036445809162231435,
+      "loss": 0.6836830139160156,
+      "mean_token_accuracy": 0.8021619468927383,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.7098110676481483,
+      "eval_loss": 0.758007287979126,
+      "eval_mean_token_accuracy": 0.7909953189030122,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 89.6101,
+      "eval_samples_per_second": 15.846,
+      "eval_steps_per_second": 1.986,
+      "step": 480
+    },
+    {
+      "entropy": 0.7632691070437432,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 1.0122262239456177,
+      "learning_rate": 0.0003642672702835562,
+      "loss": 0.7071213245391845,
+      "mean_token_accuracy": 0.8000680930912495,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7645382720432924,
+      "eval_loss": 0.7492606043815613,
+      "eval_mean_token_accuracy": 0.7976858676149604,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 89.6092,
+      "eval_samples_per_second": 15.847,
+      "eval_steps_per_second": 1.986,
+      "step": 500
+    },
+    {
+      "entropy": 0.7623726457357407,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.367774486541748,
+      "learning_rate": 0.00036402497165841384,
+      "loss": 0.6965175151824952,
+      "mean_token_accuracy": 0.8016501650214195,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7868083715438843,
+      "eval_loss": 0.7481449842453003,
+      "eval_mean_token_accuracy": 0.799330582779445,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 89.8241,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 520
+    },
+    {
+      "entropy": 0.7535207405686378,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 1.213614821434021,
+      "learning_rate": 0.00036373126431453207,
+      "loss": 0.7078310489654541,
+      "mean_token_accuracy": 0.7998907208442688,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7829091950748743,
+      "eval_loss": 0.7416072487831116,
+      "eval_mean_token_accuracy": 0.7969146250339036,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 89.4607,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 540
+    },
+    {
+      "entropy": 0.7527363449335098,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 1.0929982662200928,
+      "learning_rate": 0.00036338623136761495,
+      "loss": 0.7058369636535644,
+      "mean_token_accuracy": 0.7997087433934211,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.728419938114252,
+      "eval_loss": 0.744873046875,
+      "eval_mean_token_accuracy": 0.7966634296299366,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 89.6597,
+      "eval_samples_per_second": 15.838,
+      "eval_steps_per_second": 1.985,
+      "step": 560
+    },
+    {
+      "entropy": 0.7406099259853363,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.409364104270935,
+      "learning_rate": 0.00036298997045790515,
+      "loss": 0.6902852058410645,
+      "mean_token_accuracy": 0.805386807769537,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.6991632758231645,
+      "eval_loss": 0.7480319738388062,
+      "eval_mean_token_accuracy": 0.7970068006033308,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 89.8329,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 580
+    },
+    {
+      "entropy": 0.7459449704736472,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.7053111791610718,
+      "learning_rate": 0.00036254259372255277,
+      "loss": 0.7044640064239502,
+      "mean_token_accuracy": 0.8005018755793571,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7454566660891758,
+      "eval_loss": 0.7551484704017639,
+      "eval_mean_token_accuracy": 0.7897093731365846,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 89.8313,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 600
+    },
+    {
+      "entropy": 0.7621391348540782,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 1.174443244934082,
+      "learning_rate": 0.000362044227763882,
+      "loss": 0.7111573696136475,
+      "mean_token_accuracy": 0.798752411454916,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7664387949397055,
+      "eval_loss": 0.7365804314613342,
+      "eval_mean_token_accuracy": 0.7963101200843126,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 89.9623,
+      "eval_samples_per_second": 15.784,
+      "eval_steps_per_second": 1.979,
+      "step": 620
+    },
+    {
+      "entropy": 0.7424944408237935,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.2898184061050415,
+      "learning_rate": 0.000361495013613564,
+      "loss": 0.6913110733032226,
+      "mean_token_accuracy": 0.8009015507996082,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7255479529332579,
+      "eval_loss": 0.7345991730690002,
+      "eval_mean_token_accuracy": 0.8013592945056015,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 89.817,
+      "eval_samples_per_second": 15.81,
+      "eval_steps_per_second": 1.982,
+      "step": 640
+    },
+    {
+      "entropy": 0.7303367637097835,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.1310476064682007,
+      "learning_rate": 0.00036089510669270666,
+      "loss": 0.6875617027282714,
+      "mean_token_accuracy": 0.8035397171974182,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7101308248016271,
+      "eval_loss": 0.7394917011260986,
+      "eval_mean_token_accuracy": 0.7990792942850777,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 89.7431,
+      "eval_samples_per_second": 15.823,
+      "eval_steps_per_second": 1.983,
+      "step": 660
+    },
+    {
+      "entropy": 0.7387678384780884,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.9645389914512634,
+      "learning_rate": 0.0003602446767678725,
+      "loss": 0.6911204338073731,
+      "mean_token_accuracy": 0.8056333385407924,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7281073737010527,
+      "eval_loss": 0.7376708984375,
+      "eval_mean_token_accuracy": 0.7997356415464637,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 90.0338,
+      "eval_samples_per_second": 15.772,
+      "eval_steps_per_second": 1.977,
+      "step": 680
+    },
+    {
+      "entropy": 0.7449554048478604,
+      "epoch": 1.687990343995172,
+      "grad_norm": 1.9069629907608032,
+      "learning_rate": 0.0003595439079030364,
+      "loss": 0.7042286396026611,
+      "mean_token_accuracy": 0.8018639378249646,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7022633817088738,
+      "eval_loss": 0.7320939898490906,
+      "eval_mean_token_accuracy": 0.8006051541044471,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 89.8948,
+      "eval_samples_per_second": 15.796,
+      "eval_steps_per_second": 1.98,
+      "step": 700
+    },
+    {
+      "entropy": 0.723249051719904,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.8333877325057983,
+      "learning_rate": 0.00035879299840749777,
+      "loss": 0.7043714046478271,
+      "mean_token_accuracy": 0.8028345607221127,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.672436946898364,
+      "eval_loss": 0.713233470916748,
+      "eval_mean_token_accuracy": 0.8047760577684038,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 89.6394,
+      "eval_samples_per_second": 15.841,
+      "eval_steps_per_second": 1.986,
+      "step": 720
+    },
+    {
+      "entropy": 0.730014206469059,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.4580037593841553,
+      "learning_rate": 0.00035799216077976137,
+      "loss": 0.6985176563262939,
+      "mean_token_accuracy": 0.8015547141432762,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.6908873795123582,
+      "eval_loss": 0.7198202013969421,
+      "eval_mean_token_accuracy": 0.8024131682481659,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 89.3771,
+      "eval_samples_per_second": 15.888,
+      "eval_steps_per_second": 1.992,
+      "step": 740
+    },
+    {
+      "entropy": 0.7219605796039105,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.51682710647583,
+      "learning_rate": 0.000357141621647403,
+      "loss": 0.6844424724578857,
+      "mean_token_accuracy": 0.8032813094556331,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7113401109582922,
+      "eval_loss": 0.7195309400558472,
+      "eval_mean_token_accuracy": 0.8019187035185568,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 89.4658,
+      "eval_samples_per_second": 15.872,
+      "eval_steps_per_second": 1.99,
+      "step": 760
+    },
+    {
+      "entropy": 0.729826743900776,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 1.0838000774383545,
+      "learning_rate": 0.0003562416217029361,
+      "loss": 0.685858678817749,
+      "mean_token_accuracy": 0.8050567395985126,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.649370267484965,
+      "eval_loss": 0.7177029252052307,
+      "eval_mean_token_accuracy": 0.8050174043419656,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 89.2287,
+      "eval_samples_per_second": 15.914,
+      "eval_steps_per_second": 1.995,
+      "step": 780
+    },
+    {
+      "entropy": 0.6799364153295755,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 1.1364924907684326,
+      "learning_rate": 0.00035529241563569903,
+      "loss": 0.6714428901672364,
+      "mean_token_accuracy": 0.8087341658771038,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6603799047095052,
+      "eval_loss": 0.7201809287071228,
+      "eval_mean_token_accuracy": 0.802863868769635,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 89.2718,
+      "eval_samples_per_second": 15.906,
+      "eval_steps_per_second": 1.994,
+      "step": 800
+    },
+    {
+      "entropy": 0.7078610375523567,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 1.2191942930221558,
+      "learning_rate": 0.0003542942720597808,
+      "loss": 0.6996590614318847,
+      "mean_token_accuracy": 0.8049512051045895,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.7240545264120852,
+      "eval_loss": 0.7301309108734131,
+      "eval_mean_token_accuracy": 0.8015973166133581,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 89.3438,
+      "eval_samples_per_second": 15.894,
+      "eval_steps_per_second": 1.992,
+      "step": 820
+    },
+    {
+      "entropy": 0.6557391556826505,
+      "epoch": 2.024140012070006,
+      "grad_norm": 1.36203134059906,
+      "learning_rate": 0.0003532474734380064,
+      "loss": 0.6327113628387451,
+      "mean_token_accuracy": 0.8142829847026181,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6573808648613062,
+      "eval_loss": 0.7334365844726562,
+      "eval_mean_token_accuracy": 0.8043633893634496,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 89.3359,
+      "eval_samples_per_second": 15.895,
+      "eval_steps_per_second": 1.992,
+      "step": 840
+    },
+    {
+      "entropy": 0.6409128502011299,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 1.5804955959320068,
+      "learning_rate": 0.0003521523160020035,
+      "loss": 0.6023559093475341,
+      "mean_token_accuracy": 0.8222251623868942,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6983234788594621,
+      "eval_loss": 0.7232135534286499,
+      "eval_mean_token_accuracy": 0.8010654268639811,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 89.2615,
+      "eval_samples_per_second": 15.908,
+      "eval_steps_per_second": 1.994,
+      "step": 860
+    },
+    {
+      "entropy": 0.6151149723678827,
+      "epoch": 2.12070006035003,
+      "grad_norm": 1.1975436210632324,
+      "learning_rate": 0.00035100910966837193,
+      "loss": 0.5803564548492431,
+      "mean_token_accuracy": 0.8256325736641884,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6636555573243773,
+      "eval_loss": 0.7205662727355957,
+      "eval_mean_token_accuracy": 0.8036768369460374,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 89.2957,
+      "eval_samples_per_second": 15.902,
+      "eval_steps_per_second": 1.993,
+      "step": 880
+    },
+    {
+      "entropy": 0.6201727617532015,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9907131195068359,
+      "learning_rate": 0.0003498181779509813,
+      "loss": 0.6101872444152832,
+      "mean_token_accuracy": 0.8196512959897518,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.6473788086617931,
+      "eval_loss": 0.7062397003173828,
+      "eval_mean_token_accuracy": 0.8078362184963869,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 88.6198,
+      "eval_samples_per_second": 16.024,
+      "eval_steps_per_second": 2.009,
+      "step": 900
+    },
+    {
+      "entropy": 0.6097332935780286,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.9055633544921875,
+      "learning_rate": 0.00034857985786942026,
+      "loss": 0.5956094264984131,
+      "mean_token_accuracy": 0.8240575887262821,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.612629868341296,
+      "eval_loss": 0.7106770277023315,
+      "eval_mean_token_accuracy": 0.805324277181304,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 89.4383,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 920
+    },
+    {
+      "entropy": 0.6182701248675585,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.9513736367225647,
+      "learning_rate": 0.00034729449985362404,
+      "loss": 0.6019269466400147,
+      "mean_token_accuracy": 0.8187106013298034,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6437647456533453,
+      "eval_loss": 0.7173502445220947,
+      "eval_mean_token_accuracy": 0.8056439573175451,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.9568,
+      "eval_samples_per_second": 15.963,
+      "eval_steps_per_second": 2.001,
+      "step": 940
+    },
+    {
+      "entropy": 0.6508570514619351,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.9606207013130188,
+      "learning_rate": 0.0003459624676447067,
+      "loss": 0.6296409606933594,
+      "mean_token_accuracy": 0.8167732007801533,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6288247135248077,
+      "eval_loss": 0.7160675525665283,
+      "eval_mean_token_accuracy": 0.8066010689467527,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 89.0044,
+      "eval_samples_per_second": 15.954,
+      "eval_steps_per_second": 2.0,
+      "step": 960
+    },
+    {
+      "entropy": 0.6380081418901682,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.9677181839942932,
+      "learning_rate": 0.000344584138192027,
+      "loss": 0.6210546493530273,
+      "mean_token_accuracy": 0.8186135642230511,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6407805031604981,
+      "eval_loss": 0.7055043578147888,
+      "eval_mean_token_accuracy": 0.8070103157772107,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 89.5606,
+      "eval_samples_per_second": 15.855,
+      "eval_steps_per_second": 1.987,
+      "step": 980
+    },
+    {
+      "entropy": 0.6310251638293266,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 2.1046407222747803,
+      "learning_rate": 0.000343159901546516,
+      "loss": 0.6165830135345459,
+      "mean_token_accuracy": 0.8207855716347694,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6461507470420237,
+      "eval_loss": 0.7292136549949646,
+      "eval_mean_token_accuracy": 0.8030418261383356,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 89.315,
+      "eval_samples_per_second": 15.899,
+      "eval_steps_per_second": 1.993,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6256617344915867,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 1.1768349409103394,
+      "learning_rate": 0.0003416901607502972,
+      "loss": 0.6142709255218506,
+      "mean_token_accuracy": 0.8220669947564602,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6150523993406403,
+      "eval_loss": 0.7019563317298889,
+      "eval_mean_token_accuracy": 0.8094323099998946,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 89.419,
+      "eval_samples_per_second": 15.88,
+      "eval_steps_per_second": 1.991,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6126735735684633,
+      "epoch": 2.506940253470127,
+      "grad_norm": 1.8065496683120728,
+      "learning_rate": 0.00034017533172263055,
+      "loss": 0.6056458473205566,
+      "mean_token_accuracy": 0.8203762136399746,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6544345456562685,
+      "eval_loss": 0.7086517214775085,
+      "eval_mean_token_accuracy": 0.806709805901131,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 88.981,
+      "eval_samples_per_second": 15.958,
+      "eval_steps_per_second": 2.0,
+      "step": 1040
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.0435065133937664e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1060/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1060/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1060/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1060/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1060/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..7ce4b072fcbe5b636867e13937c56fedee3b68b4
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1060/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.02728394274790723,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1060/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1060/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1060/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1060/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1060/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1060/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1060/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1060/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..0111253307a79c355d8c8126917897c83baa926b
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1060/trainer_state.json
@@ -0,0 +1,1147 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 2.5552202776101387,
+  "eval_steps": 20,
+  "global_step": 1060,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.024795526266098,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.945375919342041,
+      "learning_rate": 1.6698127428345936e-05,
+      "loss": 1.7825420379638672,
+      "mean_token_accuracy": 0.6315708436071873,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.29921497588747,
+      "eval_loss": 1.169131875038147,
+      "eval_mean_token_accuracy": 0.7185706342204233,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 91.7898,
+      "eval_samples_per_second": 15.47,
+      "eval_steps_per_second": 1.939,
+      "step": 20
+    },
+    {
+      "entropy": 1.0346344470977784,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.7916979789733887,
+      "learning_rate": 3.427510366871008e-05,
+      "loss": 0.9480395317077637,
+      "mean_token_accuracy": 0.7543429024517536,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 0.9673695430326997,
+      "eval_loss": 0.8588430285453796,
+      "eval_mean_token_accuracy": 0.7737233002534073,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 89.8219,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 40
+    },
+    {
+      "entropy": 0.8977566614747048,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7972187995910645,
+      "learning_rate": 5.185207990907422e-05,
+      "loss": 0.8029604911804199,
+      "mean_token_accuracy": 0.7818016350269318,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8436836982041263,
+      "eval_loss": 0.7893033623695374,
+      "eval_mean_token_accuracy": 0.7858914984076211,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 89.9349,
+      "eval_samples_per_second": 15.789,
+      "eval_steps_per_second": 1.979,
+      "step": 60
+    },
+    {
+      "entropy": 0.8650321319699288,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.2128729820251465,
+      "learning_rate": 6.942905614943836e-05,
+      "loss": 0.77417893409729,
+      "mean_token_accuracy": 0.7875479467213153,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8209857180547179,
+      "eval_loss": 0.7572280168533325,
+      "eval_mean_token_accuracy": 0.7925970383574453,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 89.9459,
+      "eval_samples_per_second": 15.787,
+      "eval_steps_per_second": 1.979,
+      "step": 80
+    },
+    {
+      "entropy": 0.8096660938113928,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.1254287958145142,
+      "learning_rate": 8.700603238980251e-05,
+      "loss": 0.7170331001281738,
+      "mean_token_accuracy": 0.7994555257260799,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.792780208788561,
+      "eval_loss": 0.7442984580993652,
+      "eval_mean_token_accuracy": 0.7954704573984896,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 90.294,
+      "eval_samples_per_second": 15.726,
+      "eval_steps_per_second": 1.971,
+      "step": 100
+    },
+    {
+      "entropy": 0.7979829199612141,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2140507698059082,
+      "learning_rate": 0.00010458300863016666,
+      "loss": 0.7141861915588379,
+      "mean_token_accuracy": 0.8000014744699001,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.7938692486018277,
+      "eval_loss": 0.7279797196388245,
+      "eval_mean_token_accuracy": 0.7989798734027348,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 89.7189,
+      "eval_samples_per_second": 15.827,
+      "eval_steps_per_second": 1.984,
+      "step": 120
+    },
+    {
+      "entropy": 0.7916347607970238,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.0506458282470703,
+      "learning_rate": 0.0001221599848705308,
+      "loss": 0.7117055416107178,
+      "mean_token_accuracy": 0.7970046654343605,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.8052781063519167,
+      "eval_loss": 0.722576916217804,
+      "eval_mean_token_accuracy": 0.7982530523551984,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 90.0376,
+      "eval_samples_per_second": 15.771,
+      "eval_steps_per_second": 1.977,
+      "step": 140
+    },
+    {
+      "entropy": 0.7880666613578796,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8409207463264465,
+      "learning_rate": 0.00013973696111089492,
+      "loss": 0.7052523136138916,
+      "mean_token_accuracy": 0.8000123649835587,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7787443134891853,
+      "eval_loss": 0.7137772440910339,
+      "eval_mean_token_accuracy": 0.8014209287220173,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 90.2562,
+      "eval_samples_per_second": 15.733,
+      "eval_steps_per_second": 1.972,
+      "step": 160
+    },
+    {
+      "entropy": 0.7696007996797561,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.0182477235794067,
+      "learning_rate": 0.00015731393735125907,
+      "loss": 0.7008797645568847,
+      "mean_token_accuracy": 0.8007259473204613,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.752259682068664,
+      "eval_loss": 0.7171286344528198,
+      "eval_mean_token_accuracy": 0.80052122574174,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 90.0797,
+      "eval_samples_per_second": 15.764,
+      "eval_steps_per_second": 1.976,
+      "step": 180
+    },
+    {
+      "entropy": 0.7828126326203346,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9553632736206055,
+      "learning_rate": 0.0001748909135916232,
+      "loss": 0.7046853542327881,
+      "mean_token_accuracy": 0.7988562889397144,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7389638169427936,
+      "eval_loss": 0.7187935709953308,
+      "eval_mean_token_accuracy": 0.7995899999409579,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 89.9223,
+      "eval_samples_per_second": 15.791,
+      "eval_steps_per_second": 1.979,
+      "step": 200
+    },
+    {
+      "entropy": 0.7909286297857762,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 1.134459376335144,
+      "learning_rate": 0.00019246788983198735,
+      "loss": 0.6989312171936035,
+      "mean_token_accuracy": 0.7991565138101577,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7496965199374082,
+      "eval_loss": 0.7106321454048157,
+      "eval_mean_token_accuracy": 0.8046183442131857,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 89.6899,
+      "eval_samples_per_second": 15.832,
+      "eval_steps_per_second": 1.985,
+      "step": 220
+    },
+    {
+      "entropy": 0.7651963323354721,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.9557585716247559,
+      "learning_rate": 0.0002100448660723515,
+      "loss": 0.6984559059143066,
+      "mean_token_accuracy": 0.8019823372364044,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7615563079212488,
+      "eval_loss": 0.7098539471626282,
+      "eval_mean_token_accuracy": 0.8047233508544022,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 90.3752,
+      "eval_samples_per_second": 15.712,
+      "eval_steps_per_second": 1.97,
+      "step": 240
+    },
+    {
+      "entropy": 0.787507726252079,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.9279311299324036,
+      "learning_rate": 0.00022762184231271568,
+      "loss": 0.708138370513916,
+      "mean_token_accuracy": 0.8013308703899383,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7856640896100676,
+      "eval_loss": 0.7191570401191711,
+      "eval_mean_token_accuracy": 0.8024677249153008,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 93.7336,
+      "eval_samples_per_second": 15.149,
+      "eval_steps_per_second": 1.899,
+      "step": 260
+    },
+    {
+      "entropy": 0.768489234894514,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.7992776036262512,
+      "learning_rate": 0.0002451988185530798,
+      "loss": 0.7035849571228028,
+      "mean_token_accuracy": 0.8031105332076549,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7682498740346244,
+      "eval_loss": 0.7149022221565247,
+      "eval_mean_token_accuracy": 0.7973497995499814,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 89.7309,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 280
+    },
+    {
+      "entropy": 0.7720955617725849,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.1931514739990234,
+      "learning_rate": 0.0002627757947934439,
+      "loss": 0.6965402603149414,
+      "mean_token_accuracy": 0.8040348663926125,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7214708304807042,
+      "eval_loss": 0.7350823879241943,
+      "eval_mean_token_accuracy": 0.7942911925610532,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 89.6451,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 300
+    },
+    {
+      "entropy": 0.7930095426738262,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8994467258453369,
+      "learning_rate": 0.0002803527710338081,
+      "loss": 0.7331492900848389,
+      "mean_token_accuracy": 0.796185302734375,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.8142535488926963,
+      "eval_loss": 0.7237834334373474,
+      "eval_mean_token_accuracy": 0.799495273761535,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 90.6759,
+      "eval_samples_per_second": 15.66,
+      "eval_steps_per_second": 1.963,
+      "step": 320
+    },
+    {
+      "entropy": 0.7825102441012859,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 1.8224239349365234,
+      "learning_rate": 0.00029792974727417223,
+      "loss": 0.7331175804138184,
+      "mean_token_accuracy": 0.7977667152881622,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7327135443017724,
+      "eval_loss": 0.7388784289360046,
+      "eval_mean_token_accuracy": 0.8007041252730938,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 89.7692,
+      "eval_samples_per_second": 15.818,
+      "eval_steps_per_second": 1.983,
+      "step": 340
+    },
+    {
+      "entropy": 0.819331557303667,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 1.0407779216766357,
+      "learning_rate": 0.0003155067235145364,
+      "loss": 0.7456695556640625,
+      "mean_token_accuracy": 0.7936980701982975,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.8065585420372781,
+      "eval_loss": 0.735519528388977,
+      "eval_mean_token_accuracy": 0.7974206296245704,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 89.4336,
+      "eval_samples_per_second": 15.878,
+      "eval_steps_per_second": 1.99,
+      "step": 360
+    },
+    {
+      "entropy": 0.8114653021097183,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.1209490299224854,
+      "learning_rate": 0.0003330836997549005,
+      "loss": 0.7340899467468261,
+      "mean_token_accuracy": 0.7914272703230381,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.8025527285056168,
+      "eval_loss": 0.7295576930046082,
+      "eval_mean_token_accuracy": 0.7990576628218876,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 89.9579,
+      "eval_samples_per_second": 15.785,
+      "eval_steps_per_second": 1.979,
+      "step": 380
+    },
+    {
+      "entropy": 0.8042497783899307,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 1.2437798976898193,
+      "learning_rate": 0.0003506606759952647,
+      "loss": 0.7381744384765625,
+      "mean_token_accuracy": 0.7967362694442273,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7717011336530193,
+      "eval_loss": 0.7436173558235168,
+      "eval_mean_token_accuracy": 0.7975849482450592,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 89.8411,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 400
+    },
+    {
+      "entropy": 0.7987750010056929,
+      "epoch": 1.012070006035003,
+      "grad_norm": 1.2390918731689453,
+      "learning_rate": 0.000364721224843344,
+      "loss": 0.7317790031433106,
+      "mean_token_accuracy": 0.7986087551364651,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7333170604170038,
+      "eval_loss": 0.7456948757171631,
+      "eval_mean_token_accuracy": 0.7964897577682238,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 89.7288,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 420
+    },
+    {
+      "entropy": 0.7372976846992969,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.1552740335464478,
+      "learning_rate": 0.00036468510102269575,
+      "loss": 0.6734249114990234,
+      "mean_token_accuracy": 0.805834824591875,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.7130355526891987,
+      "eval_loss": 0.7555333375930786,
+      "eval_mean_token_accuracy": 0.7961652781186479,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 89.8713,
+      "eval_samples_per_second": 15.8,
+      "eval_steps_per_second": 1.981,
+      "step": 440
+    },
+    {
+      "entropy": 0.7465668715536594,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 1.1991469860076904,
+      "learning_rate": 0.0003645973816745066,
+      "loss": 0.6871879577636719,
+      "mean_token_accuracy": 0.8042450234293937,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7772979542110743,
+      "eval_loss": 0.7561826705932617,
+      "eval_mean_token_accuracy": 0.7953876427720102,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 89.8395,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 460
+    },
+    {
+      "entropy": 0.7477855801582336,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 1.325486660003662,
+      "learning_rate": 0.00036445809162231435,
+      "loss": 0.6836830139160156,
+      "mean_token_accuracy": 0.8021619468927383,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.7098110676481483,
+      "eval_loss": 0.758007287979126,
+      "eval_mean_token_accuracy": 0.7909953189030122,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 89.6101,
+      "eval_samples_per_second": 15.846,
+      "eval_steps_per_second": 1.986,
+      "step": 480
+    },
+    {
+      "entropy": 0.7632691070437432,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 1.0122262239456177,
+      "learning_rate": 0.0003642672702835562,
+      "loss": 0.7071213245391845,
+      "mean_token_accuracy": 0.8000680930912495,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7645382720432924,
+      "eval_loss": 0.7492606043815613,
+      "eval_mean_token_accuracy": 0.7976858676149604,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 89.6092,
+      "eval_samples_per_second": 15.847,
+      "eval_steps_per_second": 1.986,
+      "step": 500
+    },
+    {
+      "entropy": 0.7623726457357407,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.367774486541748,
+      "learning_rate": 0.00036402497165841384,
+      "loss": 0.6965175151824952,
+      "mean_token_accuracy": 0.8016501650214195,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7868083715438843,
+      "eval_loss": 0.7481449842453003,
+      "eval_mean_token_accuracy": 0.799330582779445,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 89.8241,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 520
+    },
+    {
+      "entropy": 0.7535207405686378,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 1.213614821434021,
+      "learning_rate": 0.00036373126431453207,
+      "loss": 0.7078310489654541,
+      "mean_token_accuracy": 0.7998907208442688,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7829091950748743,
+      "eval_loss": 0.7416072487831116,
+      "eval_mean_token_accuracy": 0.7969146250339036,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 89.4607,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 540
+    },
+    {
+      "entropy": 0.7527363449335098,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 1.0929982662200928,
+      "learning_rate": 0.00036338623136761495,
+      "loss": 0.7058369636535644,
+      "mean_token_accuracy": 0.7997087433934211,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.728419938114252,
+      "eval_loss": 0.744873046875,
+      "eval_mean_token_accuracy": 0.7966634296299366,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 89.6597,
+      "eval_samples_per_second": 15.838,
+      "eval_steps_per_second": 1.985,
+      "step": 560
+    },
+    {
+      "entropy": 0.7406099259853363,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.409364104270935,
+      "learning_rate": 0.00036298997045790515,
+      "loss": 0.6902852058410645,
+      "mean_token_accuracy": 0.805386807769537,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.6991632758231645,
+      "eval_loss": 0.7480319738388062,
+      "eval_mean_token_accuracy": 0.7970068006033308,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 89.8329,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 580
+    },
+    {
+      "entropy": 0.7459449704736472,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.7053111791610718,
+      "learning_rate": 0.00036254259372255277,
+      "loss": 0.7044640064239502,
+      "mean_token_accuracy": 0.8005018755793571,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7454566660891758,
+      "eval_loss": 0.7551484704017639,
+      "eval_mean_token_accuracy": 0.7897093731365846,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 89.8313,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 600
+    },
+    {
+      "entropy": 0.7621391348540782,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 1.174443244934082,
+      "learning_rate": 0.000362044227763882,
+      "loss": 0.7111573696136475,
+      "mean_token_accuracy": 0.798752411454916,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7664387949397055,
+      "eval_loss": 0.7365804314613342,
+      "eval_mean_token_accuracy": 0.7963101200843126,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 89.9623,
+      "eval_samples_per_second": 15.784,
+      "eval_steps_per_second": 1.979,
+      "step": 620
+    },
+    {
+      "entropy": 0.7424944408237935,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.2898184061050415,
+      "learning_rate": 0.000361495013613564,
+      "loss": 0.6913110733032226,
+      "mean_token_accuracy": 0.8009015507996082,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7255479529332579,
+      "eval_loss": 0.7345991730690002,
+      "eval_mean_token_accuracy": 0.8013592945056015,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 89.817,
+      "eval_samples_per_second": 15.81,
+      "eval_steps_per_second": 1.982,
+      "step": 640
+    },
+    {
+      "entropy": 0.7303367637097835,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.1310476064682007,
+      "learning_rate": 0.00036089510669270666,
+      "loss": 0.6875617027282714,
+      "mean_token_accuracy": 0.8035397171974182,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7101308248016271,
+      "eval_loss": 0.7394917011260986,
+      "eval_mean_token_accuracy": 0.7990792942850777,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 89.7431,
+      "eval_samples_per_second": 15.823,
+      "eval_steps_per_second": 1.983,
+      "step": 660
+    },
+    {
+      "entropy": 0.7387678384780884,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.9645389914512634,
+      "learning_rate": 0.0003602446767678725,
+      "loss": 0.6911204338073731,
+      "mean_token_accuracy": 0.8056333385407924,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7281073737010527,
+      "eval_loss": 0.7376708984375,
+      "eval_mean_token_accuracy": 0.7997356415464637,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 90.0338,
+      "eval_samples_per_second": 15.772,
+      "eval_steps_per_second": 1.977,
+      "step": 680
+    },
+    {
+      "entropy": 0.7449554048478604,
+      "epoch": 1.687990343995172,
+      "grad_norm": 1.9069629907608032,
+      "learning_rate": 0.0003595439079030364,
+      "loss": 0.7042286396026611,
+      "mean_token_accuracy": 0.8018639378249646,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7022633817088738,
+      "eval_loss": 0.7320939898490906,
+      "eval_mean_token_accuracy": 0.8006051541044471,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 89.8948,
+      "eval_samples_per_second": 15.796,
+      "eval_steps_per_second": 1.98,
+      "step": 700
+    },
+    {
+      "entropy": 0.723249051719904,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.8333877325057983,
+      "learning_rate": 0.00035879299840749777,
+      "loss": 0.7043714046478271,
+      "mean_token_accuracy": 0.8028345607221127,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.672436946898364,
+      "eval_loss": 0.713233470916748,
+      "eval_mean_token_accuracy": 0.8047760577684038,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 89.6394,
+      "eval_samples_per_second": 15.841,
+      "eval_steps_per_second": 1.986,
+      "step": 720
+    },
+    {
+      "entropy": 0.730014206469059,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.4580037593841553,
+      "learning_rate": 0.00035799216077976137,
+      "loss": 0.6985176563262939,
+      "mean_token_accuracy": 0.8015547141432762,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.6908873795123582,
+      "eval_loss": 0.7198202013969421,
+      "eval_mean_token_accuracy": 0.8024131682481659,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 89.3771,
+      "eval_samples_per_second": 15.888,
+      "eval_steps_per_second": 1.992,
+      "step": 740
+    },
+    {
+      "entropy": 0.7219605796039105,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.51682710647583,
+      "learning_rate": 0.000357141621647403,
+      "loss": 0.6844424724578857,
+      "mean_token_accuracy": 0.8032813094556331,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7113401109582922,
+      "eval_loss": 0.7195309400558472,
+      "eval_mean_token_accuracy": 0.8019187035185568,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 89.4658,
+      "eval_samples_per_second": 15.872,
+      "eval_steps_per_second": 1.99,
+      "step": 760
+    },
+    {
+      "entropy": 0.729826743900776,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 1.0838000774383545,
+      "learning_rate": 0.0003562416217029361,
+      "loss": 0.685858678817749,
+      "mean_token_accuracy": 0.8050567395985126,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.649370267484965,
+      "eval_loss": 0.7177029252052307,
+      "eval_mean_token_accuracy": 0.8050174043419656,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 89.2287,
+      "eval_samples_per_second": 15.914,
+      "eval_steps_per_second": 1.995,
+      "step": 780
+    },
+    {
+      "entropy": 0.6799364153295755,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 1.1364924907684326,
+      "learning_rate": 0.00035529241563569903,
+      "loss": 0.6714428901672364,
+      "mean_token_accuracy": 0.8087341658771038,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6603799047095052,
+      "eval_loss": 0.7201809287071228,
+      "eval_mean_token_accuracy": 0.802863868769635,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 89.2718,
+      "eval_samples_per_second": 15.906,
+      "eval_steps_per_second": 1.994,
+      "step": 800
+    },
+    {
+      "entropy": 0.7078610375523567,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 1.2191942930221558,
+      "learning_rate": 0.0003542942720597808,
+      "loss": 0.6996590614318847,
+      "mean_token_accuracy": 0.8049512051045895,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.7240545264120852,
+      "eval_loss": 0.7301309108734131,
+      "eval_mean_token_accuracy": 0.8015973166133581,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 89.3438,
+      "eval_samples_per_second": 15.894,
+      "eval_steps_per_second": 1.992,
+      "step": 820
+    },
+    {
+      "entropy": 0.6557391556826505,
+      "epoch": 2.024140012070006,
+      "grad_norm": 1.36203134059906,
+      "learning_rate": 0.0003532474734380064,
+      "loss": 0.6327113628387451,
+      "mean_token_accuracy": 0.8142829847026181,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6573808648613062,
+      "eval_loss": 0.7334365844726562,
+      "eval_mean_token_accuracy": 0.8043633893634496,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 89.3359,
+      "eval_samples_per_second": 15.895,
+      "eval_steps_per_second": 1.992,
+      "step": 840
+    },
+    {
+      "entropy": 0.6409128502011299,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 1.5804955959320068,
+      "learning_rate": 0.0003521523160020035,
+      "loss": 0.6023559093475341,
+      "mean_token_accuracy": 0.8222251623868942,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6983234788594621,
+      "eval_loss": 0.7232135534286499,
+      "eval_mean_token_accuracy": 0.8010654268639811,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 89.2615,
+      "eval_samples_per_second": 15.908,
+      "eval_steps_per_second": 1.994,
+      "step": 860
+    },
+    {
+      "entropy": 0.6151149723678827,
+      "epoch": 2.12070006035003,
+      "grad_norm": 1.1975436210632324,
+      "learning_rate": 0.00035100910966837193,
+      "loss": 0.5803564548492431,
+      "mean_token_accuracy": 0.8256325736641884,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6636555573243773,
+      "eval_loss": 0.7205662727355957,
+      "eval_mean_token_accuracy": 0.8036768369460374,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 89.2957,
+      "eval_samples_per_second": 15.902,
+      "eval_steps_per_second": 1.993,
+      "step": 880
+    },
+    {
+      "entropy": 0.6201727617532015,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9907131195068359,
+      "learning_rate": 0.0003498181779509813,
+      "loss": 0.6101872444152832,
+      "mean_token_accuracy": 0.8196512959897518,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.6473788086617931,
+      "eval_loss": 0.7062397003173828,
+      "eval_mean_token_accuracy": 0.8078362184963869,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 88.6198,
+      "eval_samples_per_second": 16.024,
+      "eval_steps_per_second": 2.009,
+      "step": 900
+    },
+    {
+      "entropy": 0.6097332935780286,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.9055633544921875,
+      "learning_rate": 0.00034857985786942026,
+      "loss": 0.5956094264984131,
+      "mean_token_accuracy": 0.8240575887262821,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.612629868341296,
+      "eval_loss": 0.7106770277023315,
+      "eval_mean_token_accuracy": 0.805324277181304,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 89.4383,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 920
+    },
+    {
+      "entropy": 0.6182701248675585,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.9513736367225647,
+      "learning_rate": 0.00034729449985362404,
+      "loss": 0.6019269466400147,
+      "mean_token_accuracy": 0.8187106013298034,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6437647456533453,
+      "eval_loss": 0.7173502445220947,
+      "eval_mean_token_accuracy": 0.8056439573175451,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.9568,
+      "eval_samples_per_second": 15.963,
+      "eval_steps_per_second": 2.001,
+      "step": 940
+    },
+    {
+      "entropy": 0.6508570514619351,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.9606207013130188,
+      "learning_rate": 0.0003459624676447067,
+      "loss": 0.6296409606933594,
+      "mean_token_accuracy": 0.8167732007801533,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6288247135248077,
+      "eval_loss": 0.7160675525665283,
+      "eval_mean_token_accuracy": 0.8066010689467527,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 89.0044,
+      "eval_samples_per_second": 15.954,
+      "eval_steps_per_second": 2.0,
+      "step": 960
+    },
+    {
+      "entropy": 0.6380081418901682,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.9677181839942932,
+      "learning_rate": 0.000344584138192027,
+      "loss": 0.6210546493530273,
+      "mean_token_accuracy": 0.8186135642230511,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6407805031604981,
+      "eval_loss": 0.7055043578147888,
+      "eval_mean_token_accuracy": 0.8070103157772107,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 89.5606,
+      "eval_samples_per_second": 15.855,
+      "eval_steps_per_second": 1.987,
+      "step": 980
+    },
+    {
+      "entropy": 0.6310251638293266,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 2.1046407222747803,
+      "learning_rate": 0.000343159901546516,
+      "loss": 0.6165830135345459,
+      "mean_token_accuracy": 0.8207855716347694,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6461507470420237,
+      "eval_loss": 0.7292136549949646,
+      "eval_mean_token_accuracy": 0.8030418261383356,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 89.315,
+      "eval_samples_per_second": 15.899,
+      "eval_steps_per_second": 1.993,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6256617344915867,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 1.1768349409103394,
+      "learning_rate": 0.0003416901607502972,
+      "loss": 0.6142709255218506,
+      "mean_token_accuracy": 0.8220669947564602,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6150523993406403,
+      "eval_loss": 0.7019563317298889,
+      "eval_mean_token_accuracy": 0.8094323099998946,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 89.419,
+      "eval_samples_per_second": 15.88,
+      "eval_steps_per_second": 1.991,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6126735735684633,
+      "epoch": 2.506940253470127,
+      "grad_norm": 1.8065496683120728,
+      "learning_rate": 0.00034017533172263055,
+      "loss": 0.6056458473205566,
+      "mean_token_accuracy": 0.8203762136399746,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6544345456562685,
+      "eval_loss": 0.7086517214775085,
+      "eval_mean_token_accuracy": 0.806709805901131,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 88.981,
+      "eval_samples_per_second": 15.958,
+      "eval_steps_per_second": 2.0,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6311754353344441,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 1.0576369762420654,
+      "learning_rate": 0.00033861584314221236,
+      "loss": 0.6176392555236816,
+      "mean_token_accuracy": 0.8177410490810871,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6273799200406235,
+      "eval_loss": 0.708163321018219,
+      "eval_mean_token_accuracy": 0.8032017099053672,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 89.7668,
+      "eval_samples_per_second": 15.819,
+      "eval_steps_per_second": 1.983,
+      "step": 1060
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.0627093642584269e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1080/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1080/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1080/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1080/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1080/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..7ce4b072fcbe5b636867e13937c56fedee3b68b4
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1080/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.02728394274790723,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1080/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1080/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1080/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1080/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1080/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1080/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1080/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1080/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..0257f34da4b8f3e993d73c4d394376bfa5547352
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1080/trainer_state.json
@@ -0,0 +1,1168 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 2.603500301750151,
+  "eval_steps": 20,
+  "global_step": 1080,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.024795526266098,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.945375919342041,
+      "learning_rate": 1.6698127428345936e-05,
+      "loss": 1.7825420379638672,
+      "mean_token_accuracy": 0.6315708436071873,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.29921497588747,
+      "eval_loss": 1.169131875038147,
+      "eval_mean_token_accuracy": 0.7185706342204233,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 91.7898,
+      "eval_samples_per_second": 15.47,
+      "eval_steps_per_second": 1.939,
+      "step": 20
+    },
+    {
+      "entropy": 1.0346344470977784,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.7916979789733887,
+      "learning_rate": 3.427510366871008e-05,
+      "loss": 0.9480395317077637,
+      "mean_token_accuracy": 0.7543429024517536,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 0.9673695430326997,
+      "eval_loss": 0.8588430285453796,
+      "eval_mean_token_accuracy": 0.7737233002534073,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 89.8219,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 40
+    },
+    {
+      "entropy": 0.8977566614747048,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7972187995910645,
+      "learning_rate": 5.185207990907422e-05,
+      "loss": 0.8029604911804199,
+      "mean_token_accuracy": 0.7818016350269318,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8436836982041263,
+      "eval_loss": 0.7893033623695374,
+      "eval_mean_token_accuracy": 0.7858914984076211,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 89.9349,
+      "eval_samples_per_second": 15.789,
+      "eval_steps_per_second": 1.979,
+      "step": 60
+    },
+    {
+      "entropy": 0.8650321319699288,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.2128729820251465,
+      "learning_rate": 6.942905614943836e-05,
+      "loss": 0.77417893409729,
+      "mean_token_accuracy": 0.7875479467213153,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8209857180547179,
+      "eval_loss": 0.7572280168533325,
+      "eval_mean_token_accuracy": 0.7925970383574453,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 89.9459,
+      "eval_samples_per_second": 15.787,
+      "eval_steps_per_second": 1.979,
+      "step": 80
+    },
+    {
+      "entropy": 0.8096660938113928,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.1254287958145142,
+      "learning_rate": 8.700603238980251e-05,
+      "loss": 0.7170331001281738,
+      "mean_token_accuracy": 0.7994555257260799,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.792780208788561,
+      "eval_loss": 0.7442984580993652,
+      "eval_mean_token_accuracy": 0.7954704573984896,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 90.294,
+      "eval_samples_per_second": 15.726,
+      "eval_steps_per_second": 1.971,
+      "step": 100
+    },
+    {
+      "entropy": 0.7979829199612141,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2140507698059082,
+      "learning_rate": 0.00010458300863016666,
+      "loss": 0.7141861915588379,
+      "mean_token_accuracy": 0.8000014744699001,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.7938692486018277,
+      "eval_loss": 0.7279797196388245,
+      "eval_mean_token_accuracy": 0.7989798734027348,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 89.7189,
+      "eval_samples_per_second": 15.827,
+      "eval_steps_per_second": 1.984,
+      "step": 120
+    },
+    {
+      "entropy": 0.7916347607970238,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.0506458282470703,
+      "learning_rate": 0.0001221599848705308,
+      "loss": 0.7117055416107178,
+      "mean_token_accuracy": 0.7970046654343605,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.8052781063519167,
+      "eval_loss": 0.722576916217804,
+      "eval_mean_token_accuracy": 0.7982530523551984,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 90.0376,
+      "eval_samples_per_second": 15.771,
+      "eval_steps_per_second": 1.977,
+      "step": 140
+    },
+    {
+      "entropy": 0.7880666613578796,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8409207463264465,
+      "learning_rate": 0.00013973696111089492,
+      "loss": 0.7052523136138916,
+      "mean_token_accuracy": 0.8000123649835587,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7787443134891853,
+      "eval_loss": 0.7137772440910339,
+      "eval_mean_token_accuracy": 0.8014209287220173,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 90.2562,
+      "eval_samples_per_second": 15.733,
+      "eval_steps_per_second": 1.972,
+      "step": 160
+    },
+    {
+      "entropy": 0.7696007996797561,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.0182477235794067,
+      "learning_rate": 0.00015731393735125907,
+      "loss": 0.7008797645568847,
+      "mean_token_accuracy": 0.8007259473204613,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.752259682068664,
+      "eval_loss": 0.7171286344528198,
+      "eval_mean_token_accuracy": 0.80052122574174,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 90.0797,
+      "eval_samples_per_second": 15.764,
+      "eval_steps_per_second": 1.976,
+      "step": 180
+    },
+    {
+      "entropy": 0.7828126326203346,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9553632736206055,
+      "learning_rate": 0.0001748909135916232,
+      "loss": 0.7046853542327881,
+      "mean_token_accuracy": 0.7988562889397144,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7389638169427936,
+      "eval_loss": 0.7187935709953308,
+      "eval_mean_token_accuracy": 0.7995899999409579,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 89.9223,
+      "eval_samples_per_second": 15.791,
+      "eval_steps_per_second": 1.979,
+      "step": 200
+    },
+    {
+      "entropy": 0.7909286297857762,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 1.134459376335144,
+      "learning_rate": 0.00019246788983198735,
+      "loss": 0.6989312171936035,
+      "mean_token_accuracy": 0.7991565138101577,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7496965199374082,
+      "eval_loss": 0.7106321454048157,
+      "eval_mean_token_accuracy": 0.8046183442131857,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 89.6899,
+      "eval_samples_per_second": 15.832,
+      "eval_steps_per_second": 1.985,
+      "step": 220
+    },
+    {
+      "entropy": 0.7651963323354721,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.9557585716247559,
+      "learning_rate": 0.0002100448660723515,
+      "loss": 0.6984559059143066,
+      "mean_token_accuracy": 0.8019823372364044,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7615563079212488,
+      "eval_loss": 0.7098539471626282,
+      "eval_mean_token_accuracy": 0.8047233508544022,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 90.3752,
+      "eval_samples_per_second": 15.712,
+      "eval_steps_per_second": 1.97,
+      "step": 240
+    },
+    {
+      "entropy": 0.787507726252079,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.9279311299324036,
+      "learning_rate": 0.00022762184231271568,
+      "loss": 0.708138370513916,
+      "mean_token_accuracy": 0.8013308703899383,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7856640896100676,
+      "eval_loss": 0.7191570401191711,
+      "eval_mean_token_accuracy": 0.8024677249153008,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 93.7336,
+      "eval_samples_per_second": 15.149,
+      "eval_steps_per_second": 1.899,
+      "step": 260
+    },
+    {
+      "entropy": 0.768489234894514,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.7992776036262512,
+      "learning_rate": 0.0002451988185530798,
+      "loss": 0.7035849571228028,
+      "mean_token_accuracy": 0.8031105332076549,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7682498740346244,
+      "eval_loss": 0.7149022221565247,
+      "eval_mean_token_accuracy": 0.7973497995499814,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 89.7309,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 280
+    },
+    {
+      "entropy": 0.7720955617725849,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.1931514739990234,
+      "learning_rate": 0.0002627757947934439,
+      "loss": 0.6965402603149414,
+      "mean_token_accuracy": 0.8040348663926125,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7214708304807042,
+      "eval_loss": 0.7350823879241943,
+      "eval_mean_token_accuracy": 0.7942911925610532,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 89.6451,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 300
+    },
+    {
+      "entropy": 0.7930095426738262,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8994467258453369,
+      "learning_rate": 0.0002803527710338081,
+      "loss": 0.7331492900848389,
+      "mean_token_accuracy": 0.796185302734375,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.8142535488926963,
+      "eval_loss": 0.7237834334373474,
+      "eval_mean_token_accuracy": 0.799495273761535,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 90.6759,
+      "eval_samples_per_second": 15.66,
+      "eval_steps_per_second": 1.963,
+      "step": 320
+    },
+    {
+      "entropy": 0.7825102441012859,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 1.8224239349365234,
+      "learning_rate": 0.00029792974727417223,
+      "loss": 0.7331175804138184,
+      "mean_token_accuracy": 0.7977667152881622,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7327135443017724,
+      "eval_loss": 0.7388784289360046,
+      "eval_mean_token_accuracy": 0.8007041252730938,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 89.7692,
+      "eval_samples_per_second": 15.818,
+      "eval_steps_per_second": 1.983,
+      "step": 340
+    },
+    {
+      "entropy": 0.819331557303667,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 1.0407779216766357,
+      "learning_rate": 0.0003155067235145364,
+      "loss": 0.7456695556640625,
+      "mean_token_accuracy": 0.7936980701982975,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.8065585420372781,
+      "eval_loss": 0.735519528388977,
+      "eval_mean_token_accuracy": 0.7974206296245704,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 89.4336,
+      "eval_samples_per_second": 15.878,
+      "eval_steps_per_second": 1.99,
+      "step": 360
+    },
+    {
+      "entropy": 0.8114653021097183,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.1209490299224854,
+      "learning_rate": 0.0003330836997549005,
+      "loss": 0.7340899467468261,
+      "mean_token_accuracy": 0.7914272703230381,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.8025527285056168,
+      "eval_loss": 0.7295576930046082,
+      "eval_mean_token_accuracy": 0.7990576628218876,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 89.9579,
+      "eval_samples_per_second": 15.785,
+      "eval_steps_per_second": 1.979,
+      "step": 380
+    },
+    {
+      "entropy": 0.8042497783899307,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 1.2437798976898193,
+      "learning_rate": 0.0003506606759952647,
+      "loss": 0.7381744384765625,
+      "mean_token_accuracy": 0.7967362694442273,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7717011336530193,
+      "eval_loss": 0.7436173558235168,
+      "eval_mean_token_accuracy": 0.7975849482450592,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 89.8411,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 400
+    },
+    {
+      "entropy": 0.7987750010056929,
+      "epoch": 1.012070006035003,
+      "grad_norm": 1.2390918731689453,
+      "learning_rate": 0.000364721224843344,
+      "loss": 0.7317790031433106,
+      "mean_token_accuracy": 0.7986087551364651,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7333170604170038,
+      "eval_loss": 0.7456948757171631,
+      "eval_mean_token_accuracy": 0.7964897577682238,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 89.7288,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 420
+    },
+    {
+      "entropy": 0.7372976846992969,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.1552740335464478,
+      "learning_rate": 0.00036468510102269575,
+      "loss": 0.6734249114990234,
+      "mean_token_accuracy": 0.805834824591875,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.7130355526891987,
+      "eval_loss": 0.7555333375930786,
+      "eval_mean_token_accuracy": 0.7961652781186479,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 89.8713,
+      "eval_samples_per_second": 15.8,
+      "eval_steps_per_second": 1.981,
+      "step": 440
+    },
+    {
+      "entropy": 0.7465668715536594,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 1.1991469860076904,
+      "learning_rate": 0.0003645973816745066,
+      "loss": 0.6871879577636719,
+      "mean_token_accuracy": 0.8042450234293937,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7772979542110743,
+      "eval_loss": 0.7561826705932617,
+      "eval_mean_token_accuracy": 0.7953876427720102,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 89.8395,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 460
+    },
+    {
+      "entropy": 0.7477855801582336,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 1.325486660003662,
+      "learning_rate": 0.00036445809162231435,
+      "loss": 0.6836830139160156,
+      "mean_token_accuracy": 0.8021619468927383,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.7098110676481483,
+      "eval_loss": 0.758007287979126,
+      "eval_mean_token_accuracy": 0.7909953189030122,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 89.6101,
+      "eval_samples_per_second": 15.846,
+      "eval_steps_per_second": 1.986,
+      "step": 480
+    },
+    {
+      "entropy": 0.7632691070437432,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 1.0122262239456177,
+      "learning_rate": 0.0003642672702835562,
+      "loss": 0.7071213245391845,
+      "mean_token_accuracy": 0.8000680930912495,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7645382720432924,
+      "eval_loss": 0.7492606043815613,
+      "eval_mean_token_accuracy": 0.7976858676149604,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 89.6092,
+      "eval_samples_per_second": 15.847,
+      "eval_steps_per_second": 1.986,
+      "step": 500
+    },
+    {
+      "entropy": 0.7623726457357407,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.367774486541748,
+      "learning_rate": 0.00036402497165841384,
+      "loss": 0.6965175151824952,
+      "mean_token_accuracy": 0.8016501650214195,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7868083715438843,
+      "eval_loss": 0.7481449842453003,
+      "eval_mean_token_accuracy": 0.799330582779445,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 89.8241,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 520
+    },
+    {
+      "entropy": 0.7535207405686378,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 1.213614821434021,
+      "learning_rate": 0.00036373126431453207,
+      "loss": 0.7078310489654541,
+      "mean_token_accuracy": 0.7998907208442688,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7829091950748743,
+      "eval_loss": 0.7416072487831116,
+      "eval_mean_token_accuracy": 0.7969146250339036,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 89.4607,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 540
+    },
+    {
+      "entropy": 0.7527363449335098,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 1.0929982662200928,
+      "learning_rate": 0.00036338623136761495,
+      "loss": 0.7058369636535644,
+      "mean_token_accuracy": 0.7997087433934211,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.728419938114252,
+      "eval_loss": 0.744873046875,
+      "eval_mean_token_accuracy": 0.7966634296299366,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 89.6597,
+      "eval_samples_per_second": 15.838,
+      "eval_steps_per_second": 1.985,
+      "step": 560
+    },
+    {
+      "entropy": 0.7406099259853363,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.409364104270935,
+      "learning_rate": 0.00036298997045790515,
+      "loss": 0.6902852058410645,
+      "mean_token_accuracy": 0.805386807769537,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.6991632758231645,
+      "eval_loss": 0.7480319738388062,
+      "eval_mean_token_accuracy": 0.7970068006033308,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 89.8329,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 580
+    },
+    {
+      "entropy": 0.7459449704736472,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.7053111791610718,
+      "learning_rate": 0.00036254259372255277,
+      "loss": 0.7044640064239502,
+      "mean_token_accuracy": 0.8005018755793571,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7454566660891758,
+      "eval_loss": 0.7551484704017639,
+      "eval_mean_token_accuracy": 0.7897093731365846,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 89.8313,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 600
+    },
+    {
+      "entropy": 0.7621391348540782,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 1.174443244934082,
+      "learning_rate": 0.000362044227763882,
+      "loss": 0.7111573696136475,
+      "mean_token_accuracy": 0.798752411454916,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7664387949397055,
+      "eval_loss": 0.7365804314613342,
+      "eval_mean_token_accuracy": 0.7963101200843126,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 89.9623,
+      "eval_samples_per_second": 15.784,
+      "eval_steps_per_second": 1.979,
+      "step": 620
+    },
+    {
+      "entropy": 0.7424944408237935,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.2898184061050415,
+      "learning_rate": 0.000361495013613564,
+      "loss": 0.6913110733032226,
+      "mean_token_accuracy": 0.8009015507996082,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7255479529332579,
+      "eval_loss": 0.7345991730690002,
+      "eval_mean_token_accuracy": 0.8013592945056015,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 89.817,
+      "eval_samples_per_second": 15.81,
+      "eval_steps_per_second": 1.982,
+      "step": 640
+    },
+    {
+      "entropy": 0.7303367637097835,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.1310476064682007,
+      "learning_rate": 0.00036089510669270666,
+      "loss": 0.6875617027282714,
+      "mean_token_accuracy": 0.8035397171974182,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7101308248016271,
+      "eval_loss": 0.7394917011260986,
+      "eval_mean_token_accuracy": 0.7990792942850777,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 89.7431,
+      "eval_samples_per_second": 15.823,
+      "eval_steps_per_second": 1.983,
+      "step": 660
+    },
+    {
+      "entropy": 0.7387678384780884,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.9645389914512634,
+      "learning_rate": 0.0003602446767678725,
+      "loss": 0.6911204338073731,
+      "mean_token_accuracy": 0.8056333385407924,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7281073737010527,
+      "eval_loss": 0.7376708984375,
+      "eval_mean_token_accuracy": 0.7997356415464637,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 90.0338,
+      "eval_samples_per_second": 15.772,
+      "eval_steps_per_second": 1.977,
+      "step": 680
+    },
+    {
+      "entropy": 0.7449554048478604,
+      "epoch": 1.687990343995172,
+      "grad_norm": 1.9069629907608032,
+      "learning_rate": 0.0003595439079030364,
+      "loss": 0.7042286396026611,
+      "mean_token_accuracy": 0.8018639378249646,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7022633817088738,
+      "eval_loss": 0.7320939898490906,
+      "eval_mean_token_accuracy": 0.8006051541044471,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 89.8948,
+      "eval_samples_per_second": 15.796,
+      "eval_steps_per_second": 1.98,
+      "step": 700
+    },
+    {
+      "entropy": 0.723249051719904,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.8333877325057983,
+      "learning_rate": 0.00035879299840749777,
+      "loss": 0.7043714046478271,
+      "mean_token_accuracy": 0.8028345607221127,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.672436946898364,
+      "eval_loss": 0.713233470916748,
+      "eval_mean_token_accuracy": 0.8047760577684038,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 89.6394,
+      "eval_samples_per_second": 15.841,
+      "eval_steps_per_second": 1.986,
+      "step": 720
+    },
+    {
+      "entropy": 0.730014206469059,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.4580037593841553,
+      "learning_rate": 0.00035799216077976137,
+      "loss": 0.6985176563262939,
+      "mean_token_accuracy": 0.8015547141432762,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.6908873795123582,
+      "eval_loss": 0.7198202013969421,
+      "eval_mean_token_accuracy": 0.8024131682481659,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 89.3771,
+      "eval_samples_per_second": 15.888,
+      "eval_steps_per_second": 1.992,
+      "step": 740
+    },
+    {
+      "entropy": 0.7219605796039105,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.51682710647583,
+      "learning_rate": 0.000357141621647403,
+      "loss": 0.6844424724578857,
+      "mean_token_accuracy": 0.8032813094556331,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7113401109582922,
+      "eval_loss": 0.7195309400558472,
+      "eval_mean_token_accuracy": 0.8019187035185568,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 89.4658,
+      "eval_samples_per_second": 15.872,
+      "eval_steps_per_second": 1.99,
+      "step": 760
+    },
+    {
+      "entropy": 0.729826743900776,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 1.0838000774383545,
+      "learning_rate": 0.0003562416217029361,
+      "loss": 0.685858678817749,
+      "mean_token_accuracy": 0.8050567395985126,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.649370267484965,
+      "eval_loss": 0.7177029252052307,
+      "eval_mean_token_accuracy": 0.8050174043419656,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 89.2287,
+      "eval_samples_per_second": 15.914,
+      "eval_steps_per_second": 1.995,
+      "step": 780
+    },
+    {
+      "entropy": 0.6799364153295755,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 1.1364924907684326,
+      "learning_rate": 0.00035529241563569903,
+      "loss": 0.6714428901672364,
+      "mean_token_accuracy": 0.8087341658771038,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6603799047095052,
+      "eval_loss": 0.7201809287071228,
+      "eval_mean_token_accuracy": 0.802863868769635,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 89.2718,
+      "eval_samples_per_second": 15.906,
+      "eval_steps_per_second": 1.994,
+      "step": 800
+    },
+    {
+      "entropy": 0.7078610375523567,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 1.2191942930221558,
+      "learning_rate": 0.0003542942720597808,
+      "loss": 0.6996590614318847,
+      "mean_token_accuracy": 0.8049512051045895,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.7240545264120852,
+      "eval_loss": 0.7301309108734131,
+      "eval_mean_token_accuracy": 0.8015973166133581,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 89.3438,
+      "eval_samples_per_second": 15.894,
+      "eval_steps_per_second": 1.992,
+      "step": 820
+    },
+    {
+      "entropy": 0.6557391556826505,
+      "epoch": 2.024140012070006,
+      "grad_norm": 1.36203134059906,
+      "learning_rate": 0.0003532474734380064,
+      "loss": 0.6327113628387451,
+      "mean_token_accuracy": 0.8142829847026181,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6573808648613062,
+      "eval_loss": 0.7334365844726562,
+      "eval_mean_token_accuracy": 0.8043633893634496,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 89.3359,
+      "eval_samples_per_second": 15.895,
+      "eval_steps_per_second": 1.992,
+      "step": 840
+    },
+    {
+      "entropy": 0.6409128502011299,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 1.5804955959320068,
+      "learning_rate": 0.0003521523160020035,
+      "loss": 0.6023559093475341,
+      "mean_token_accuracy": 0.8222251623868942,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6983234788594621,
+      "eval_loss": 0.7232135534286499,
+      "eval_mean_token_accuracy": 0.8010654268639811,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 89.2615,
+      "eval_samples_per_second": 15.908,
+      "eval_steps_per_second": 1.994,
+      "step": 860
+    },
+    {
+      "entropy": 0.6151149723678827,
+      "epoch": 2.12070006035003,
+      "grad_norm": 1.1975436210632324,
+      "learning_rate": 0.00035100910966837193,
+      "loss": 0.5803564548492431,
+      "mean_token_accuracy": 0.8256325736641884,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6636555573243773,
+      "eval_loss": 0.7205662727355957,
+      "eval_mean_token_accuracy": 0.8036768369460374,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 89.2957,
+      "eval_samples_per_second": 15.902,
+      "eval_steps_per_second": 1.993,
+      "step": 880
+    },
+    {
+      "entropy": 0.6201727617532015,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9907131195068359,
+      "learning_rate": 0.0003498181779509813,
+      "loss": 0.6101872444152832,
+      "mean_token_accuracy": 0.8196512959897518,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.6473788086617931,
+      "eval_loss": 0.7062397003173828,
+      "eval_mean_token_accuracy": 0.8078362184963869,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 88.6198,
+      "eval_samples_per_second": 16.024,
+      "eval_steps_per_second": 2.009,
+      "step": 900
+    },
+    {
+      "entropy": 0.6097332935780286,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.9055633544921875,
+      "learning_rate": 0.00034857985786942026,
+      "loss": 0.5956094264984131,
+      "mean_token_accuracy": 0.8240575887262821,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.612629868341296,
+      "eval_loss": 0.7106770277023315,
+      "eval_mean_token_accuracy": 0.805324277181304,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 89.4383,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 920
+    },
+    {
+      "entropy": 0.6182701248675585,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.9513736367225647,
+      "learning_rate": 0.00034729449985362404,
+      "loss": 0.6019269466400147,
+      "mean_token_accuracy": 0.8187106013298034,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6437647456533453,
+      "eval_loss": 0.7173502445220947,
+      "eval_mean_token_accuracy": 0.8056439573175451,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.9568,
+      "eval_samples_per_second": 15.963,
+      "eval_steps_per_second": 2.001,
+      "step": 940
+    },
+    {
+      "entropy": 0.6508570514619351,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.9606207013130188,
+      "learning_rate": 0.0003459624676447067,
+      "loss": 0.6296409606933594,
+      "mean_token_accuracy": 0.8167732007801533,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6288247135248077,
+      "eval_loss": 0.7160675525665283,
+      "eval_mean_token_accuracy": 0.8066010689467527,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 89.0044,
+      "eval_samples_per_second": 15.954,
+      "eval_steps_per_second": 2.0,
+      "step": 960
+    },
+    {
+      "entropy": 0.6380081418901682,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.9677181839942932,
+      "learning_rate": 0.000344584138192027,
+      "loss": 0.6210546493530273,
+      "mean_token_accuracy": 0.8186135642230511,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6407805031604981,
+      "eval_loss": 0.7055043578147888,
+      "eval_mean_token_accuracy": 0.8070103157772107,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 89.5606,
+      "eval_samples_per_second": 15.855,
+      "eval_steps_per_second": 1.987,
+      "step": 980
+    },
+    {
+      "entropy": 0.6310251638293266,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 2.1046407222747803,
+      "learning_rate": 0.000343159901546516,
+      "loss": 0.6165830135345459,
+      "mean_token_accuracy": 0.8207855716347694,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6461507470420237,
+      "eval_loss": 0.7292136549949646,
+      "eval_mean_token_accuracy": 0.8030418261383356,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 89.315,
+      "eval_samples_per_second": 15.899,
+      "eval_steps_per_second": 1.993,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6256617344915867,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 1.1768349409103394,
+      "learning_rate": 0.0003416901607502972,
+      "loss": 0.6142709255218506,
+      "mean_token_accuracy": 0.8220669947564602,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6150523993406403,
+      "eval_loss": 0.7019563317298889,
+      "eval_mean_token_accuracy": 0.8094323099998946,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 89.419,
+      "eval_samples_per_second": 15.88,
+      "eval_steps_per_second": 1.991,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6126735735684633,
+      "epoch": 2.506940253470127,
+      "grad_norm": 1.8065496683120728,
+      "learning_rate": 0.00034017533172263055,
+      "loss": 0.6056458473205566,
+      "mean_token_accuracy": 0.8203762136399746,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6544345456562685,
+      "eval_loss": 0.7086517214775085,
+      "eval_mean_token_accuracy": 0.806709805901131,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 88.981,
+      "eval_samples_per_second": 15.958,
+      "eval_steps_per_second": 2.0,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6311754353344441,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 1.0576369762420654,
+      "learning_rate": 0.00033861584314221236,
+      "loss": 0.6176392555236816,
+      "mean_token_accuracy": 0.8177410490810871,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6273799200406235,
+      "eval_loss": 0.708163321018219,
+      "eval_mean_token_accuracy": 0.8032017099053672,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 89.7668,
+      "eval_samples_per_second": 15.819,
+      "eval_steps_per_second": 1.983,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6200249589979648,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.997065007686615,
+      "learning_rate": 0.0003370121363258637,
+      "loss": 0.6203888893127442,
+      "mean_token_accuracy": 0.8196275025606156,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6457576811983344,
+      "eval_loss": 0.6972290277481079,
+      "eval_mean_token_accuracy": 0.8095980849158898,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 89.2027,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 1080
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.0829722584469709e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1100/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1100/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1100/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1100/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1100/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..7ce4b072fcbe5b636867e13937c56fedee3b68b4
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1100/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.02728394274790723,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1100/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1100/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1100/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1100/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1100/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1100/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1100/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1100/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..275a789139d75e8da3fd3217959dd0f26bcf863f
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1100/trainer_state.json
@@ -0,0 +1,1189 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 2.651780325890163,
+  "eval_steps": 20,
+  "global_step": 1100,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.024795526266098,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.945375919342041,
+      "learning_rate": 1.6698127428345936e-05,
+      "loss": 1.7825420379638672,
+      "mean_token_accuracy": 0.6315708436071873,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.29921497588747,
+      "eval_loss": 1.169131875038147,
+      "eval_mean_token_accuracy": 0.7185706342204233,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 91.7898,
+      "eval_samples_per_second": 15.47,
+      "eval_steps_per_second": 1.939,
+      "step": 20
+    },
+    {
+      "entropy": 1.0346344470977784,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.7916979789733887,
+      "learning_rate": 3.427510366871008e-05,
+      "loss": 0.9480395317077637,
+      "mean_token_accuracy": 0.7543429024517536,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 0.9673695430326997,
+      "eval_loss": 0.8588430285453796,
+      "eval_mean_token_accuracy": 0.7737233002534073,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 89.8219,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 40
+    },
+    {
+      "entropy": 0.8977566614747048,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7972187995910645,
+      "learning_rate": 5.185207990907422e-05,
+      "loss": 0.8029604911804199,
+      "mean_token_accuracy": 0.7818016350269318,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8436836982041263,
+      "eval_loss": 0.7893033623695374,
+      "eval_mean_token_accuracy": 0.7858914984076211,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 89.9349,
+      "eval_samples_per_second": 15.789,
+      "eval_steps_per_second": 1.979,
+      "step": 60
+    },
+    {
+      "entropy": 0.8650321319699288,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.2128729820251465,
+      "learning_rate": 6.942905614943836e-05,
+      "loss": 0.77417893409729,
+      "mean_token_accuracy": 0.7875479467213153,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8209857180547179,
+      "eval_loss": 0.7572280168533325,
+      "eval_mean_token_accuracy": 0.7925970383574453,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 89.9459,
+      "eval_samples_per_second": 15.787,
+      "eval_steps_per_second": 1.979,
+      "step": 80
+    },
+    {
+      "entropy": 0.8096660938113928,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.1254287958145142,
+      "learning_rate": 8.700603238980251e-05,
+      "loss": 0.7170331001281738,
+      "mean_token_accuracy": 0.7994555257260799,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.792780208788561,
+      "eval_loss": 0.7442984580993652,
+      "eval_mean_token_accuracy": 0.7954704573984896,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 90.294,
+      "eval_samples_per_second": 15.726,
+      "eval_steps_per_second": 1.971,
+      "step": 100
+    },
+    {
+      "entropy": 0.7979829199612141,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2140507698059082,
+      "learning_rate": 0.00010458300863016666,
+      "loss": 0.7141861915588379,
+      "mean_token_accuracy": 0.8000014744699001,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.7938692486018277,
+      "eval_loss": 0.7279797196388245,
+      "eval_mean_token_accuracy": 0.7989798734027348,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 89.7189,
+      "eval_samples_per_second": 15.827,
+      "eval_steps_per_second": 1.984,
+      "step": 120
+    },
+    {
+      "entropy": 0.7916347607970238,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.0506458282470703,
+      "learning_rate": 0.0001221599848705308,
+      "loss": 0.7117055416107178,
+      "mean_token_accuracy": 0.7970046654343605,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.8052781063519167,
+      "eval_loss": 0.722576916217804,
+      "eval_mean_token_accuracy": 0.7982530523551984,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 90.0376,
+      "eval_samples_per_second": 15.771,
+      "eval_steps_per_second": 1.977,
+      "step": 140
+    },
+    {
+      "entropy": 0.7880666613578796,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8409207463264465,
+      "learning_rate": 0.00013973696111089492,
+      "loss": 0.7052523136138916,
+      "mean_token_accuracy": 0.8000123649835587,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7787443134891853,
+      "eval_loss": 0.7137772440910339,
+      "eval_mean_token_accuracy": 0.8014209287220173,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 90.2562,
+      "eval_samples_per_second": 15.733,
+      "eval_steps_per_second": 1.972,
+      "step": 160
+    },
+    {
+      "entropy": 0.7696007996797561,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.0182477235794067,
+      "learning_rate": 0.00015731393735125907,
+      "loss": 0.7008797645568847,
+      "mean_token_accuracy": 0.8007259473204613,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.752259682068664,
+      "eval_loss": 0.7171286344528198,
+      "eval_mean_token_accuracy": 0.80052122574174,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 90.0797,
+      "eval_samples_per_second": 15.764,
+      "eval_steps_per_second": 1.976,
+      "step": 180
+    },
+    {
+      "entropy": 0.7828126326203346,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9553632736206055,
+      "learning_rate": 0.0001748909135916232,
+      "loss": 0.7046853542327881,
+      "mean_token_accuracy": 0.7988562889397144,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7389638169427936,
+      "eval_loss": 0.7187935709953308,
+      "eval_mean_token_accuracy": 0.7995899999409579,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 89.9223,
+      "eval_samples_per_second": 15.791,
+      "eval_steps_per_second": 1.979,
+      "step": 200
+    },
+    {
+      "entropy": 0.7909286297857762,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 1.134459376335144,
+      "learning_rate": 0.00019246788983198735,
+      "loss": 0.6989312171936035,
+      "mean_token_accuracy": 0.7991565138101577,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7496965199374082,
+      "eval_loss": 0.7106321454048157,
+      "eval_mean_token_accuracy": 0.8046183442131857,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 89.6899,
+      "eval_samples_per_second": 15.832,
+      "eval_steps_per_second": 1.985,
+      "step": 220
+    },
+    {
+      "entropy": 0.7651963323354721,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.9557585716247559,
+      "learning_rate": 0.0002100448660723515,
+      "loss": 0.6984559059143066,
+      "mean_token_accuracy": 0.8019823372364044,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7615563079212488,
+      "eval_loss": 0.7098539471626282,
+      "eval_mean_token_accuracy": 0.8047233508544022,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 90.3752,
+      "eval_samples_per_second": 15.712,
+      "eval_steps_per_second": 1.97,
+      "step": 240
+    },
+    {
+      "entropy": 0.787507726252079,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.9279311299324036,
+      "learning_rate": 0.00022762184231271568,
+      "loss": 0.708138370513916,
+      "mean_token_accuracy": 0.8013308703899383,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7856640896100676,
+      "eval_loss": 0.7191570401191711,
+      "eval_mean_token_accuracy": 0.8024677249153008,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 93.7336,
+      "eval_samples_per_second": 15.149,
+      "eval_steps_per_second": 1.899,
+      "step": 260
+    },
+    {
+      "entropy": 0.768489234894514,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.7992776036262512,
+      "learning_rate": 0.0002451988185530798,
+      "loss": 0.7035849571228028,
+      "mean_token_accuracy": 0.8031105332076549,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7682498740346244,
+      "eval_loss": 0.7149022221565247,
+      "eval_mean_token_accuracy": 0.7973497995499814,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 89.7309,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 280
+    },
+    {
+      "entropy": 0.7720955617725849,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.1931514739990234,
+      "learning_rate": 0.0002627757947934439,
+      "loss": 0.6965402603149414,
+      "mean_token_accuracy": 0.8040348663926125,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7214708304807042,
+      "eval_loss": 0.7350823879241943,
+      "eval_mean_token_accuracy": 0.7942911925610532,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 89.6451,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 300
+    },
+    {
+      "entropy": 0.7930095426738262,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8994467258453369,
+      "learning_rate": 0.0002803527710338081,
+      "loss": 0.7331492900848389,
+      "mean_token_accuracy": 0.796185302734375,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.8142535488926963,
+      "eval_loss": 0.7237834334373474,
+      "eval_mean_token_accuracy": 0.799495273761535,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 90.6759,
+      "eval_samples_per_second": 15.66,
+      "eval_steps_per_second": 1.963,
+      "step": 320
+    },
+    {
+      "entropy": 0.7825102441012859,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 1.8224239349365234,
+      "learning_rate": 0.00029792974727417223,
+      "loss": 0.7331175804138184,
+      "mean_token_accuracy": 0.7977667152881622,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7327135443017724,
+      "eval_loss": 0.7388784289360046,
+      "eval_mean_token_accuracy": 0.8007041252730938,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 89.7692,
+      "eval_samples_per_second": 15.818,
+      "eval_steps_per_second": 1.983,
+      "step": 340
+    },
+    {
+      "entropy": 0.819331557303667,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 1.0407779216766357,
+      "learning_rate": 0.0003155067235145364,
+      "loss": 0.7456695556640625,
+      "mean_token_accuracy": 0.7936980701982975,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.8065585420372781,
+      "eval_loss": 0.735519528388977,
+      "eval_mean_token_accuracy": 0.7974206296245704,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 89.4336,
+      "eval_samples_per_second": 15.878,
+      "eval_steps_per_second": 1.99,
+      "step": 360
+    },
+    {
+      "entropy": 0.8114653021097183,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.1209490299224854,
+      "learning_rate": 0.0003330836997549005,
+      "loss": 0.7340899467468261,
+      "mean_token_accuracy": 0.7914272703230381,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.8025527285056168,
+      "eval_loss": 0.7295576930046082,
+      "eval_mean_token_accuracy": 0.7990576628218876,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 89.9579,
+      "eval_samples_per_second": 15.785,
+      "eval_steps_per_second": 1.979,
+      "step": 380
+    },
+    {
+      "entropy": 0.8042497783899307,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 1.2437798976898193,
+      "learning_rate": 0.0003506606759952647,
+      "loss": 0.7381744384765625,
+      "mean_token_accuracy": 0.7967362694442273,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7717011336530193,
+      "eval_loss": 0.7436173558235168,
+      "eval_mean_token_accuracy": 0.7975849482450592,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 89.8411,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 400
+    },
+    {
+      "entropy": 0.7987750010056929,
+      "epoch": 1.012070006035003,
+      "grad_norm": 1.2390918731689453,
+      "learning_rate": 0.000364721224843344,
+      "loss": 0.7317790031433106,
+      "mean_token_accuracy": 0.7986087551364651,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7333170604170038,
+      "eval_loss": 0.7456948757171631,
+      "eval_mean_token_accuracy": 0.7964897577682238,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 89.7288,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 420
+    },
+    {
+      "entropy": 0.7372976846992969,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.1552740335464478,
+      "learning_rate": 0.00036468510102269575,
+      "loss": 0.6734249114990234,
+      "mean_token_accuracy": 0.805834824591875,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.7130355526891987,
+      "eval_loss": 0.7555333375930786,
+      "eval_mean_token_accuracy": 0.7961652781186479,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 89.8713,
+      "eval_samples_per_second": 15.8,
+      "eval_steps_per_second": 1.981,
+      "step": 440
+    },
+    {
+      "entropy": 0.7465668715536594,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 1.1991469860076904,
+      "learning_rate": 0.0003645973816745066,
+      "loss": 0.6871879577636719,
+      "mean_token_accuracy": 0.8042450234293937,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7772979542110743,
+      "eval_loss": 0.7561826705932617,
+      "eval_mean_token_accuracy": 0.7953876427720102,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 89.8395,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 460
+    },
+    {
+      "entropy": 0.7477855801582336,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 1.325486660003662,
+      "learning_rate": 0.00036445809162231435,
+      "loss": 0.6836830139160156,
+      "mean_token_accuracy": 0.8021619468927383,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.7098110676481483,
+      "eval_loss": 0.758007287979126,
+      "eval_mean_token_accuracy": 0.7909953189030122,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 89.6101,
+      "eval_samples_per_second": 15.846,
+      "eval_steps_per_second": 1.986,
+      "step": 480
+    },
+    {
+      "entropy": 0.7632691070437432,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 1.0122262239456177,
+      "learning_rate": 0.0003642672702835562,
+      "loss": 0.7071213245391845,
+      "mean_token_accuracy": 0.8000680930912495,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7645382720432924,
+      "eval_loss": 0.7492606043815613,
+      "eval_mean_token_accuracy": 0.7976858676149604,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 89.6092,
+      "eval_samples_per_second": 15.847,
+      "eval_steps_per_second": 1.986,
+      "step": 500
+    },
+    {
+      "entropy": 0.7623726457357407,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.367774486541748,
+      "learning_rate": 0.00036402497165841384,
+      "loss": 0.6965175151824952,
+      "mean_token_accuracy": 0.8016501650214195,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7868083715438843,
+      "eval_loss": 0.7481449842453003,
+      "eval_mean_token_accuracy": 0.799330582779445,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 89.8241,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 520
+    },
+    {
+      "entropy": 0.7535207405686378,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 1.213614821434021,
+      "learning_rate": 0.00036373126431453207,
+      "loss": 0.7078310489654541,
+      "mean_token_accuracy": 0.7998907208442688,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7829091950748743,
+      "eval_loss": 0.7416072487831116,
+      "eval_mean_token_accuracy": 0.7969146250339036,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 89.4607,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 540
+    },
+    {
+      "entropy": 0.7527363449335098,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 1.0929982662200928,
+      "learning_rate": 0.00036338623136761495,
+      "loss": 0.7058369636535644,
+      "mean_token_accuracy": 0.7997087433934211,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.728419938114252,
+      "eval_loss": 0.744873046875,
+      "eval_mean_token_accuracy": 0.7966634296299366,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 89.6597,
+      "eval_samples_per_second": 15.838,
+      "eval_steps_per_second": 1.985,
+      "step": 560
+    },
+    {
+      "entropy": 0.7406099259853363,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.409364104270935,
+      "learning_rate": 0.00036298997045790515,
+      "loss": 0.6902852058410645,
+      "mean_token_accuracy": 0.805386807769537,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.6991632758231645,
+      "eval_loss": 0.7480319738388062,
+      "eval_mean_token_accuracy": 0.7970068006033308,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 89.8329,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 580
+    },
+    {
+      "entropy": 0.7459449704736472,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.7053111791610718,
+      "learning_rate": 0.00036254259372255277,
+      "loss": 0.7044640064239502,
+      "mean_token_accuracy": 0.8005018755793571,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7454566660891758,
+      "eval_loss": 0.7551484704017639,
+      "eval_mean_token_accuracy": 0.7897093731365846,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 89.8313,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 600
+    },
+    {
+      "entropy": 0.7621391348540782,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 1.174443244934082,
+      "learning_rate": 0.000362044227763882,
+      "loss": 0.7111573696136475,
+      "mean_token_accuracy": 0.798752411454916,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7664387949397055,
+      "eval_loss": 0.7365804314613342,
+      "eval_mean_token_accuracy": 0.7963101200843126,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 89.9623,
+      "eval_samples_per_second": 15.784,
+      "eval_steps_per_second": 1.979,
+      "step": 620
+    },
+    {
+      "entropy": 0.7424944408237935,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.2898184061050415,
+      "learning_rate": 0.000361495013613564,
+      "loss": 0.6913110733032226,
+      "mean_token_accuracy": 0.8009015507996082,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7255479529332579,
+      "eval_loss": 0.7345991730690002,
+      "eval_mean_token_accuracy": 0.8013592945056015,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 89.817,
+      "eval_samples_per_second": 15.81,
+      "eval_steps_per_second": 1.982,
+      "step": 640
+    },
+    {
+      "entropy": 0.7303367637097835,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.1310476064682007,
+      "learning_rate": 0.00036089510669270666,
+      "loss": 0.6875617027282714,
+      "mean_token_accuracy": 0.8035397171974182,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7101308248016271,
+      "eval_loss": 0.7394917011260986,
+      "eval_mean_token_accuracy": 0.7990792942850777,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 89.7431,
+      "eval_samples_per_second": 15.823,
+      "eval_steps_per_second": 1.983,
+      "step": 660
+    },
+    {
+      "entropy": 0.7387678384780884,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.9645389914512634,
+      "learning_rate": 0.0003602446767678725,
+      "loss": 0.6911204338073731,
+      "mean_token_accuracy": 0.8056333385407924,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7281073737010527,
+      "eval_loss": 0.7376708984375,
+      "eval_mean_token_accuracy": 0.7997356415464637,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 90.0338,
+      "eval_samples_per_second": 15.772,
+      "eval_steps_per_second": 1.977,
+      "step": 680
+    },
+    {
+      "entropy": 0.7449554048478604,
+      "epoch": 1.687990343995172,
+      "grad_norm": 1.9069629907608032,
+      "learning_rate": 0.0003595439079030364,
+      "loss": 0.7042286396026611,
+      "mean_token_accuracy": 0.8018639378249646,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7022633817088738,
+      "eval_loss": 0.7320939898490906,
+      "eval_mean_token_accuracy": 0.8006051541044471,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 89.8948,
+      "eval_samples_per_second": 15.796,
+      "eval_steps_per_second": 1.98,
+      "step": 700
+    },
+    {
+      "entropy": 0.723249051719904,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.8333877325057983,
+      "learning_rate": 0.00035879299840749777,
+      "loss": 0.7043714046478271,
+      "mean_token_accuracy": 0.8028345607221127,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.672436946898364,
+      "eval_loss": 0.713233470916748,
+      "eval_mean_token_accuracy": 0.8047760577684038,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 89.6394,
+      "eval_samples_per_second": 15.841,
+      "eval_steps_per_second": 1.986,
+      "step": 720
+    },
+    {
+      "entropy": 0.730014206469059,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.4580037593841553,
+      "learning_rate": 0.00035799216077976137,
+      "loss": 0.6985176563262939,
+      "mean_token_accuracy": 0.8015547141432762,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.6908873795123582,
+      "eval_loss": 0.7198202013969421,
+      "eval_mean_token_accuracy": 0.8024131682481659,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 89.3771,
+      "eval_samples_per_second": 15.888,
+      "eval_steps_per_second": 1.992,
+      "step": 740
+    },
+    {
+      "entropy": 0.7219605796039105,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.51682710647583,
+      "learning_rate": 0.000357141621647403,
+      "loss": 0.6844424724578857,
+      "mean_token_accuracy": 0.8032813094556331,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7113401109582922,
+      "eval_loss": 0.7195309400558472,
+      "eval_mean_token_accuracy": 0.8019187035185568,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 89.4658,
+      "eval_samples_per_second": 15.872,
+      "eval_steps_per_second": 1.99,
+      "step": 760
+    },
+    {
+      "entropy": 0.729826743900776,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 1.0838000774383545,
+      "learning_rate": 0.0003562416217029361,
+      "loss": 0.685858678817749,
+      "mean_token_accuracy": 0.8050567395985126,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.649370267484965,
+      "eval_loss": 0.7177029252052307,
+      "eval_mean_token_accuracy": 0.8050174043419656,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 89.2287,
+      "eval_samples_per_second": 15.914,
+      "eval_steps_per_second": 1.995,
+      "step": 780
+    },
+    {
+      "entropy": 0.6799364153295755,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 1.1364924907684326,
+      "learning_rate": 0.00035529241563569903,
+      "loss": 0.6714428901672364,
+      "mean_token_accuracy": 0.8087341658771038,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6603799047095052,
+      "eval_loss": 0.7201809287071228,
+      "eval_mean_token_accuracy": 0.802863868769635,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 89.2718,
+      "eval_samples_per_second": 15.906,
+      "eval_steps_per_second": 1.994,
+      "step": 800
+    },
+    {
+      "entropy": 0.7078610375523567,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 1.2191942930221558,
+      "learning_rate": 0.0003542942720597808,
+      "loss": 0.6996590614318847,
+      "mean_token_accuracy": 0.8049512051045895,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.7240545264120852,
+      "eval_loss": 0.7301309108734131,
+      "eval_mean_token_accuracy": 0.8015973166133581,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 89.3438,
+      "eval_samples_per_second": 15.894,
+      "eval_steps_per_second": 1.992,
+      "step": 820
+    },
+    {
+      "entropy": 0.6557391556826505,
+      "epoch": 2.024140012070006,
+      "grad_norm": 1.36203134059906,
+      "learning_rate": 0.0003532474734380064,
+      "loss": 0.6327113628387451,
+      "mean_token_accuracy": 0.8142829847026181,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6573808648613062,
+      "eval_loss": 0.7334365844726562,
+      "eval_mean_token_accuracy": 0.8043633893634496,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 89.3359,
+      "eval_samples_per_second": 15.895,
+      "eval_steps_per_second": 1.992,
+      "step": 840
+    },
+    {
+      "entropy": 0.6409128502011299,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 1.5804955959320068,
+      "learning_rate": 0.0003521523160020035,
+      "loss": 0.6023559093475341,
+      "mean_token_accuracy": 0.8222251623868942,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6983234788594621,
+      "eval_loss": 0.7232135534286499,
+      "eval_mean_token_accuracy": 0.8010654268639811,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 89.2615,
+      "eval_samples_per_second": 15.908,
+      "eval_steps_per_second": 1.994,
+      "step": 860
+    },
+    {
+      "entropy": 0.6151149723678827,
+      "epoch": 2.12070006035003,
+      "grad_norm": 1.1975436210632324,
+      "learning_rate": 0.00035100910966837193,
+      "loss": 0.5803564548492431,
+      "mean_token_accuracy": 0.8256325736641884,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6636555573243773,
+      "eval_loss": 0.7205662727355957,
+      "eval_mean_token_accuracy": 0.8036768369460374,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 89.2957,
+      "eval_samples_per_second": 15.902,
+      "eval_steps_per_second": 1.993,
+      "step": 880
+    },
+    {
+      "entropy": 0.6201727617532015,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9907131195068359,
+      "learning_rate": 0.0003498181779509813,
+      "loss": 0.6101872444152832,
+      "mean_token_accuracy": 0.8196512959897518,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.6473788086617931,
+      "eval_loss": 0.7062397003173828,
+      "eval_mean_token_accuracy": 0.8078362184963869,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 88.6198,
+      "eval_samples_per_second": 16.024,
+      "eval_steps_per_second": 2.009,
+      "step": 900
+    },
+    {
+      "entropy": 0.6097332935780286,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.9055633544921875,
+      "learning_rate": 0.00034857985786942026,
+      "loss": 0.5956094264984131,
+      "mean_token_accuracy": 0.8240575887262821,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.612629868341296,
+      "eval_loss": 0.7106770277023315,
+      "eval_mean_token_accuracy": 0.805324277181304,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 89.4383,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 920
+    },
+    {
+      "entropy": 0.6182701248675585,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.9513736367225647,
+      "learning_rate": 0.00034729449985362404,
+      "loss": 0.6019269466400147,
+      "mean_token_accuracy": 0.8187106013298034,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6437647456533453,
+      "eval_loss": 0.7173502445220947,
+      "eval_mean_token_accuracy": 0.8056439573175451,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.9568,
+      "eval_samples_per_second": 15.963,
+      "eval_steps_per_second": 2.001,
+      "step": 940
+    },
+    {
+      "entropy": 0.6508570514619351,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.9606207013130188,
+      "learning_rate": 0.0003459624676447067,
+      "loss": 0.6296409606933594,
+      "mean_token_accuracy": 0.8167732007801533,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6288247135248077,
+      "eval_loss": 0.7160675525665283,
+      "eval_mean_token_accuracy": 0.8066010689467527,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 89.0044,
+      "eval_samples_per_second": 15.954,
+      "eval_steps_per_second": 2.0,
+      "step": 960
+    },
+    {
+      "entropy": 0.6380081418901682,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.9677181839942932,
+      "learning_rate": 0.000344584138192027,
+      "loss": 0.6210546493530273,
+      "mean_token_accuracy": 0.8186135642230511,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6407805031604981,
+      "eval_loss": 0.7055043578147888,
+      "eval_mean_token_accuracy": 0.8070103157772107,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 89.5606,
+      "eval_samples_per_second": 15.855,
+      "eval_steps_per_second": 1.987,
+      "step": 980
+    },
+    {
+      "entropy": 0.6310251638293266,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 2.1046407222747803,
+      "learning_rate": 0.000343159901546516,
+      "loss": 0.6165830135345459,
+      "mean_token_accuracy": 0.8207855716347694,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6461507470420237,
+      "eval_loss": 0.7292136549949646,
+      "eval_mean_token_accuracy": 0.8030418261383356,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 89.315,
+      "eval_samples_per_second": 15.899,
+      "eval_steps_per_second": 1.993,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6256617344915867,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 1.1768349409103394,
+      "learning_rate": 0.0003416901607502972,
+      "loss": 0.6142709255218506,
+      "mean_token_accuracy": 0.8220669947564602,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6150523993406403,
+      "eval_loss": 0.7019563317298889,
+      "eval_mean_token_accuracy": 0.8094323099998946,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 89.419,
+      "eval_samples_per_second": 15.88,
+      "eval_steps_per_second": 1.991,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6126735735684633,
+      "epoch": 2.506940253470127,
+      "grad_norm": 1.8065496683120728,
+      "learning_rate": 0.00034017533172263055,
+      "loss": 0.6056458473205566,
+      "mean_token_accuracy": 0.8203762136399746,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6544345456562685,
+      "eval_loss": 0.7086517214775085,
+      "eval_mean_token_accuracy": 0.806709805901131,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 88.981,
+      "eval_samples_per_second": 15.958,
+      "eval_steps_per_second": 2.0,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6311754353344441,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 1.0576369762420654,
+      "learning_rate": 0.00033861584314221236,
+      "loss": 0.6176392555236816,
+      "mean_token_accuracy": 0.8177410490810871,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6273799200406235,
+      "eval_loss": 0.708163321018219,
+      "eval_mean_token_accuracy": 0.8032017099053672,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 89.7668,
+      "eval_samples_per_second": 15.819,
+      "eval_steps_per_second": 1.983,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6200249589979648,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.997065007686615,
+      "learning_rate": 0.0003370121363258637,
+      "loss": 0.6203888893127442,
+      "mean_token_accuracy": 0.8196275025606156,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6457576811983344,
+      "eval_loss": 0.6972290277481079,
+      "eval_mean_token_accuracy": 0.8095980849158898,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 89.2027,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 1080
+    },
+    {
+      "entropy": 0.626810473203659,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.865475058555603,
+      "learning_rate": 0.0003353646651036438,
+      "loss": 0.6102585792541504,
+      "mean_token_accuracy": 0.819708751142025,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.6412022431914726,
+      "eval_loss": 0.7036139965057373,
+      "eval_mean_token_accuracy": 0.8070317135098275,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 89.4733,
+      "eval_samples_per_second": 15.871,
+      "eval_steps_per_second": 1.989,
+      "step": 1100
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.102976649267794e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1120/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1120/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1120/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1120/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1120/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..7ce4b072fcbe5b636867e13937c56fedee3b68b4
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1120/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.02728394274790723,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1120/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1120/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1120/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1120/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1120/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1120/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1120/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1120/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..4198fd69705d2296cbfd15be632f706c18195298
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1120/trainer_state.json
@@ -0,0 +1,1210 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 2.700060350030175,
+  "eval_steps": 20,
+  "global_step": 1120,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.024795526266098,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.945375919342041,
+      "learning_rate": 1.6698127428345936e-05,
+      "loss": 1.7825420379638672,
+      "mean_token_accuracy": 0.6315708436071873,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.29921497588747,
+      "eval_loss": 1.169131875038147,
+      "eval_mean_token_accuracy": 0.7185706342204233,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 91.7898,
+      "eval_samples_per_second": 15.47,
+      "eval_steps_per_second": 1.939,
+      "step": 20
+    },
+    {
+      "entropy": 1.0346344470977784,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.7916979789733887,
+      "learning_rate": 3.427510366871008e-05,
+      "loss": 0.9480395317077637,
+      "mean_token_accuracy": 0.7543429024517536,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 0.9673695430326997,
+      "eval_loss": 0.8588430285453796,
+      "eval_mean_token_accuracy": 0.7737233002534073,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 89.8219,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 40
+    },
+    {
+      "entropy": 0.8977566614747048,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7972187995910645,
+      "learning_rate": 5.185207990907422e-05,
+      "loss": 0.8029604911804199,
+      "mean_token_accuracy": 0.7818016350269318,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8436836982041263,
+      "eval_loss": 0.7893033623695374,
+      "eval_mean_token_accuracy": 0.7858914984076211,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 89.9349,
+      "eval_samples_per_second": 15.789,
+      "eval_steps_per_second": 1.979,
+      "step": 60
+    },
+    {
+      "entropy": 0.8650321319699288,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.2128729820251465,
+      "learning_rate": 6.942905614943836e-05,
+      "loss": 0.77417893409729,
+      "mean_token_accuracy": 0.7875479467213153,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8209857180547179,
+      "eval_loss": 0.7572280168533325,
+      "eval_mean_token_accuracy": 0.7925970383574453,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 89.9459,
+      "eval_samples_per_second": 15.787,
+      "eval_steps_per_second": 1.979,
+      "step": 80
+    },
+    {
+      "entropy": 0.8096660938113928,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.1254287958145142,
+      "learning_rate": 8.700603238980251e-05,
+      "loss": 0.7170331001281738,
+      "mean_token_accuracy": 0.7994555257260799,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.792780208788561,
+      "eval_loss": 0.7442984580993652,
+      "eval_mean_token_accuracy": 0.7954704573984896,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 90.294,
+      "eval_samples_per_second": 15.726,
+      "eval_steps_per_second": 1.971,
+      "step": 100
+    },
+    {
+      "entropy": 0.7979829199612141,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2140507698059082,
+      "learning_rate": 0.00010458300863016666,
+      "loss": 0.7141861915588379,
+      "mean_token_accuracy": 0.8000014744699001,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.7938692486018277,
+      "eval_loss": 0.7279797196388245,
+      "eval_mean_token_accuracy": 0.7989798734027348,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 89.7189,
+      "eval_samples_per_second": 15.827,
+      "eval_steps_per_second": 1.984,
+      "step": 120
+    },
+    {
+      "entropy": 0.7916347607970238,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.0506458282470703,
+      "learning_rate": 0.0001221599848705308,
+      "loss": 0.7117055416107178,
+      "mean_token_accuracy": 0.7970046654343605,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.8052781063519167,
+      "eval_loss": 0.722576916217804,
+      "eval_mean_token_accuracy": 0.7982530523551984,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 90.0376,
+      "eval_samples_per_second": 15.771,
+      "eval_steps_per_second": 1.977,
+      "step": 140
+    },
+    {
+      "entropy": 0.7880666613578796,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8409207463264465,
+      "learning_rate": 0.00013973696111089492,
+      "loss": 0.7052523136138916,
+      "mean_token_accuracy": 0.8000123649835587,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7787443134891853,
+      "eval_loss": 0.7137772440910339,
+      "eval_mean_token_accuracy": 0.8014209287220173,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 90.2562,
+      "eval_samples_per_second": 15.733,
+      "eval_steps_per_second": 1.972,
+      "step": 160
+    },
+    {
+      "entropy": 0.7696007996797561,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.0182477235794067,
+      "learning_rate": 0.00015731393735125907,
+      "loss": 0.7008797645568847,
+      "mean_token_accuracy": 0.8007259473204613,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.752259682068664,
+      "eval_loss": 0.7171286344528198,
+      "eval_mean_token_accuracy": 0.80052122574174,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 90.0797,
+      "eval_samples_per_second": 15.764,
+      "eval_steps_per_second": 1.976,
+      "step": 180
+    },
+    {
+      "entropy": 0.7828126326203346,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9553632736206055,
+      "learning_rate": 0.0001748909135916232,
+      "loss": 0.7046853542327881,
+      "mean_token_accuracy": 0.7988562889397144,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7389638169427936,
+      "eval_loss": 0.7187935709953308,
+      "eval_mean_token_accuracy": 0.7995899999409579,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 89.9223,
+      "eval_samples_per_second": 15.791,
+      "eval_steps_per_second": 1.979,
+      "step": 200
+    },
+    {
+      "entropy": 0.7909286297857762,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 1.134459376335144,
+      "learning_rate": 0.00019246788983198735,
+      "loss": 0.6989312171936035,
+      "mean_token_accuracy": 0.7991565138101577,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7496965199374082,
+      "eval_loss": 0.7106321454048157,
+      "eval_mean_token_accuracy": 0.8046183442131857,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 89.6899,
+      "eval_samples_per_second": 15.832,
+      "eval_steps_per_second": 1.985,
+      "step": 220
+    },
+    {
+      "entropy": 0.7651963323354721,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.9557585716247559,
+      "learning_rate": 0.0002100448660723515,
+      "loss": 0.6984559059143066,
+      "mean_token_accuracy": 0.8019823372364044,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7615563079212488,
+      "eval_loss": 0.7098539471626282,
+      "eval_mean_token_accuracy": 0.8047233508544022,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 90.3752,
+      "eval_samples_per_second": 15.712,
+      "eval_steps_per_second": 1.97,
+      "step": 240
+    },
+    {
+      "entropy": 0.787507726252079,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.9279311299324036,
+      "learning_rate": 0.00022762184231271568,
+      "loss": 0.708138370513916,
+      "mean_token_accuracy": 0.8013308703899383,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7856640896100676,
+      "eval_loss": 0.7191570401191711,
+      "eval_mean_token_accuracy": 0.8024677249153008,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 93.7336,
+      "eval_samples_per_second": 15.149,
+      "eval_steps_per_second": 1.899,
+      "step": 260
+    },
+    {
+      "entropy": 0.768489234894514,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.7992776036262512,
+      "learning_rate": 0.0002451988185530798,
+      "loss": 0.7035849571228028,
+      "mean_token_accuracy": 0.8031105332076549,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7682498740346244,
+      "eval_loss": 0.7149022221565247,
+      "eval_mean_token_accuracy": 0.7973497995499814,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 89.7309,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 280
+    },
+    {
+      "entropy": 0.7720955617725849,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.1931514739990234,
+      "learning_rate": 0.0002627757947934439,
+      "loss": 0.6965402603149414,
+      "mean_token_accuracy": 0.8040348663926125,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7214708304807042,
+      "eval_loss": 0.7350823879241943,
+      "eval_mean_token_accuracy": 0.7942911925610532,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 89.6451,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 300
+    },
+    {
+      "entropy": 0.7930095426738262,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8994467258453369,
+      "learning_rate": 0.0002803527710338081,
+      "loss": 0.7331492900848389,
+      "mean_token_accuracy": 0.796185302734375,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.8142535488926963,
+      "eval_loss": 0.7237834334373474,
+      "eval_mean_token_accuracy": 0.799495273761535,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 90.6759,
+      "eval_samples_per_second": 15.66,
+      "eval_steps_per_second": 1.963,
+      "step": 320
+    },
+    {
+      "entropy": 0.7825102441012859,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 1.8224239349365234,
+      "learning_rate": 0.00029792974727417223,
+      "loss": 0.7331175804138184,
+      "mean_token_accuracy": 0.7977667152881622,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7327135443017724,
+      "eval_loss": 0.7388784289360046,
+      "eval_mean_token_accuracy": 0.8007041252730938,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 89.7692,
+      "eval_samples_per_second": 15.818,
+      "eval_steps_per_second": 1.983,
+      "step": 340
+    },
+    {
+      "entropy": 0.819331557303667,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 1.0407779216766357,
+      "learning_rate": 0.0003155067235145364,
+      "loss": 0.7456695556640625,
+      "mean_token_accuracy": 0.7936980701982975,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.8065585420372781,
+      "eval_loss": 0.735519528388977,
+      "eval_mean_token_accuracy": 0.7974206296245704,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 89.4336,
+      "eval_samples_per_second": 15.878,
+      "eval_steps_per_second": 1.99,
+      "step": 360
+    },
+    {
+      "entropy": 0.8114653021097183,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.1209490299224854,
+      "learning_rate": 0.0003330836997549005,
+      "loss": 0.7340899467468261,
+      "mean_token_accuracy": 0.7914272703230381,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.8025527285056168,
+      "eval_loss": 0.7295576930046082,
+      "eval_mean_token_accuracy": 0.7990576628218876,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 89.9579,
+      "eval_samples_per_second": 15.785,
+      "eval_steps_per_second": 1.979,
+      "step": 380
+    },
+    {
+      "entropy": 0.8042497783899307,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 1.2437798976898193,
+      "learning_rate": 0.0003506606759952647,
+      "loss": 0.7381744384765625,
+      "mean_token_accuracy": 0.7967362694442273,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7717011336530193,
+      "eval_loss": 0.7436173558235168,
+      "eval_mean_token_accuracy": 0.7975849482450592,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 89.8411,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 400
+    },
+    {
+      "entropy": 0.7987750010056929,
+      "epoch": 1.012070006035003,
+      "grad_norm": 1.2390918731689453,
+      "learning_rate": 0.000364721224843344,
+      "loss": 0.7317790031433106,
+      "mean_token_accuracy": 0.7986087551364651,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7333170604170038,
+      "eval_loss": 0.7456948757171631,
+      "eval_mean_token_accuracy": 0.7964897577682238,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 89.7288,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 420
+    },
+    {
+      "entropy": 0.7372976846992969,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.1552740335464478,
+      "learning_rate": 0.00036468510102269575,
+      "loss": 0.6734249114990234,
+      "mean_token_accuracy": 0.805834824591875,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.7130355526891987,
+      "eval_loss": 0.7555333375930786,
+      "eval_mean_token_accuracy": 0.7961652781186479,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 89.8713,
+      "eval_samples_per_second": 15.8,
+      "eval_steps_per_second": 1.981,
+      "step": 440
+    },
+    {
+      "entropy": 0.7465668715536594,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 1.1991469860076904,
+      "learning_rate": 0.0003645973816745066,
+      "loss": 0.6871879577636719,
+      "mean_token_accuracy": 0.8042450234293937,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7772979542110743,
+      "eval_loss": 0.7561826705932617,
+      "eval_mean_token_accuracy": 0.7953876427720102,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 89.8395,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 460
+    },
+    {
+      "entropy": 0.7477855801582336,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 1.325486660003662,
+      "learning_rate": 0.00036445809162231435,
+      "loss": 0.6836830139160156,
+      "mean_token_accuracy": 0.8021619468927383,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.7098110676481483,
+      "eval_loss": 0.758007287979126,
+      "eval_mean_token_accuracy": 0.7909953189030122,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 89.6101,
+      "eval_samples_per_second": 15.846,
+      "eval_steps_per_second": 1.986,
+      "step": 480
+    },
+    {
+      "entropy": 0.7632691070437432,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 1.0122262239456177,
+      "learning_rate": 0.0003642672702835562,
+      "loss": 0.7071213245391845,
+      "mean_token_accuracy": 0.8000680930912495,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7645382720432924,
+      "eval_loss": 0.7492606043815613,
+      "eval_mean_token_accuracy": 0.7976858676149604,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 89.6092,
+      "eval_samples_per_second": 15.847,
+      "eval_steps_per_second": 1.986,
+      "step": 500
+    },
+    {
+      "entropy": 0.7623726457357407,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.367774486541748,
+      "learning_rate": 0.00036402497165841384,
+      "loss": 0.6965175151824952,
+      "mean_token_accuracy": 0.8016501650214195,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7868083715438843,
+      "eval_loss": 0.7481449842453003,
+      "eval_mean_token_accuracy": 0.799330582779445,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 89.8241,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 520
+    },
+    {
+      "entropy": 0.7535207405686378,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 1.213614821434021,
+      "learning_rate": 0.00036373126431453207,
+      "loss": 0.7078310489654541,
+      "mean_token_accuracy": 0.7998907208442688,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7829091950748743,
+      "eval_loss": 0.7416072487831116,
+      "eval_mean_token_accuracy": 0.7969146250339036,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 89.4607,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 540
+    },
+    {
+      "entropy": 0.7527363449335098,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 1.0929982662200928,
+      "learning_rate": 0.00036338623136761495,
+      "loss": 0.7058369636535644,
+      "mean_token_accuracy": 0.7997087433934211,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.728419938114252,
+      "eval_loss": 0.744873046875,
+      "eval_mean_token_accuracy": 0.7966634296299366,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 89.6597,
+      "eval_samples_per_second": 15.838,
+      "eval_steps_per_second": 1.985,
+      "step": 560
+    },
+    {
+      "entropy": 0.7406099259853363,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.409364104270935,
+      "learning_rate": 0.00036298997045790515,
+      "loss": 0.6902852058410645,
+      "mean_token_accuracy": 0.805386807769537,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.6991632758231645,
+      "eval_loss": 0.7480319738388062,
+      "eval_mean_token_accuracy": 0.7970068006033308,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 89.8329,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 580
+    },
+    {
+      "entropy": 0.7459449704736472,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.7053111791610718,
+      "learning_rate": 0.00036254259372255277,
+      "loss": 0.7044640064239502,
+      "mean_token_accuracy": 0.8005018755793571,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7454566660891758,
+      "eval_loss": 0.7551484704017639,
+      "eval_mean_token_accuracy": 0.7897093731365846,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 89.8313,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 600
+    },
+    {
+      "entropy": 0.7621391348540782,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 1.174443244934082,
+      "learning_rate": 0.000362044227763882,
+      "loss": 0.7111573696136475,
+      "mean_token_accuracy": 0.798752411454916,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7664387949397055,
+      "eval_loss": 0.7365804314613342,
+      "eval_mean_token_accuracy": 0.7963101200843126,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 89.9623,
+      "eval_samples_per_second": 15.784,
+      "eval_steps_per_second": 1.979,
+      "step": 620
+    },
+    {
+      "entropy": 0.7424944408237935,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.2898184061050415,
+      "learning_rate": 0.000361495013613564,
+      "loss": 0.6913110733032226,
+      "mean_token_accuracy": 0.8009015507996082,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7255479529332579,
+      "eval_loss": 0.7345991730690002,
+      "eval_mean_token_accuracy": 0.8013592945056015,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 89.817,
+      "eval_samples_per_second": 15.81,
+      "eval_steps_per_second": 1.982,
+      "step": 640
+    },
+    {
+      "entropy": 0.7303367637097835,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.1310476064682007,
+      "learning_rate": 0.00036089510669270666,
+      "loss": 0.6875617027282714,
+      "mean_token_accuracy": 0.8035397171974182,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7101308248016271,
+      "eval_loss": 0.7394917011260986,
+      "eval_mean_token_accuracy": 0.7990792942850777,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 89.7431,
+      "eval_samples_per_second": 15.823,
+      "eval_steps_per_second": 1.983,
+      "step": 660
+    },
+    {
+      "entropy": 0.7387678384780884,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.9645389914512634,
+      "learning_rate": 0.0003602446767678725,
+      "loss": 0.6911204338073731,
+      "mean_token_accuracy": 0.8056333385407924,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7281073737010527,
+      "eval_loss": 0.7376708984375,
+      "eval_mean_token_accuracy": 0.7997356415464637,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 90.0338,
+      "eval_samples_per_second": 15.772,
+      "eval_steps_per_second": 1.977,
+      "step": 680
+    },
+    {
+      "entropy": 0.7449554048478604,
+      "epoch": 1.687990343995172,
+      "grad_norm": 1.9069629907608032,
+      "learning_rate": 0.0003595439079030364,
+      "loss": 0.7042286396026611,
+      "mean_token_accuracy": 0.8018639378249646,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7022633817088738,
+      "eval_loss": 0.7320939898490906,
+      "eval_mean_token_accuracy": 0.8006051541044471,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 89.8948,
+      "eval_samples_per_second": 15.796,
+      "eval_steps_per_second": 1.98,
+      "step": 700
+    },
+    {
+      "entropy": 0.723249051719904,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.8333877325057983,
+      "learning_rate": 0.00035879299840749777,
+      "loss": 0.7043714046478271,
+      "mean_token_accuracy": 0.8028345607221127,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.672436946898364,
+      "eval_loss": 0.713233470916748,
+      "eval_mean_token_accuracy": 0.8047760577684038,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 89.6394,
+      "eval_samples_per_second": 15.841,
+      "eval_steps_per_second": 1.986,
+      "step": 720
+    },
+    {
+      "entropy": 0.730014206469059,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.4580037593841553,
+      "learning_rate": 0.00035799216077976137,
+      "loss": 0.6985176563262939,
+      "mean_token_accuracy": 0.8015547141432762,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.6908873795123582,
+      "eval_loss": 0.7198202013969421,
+      "eval_mean_token_accuracy": 0.8024131682481659,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 89.3771,
+      "eval_samples_per_second": 15.888,
+      "eval_steps_per_second": 1.992,
+      "step": 740
+    },
+    {
+      "entropy": 0.7219605796039105,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.51682710647583,
+      "learning_rate": 0.000357141621647403,
+      "loss": 0.6844424724578857,
+      "mean_token_accuracy": 0.8032813094556331,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7113401109582922,
+      "eval_loss": 0.7195309400558472,
+      "eval_mean_token_accuracy": 0.8019187035185568,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 89.4658,
+      "eval_samples_per_second": 15.872,
+      "eval_steps_per_second": 1.99,
+      "step": 760
+    },
+    {
+      "entropy": 0.729826743900776,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 1.0838000774383545,
+      "learning_rate": 0.0003562416217029361,
+      "loss": 0.685858678817749,
+      "mean_token_accuracy": 0.8050567395985126,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.649370267484965,
+      "eval_loss": 0.7177029252052307,
+      "eval_mean_token_accuracy": 0.8050174043419656,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 89.2287,
+      "eval_samples_per_second": 15.914,
+      "eval_steps_per_second": 1.995,
+      "step": 780
+    },
+    {
+      "entropy": 0.6799364153295755,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 1.1364924907684326,
+      "learning_rate": 0.00035529241563569903,
+      "loss": 0.6714428901672364,
+      "mean_token_accuracy": 0.8087341658771038,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6603799047095052,
+      "eval_loss": 0.7201809287071228,
+      "eval_mean_token_accuracy": 0.802863868769635,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 89.2718,
+      "eval_samples_per_second": 15.906,
+      "eval_steps_per_second": 1.994,
+      "step": 800
+    },
+    {
+      "entropy": 0.7078610375523567,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 1.2191942930221558,
+      "learning_rate": 0.0003542942720597808,
+      "loss": 0.6996590614318847,
+      "mean_token_accuracy": 0.8049512051045895,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.7240545264120852,
+      "eval_loss": 0.7301309108734131,
+      "eval_mean_token_accuracy": 0.8015973166133581,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 89.3438,
+      "eval_samples_per_second": 15.894,
+      "eval_steps_per_second": 1.992,
+      "step": 820
+    },
+    {
+      "entropy": 0.6557391556826505,
+      "epoch": 2.024140012070006,
+      "grad_norm": 1.36203134059906,
+      "learning_rate": 0.0003532474734380064,
+      "loss": 0.6327113628387451,
+      "mean_token_accuracy": 0.8142829847026181,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6573808648613062,
+      "eval_loss": 0.7334365844726562,
+      "eval_mean_token_accuracy": 0.8043633893634496,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 89.3359,
+      "eval_samples_per_second": 15.895,
+      "eval_steps_per_second": 1.992,
+      "step": 840
+    },
+    {
+      "entropy": 0.6409128502011299,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 1.5804955959320068,
+      "learning_rate": 0.0003521523160020035,
+      "loss": 0.6023559093475341,
+      "mean_token_accuracy": 0.8222251623868942,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6983234788594621,
+      "eval_loss": 0.7232135534286499,
+      "eval_mean_token_accuracy": 0.8010654268639811,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 89.2615,
+      "eval_samples_per_second": 15.908,
+      "eval_steps_per_second": 1.994,
+      "step": 860
+    },
+    {
+      "entropy": 0.6151149723678827,
+      "epoch": 2.12070006035003,
+      "grad_norm": 1.1975436210632324,
+      "learning_rate": 0.00035100910966837193,
+      "loss": 0.5803564548492431,
+      "mean_token_accuracy": 0.8256325736641884,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6636555573243773,
+      "eval_loss": 0.7205662727355957,
+      "eval_mean_token_accuracy": 0.8036768369460374,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 89.2957,
+      "eval_samples_per_second": 15.902,
+      "eval_steps_per_second": 1.993,
+      "step": 880
+    },
+    {
+      "entropy": 0.6201727617532015,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9907131195068359,
+      "learning_rate": 0.0003498181779509813,
+      "loss": 0.6101872444152832,
+      "mean_token_accuracy": 0.8196512959897518,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.6473788086617931,
+      "eval_loss": 0.7062397003173828,
+      "eval_mean_token_accuracy": 0.8078362184963869,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 88.6198,
+      "eval_samples_per_second": 16.024,
+      "eval_steps_per_second": 2.009,
+      "step": 900
+    },
+    {
+      "entropy": 0.6097332935780286,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.9055633544921875,
+      "learning_rate": 0.00034857985786942026,
+      "loss": 0.5956094264984131,
+      "mean_token_accuracy": 0.8240575887262821,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.612629868341296,
+      "eval_loss": 0.7106770277023315,
+      "eval_mean_token_accuracy": 0.805324277181304,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 89.4383,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 920
+    },
+    {
+      "entropy": 0.6182701248675585,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.9513736367225647,
+      "learning_rate": 0.00034729449985362404,
+      "loss": 0.6019269466400147,
+      "mean_token_accuracy": 0.8187106013298034,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6437647456533453,
+      "eval_loss": 0.7173502445220947,
+      "eval_mean_token_accuracy": 0.8056439573175451,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.9568,
+      "eval_samples_per_second": 15.963,
+      "eval_steps_per_second": 2.001,
+      "step": 940
+    },
+    {
+      "entropy": 0.6508570514619351,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.9606207013130188,
+      "learning_rate": 0.0003459624676447067,
+      "loss": 0.6296409606933594,
+      "mean_token_accuracy": 0.8167732007801533,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6288247135248077,
+      "eval_loss": 0.7160675525665283,
+      "eval_mean_token_accuracy": 0.8066010689467527,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 89.0044,
+      "eval_samples_per_second": 15.954,
+      "eval_steps_per_second": 2.0,
+      "step": 960
+    },
+    {
+      "entropy": 0.6380081418901682,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.9677181839942932,
+      "learning_rate": 0.000344584138192027,
+      "loss": 0.6210546493530273,
+      "mean_token_accuracy": 0.8186135642230511,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6407805031604981,
+      "eval_loss": 0.7055043578147888,
+      "eval_mean_token_accuracy": 0.8070103157772107,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 89.5606,
+      "eval_samples_per_second": 15.855,
+      "eval_steps_per_second": 1.987,
+      "step": 980
+    },
+    {
+      "entropy": 0.6310251638293266,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 2.1046407222747803,
+      "learning_rate": 0.000343159901546516,
+      "loss": 0.6165830135345459,
+      "mean_token_accuracy": 0.8207855716347694,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6461507470420237,
+      "eval_loss": 0.7292136549949646,
+      "eval_mean_token_accuracy": 0.8030418261383356,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 89.315,
+      "eval_samples_per_second": 15.899,
+      "eval_steps_per_second": 1.993,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6256617344915867,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 1.1768349409103394,
+      "learning_rate": 0.0003416901607502972,
+      "loss": 0.6142709255218506,
+      "mean_token_accuracy": 0.8220669947564602,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6150523993406403,
+      "eval_loss": 0.7019563317298889,
+      "eval_mean_token_accuracy": 0.8094323099998946,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 89.419,
+      "eval_samples_per_second": 15.88,
+      "eval_steps_per_second": 1.991,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6126735735684633,
+      "epoch": 2.506940253470127,
+      "grad_norm": 1.8065496683120728,
+      "learning_rate": 0.00034017533172263055,
+      "loss": 0.6056458473205566,
+      "mean_token_accuracy": 0.8203762136399746,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6544345456562685,
+      "eval_loss": 0.7086517214775085,
+      "eval_mean_token_accuracy": 0.806709805901131,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 88.981,
+      "eval_samples_per_second": 15.958,
+      "eval_steps_per_second": 2.0,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6311754353344441,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 1.0576369762420654,
+      "learning_rate": 0.00033861584314221236,
+      "loss": 0.6176392555236816,
+      "mean_token_accuracy": 0.8177410490810871,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6273799200406235,
+      "eval_loss": 0.708163321018219,
+      "eval_mean_token_accuracy": 0.8032017099053672,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 89.7668,
+      "eval_samples_per_second": 15.819,
+      "eval_steps_per_second": 1.983,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6200249589979648,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.997065007686615,
+      "learning_rate": 0.0003370121363258637,
+      "loss": 0.6203888893127442,
+      "mean_token_accuracy": 0.8196275025606156,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6457576811983344,
+      "eval_loss": 0.6972290277481079,
+      "eval_mean_token_accuracy": 0.8095980849158898,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 89.2027,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 1080
+    },
+    {
+      "entropy": 0.626810473203659,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.865475058555603,
+      "learning_rate": 0.0003353646651036438,
+      "loss": 0.6102585792541504,
+      "mean_token_accuracy": 0.819708751142025,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.6412022431914726,
+      "eval_loss": 0.7036139965057373,
+      "eval_mean_token_accuracy": 0.8070317135098275,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 89.4733,
+      "eval_samples_per_second": 15.871,
+      "eval_steps_per_second": 1.989,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6581619590520859,
+      "epoch": 2.700060350030175,
+      "grad_norm": 1.00307035446167,
+      "learning_rate": 0.00033367389569042064,
+      "loss": 0.6397720336914062,
+      "mean_token_accuracy": 0.81413309648633,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6288736402318719,
+      "eval_loss": 0.6966825723648071,
+      "eval_mean_token_accuracy": 0.8094039669867312,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 88.7341,
+      "eval_samples_per_second": 16.003,
+      "eval_steps_per_second": 2.006,
+      "step": 1120
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.123058232066478e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1140/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1140/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1140/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1140/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1140/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..7ce4b072fcbe5b636867e13937c56fedee3b68b4
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1140/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.02728394274790723,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1140/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1140/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1140/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1140/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1140/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1140/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1140/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1140/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..46038d46b4e3b8a28d223c56bb29038a5f1735e2
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1140/trainer_state.json
@@ -0,0 +1,1231 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 2.748340374170187,
+  "eval_steps": 20,
+  "global_step": 1140,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.024795526266098,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.945375919342041,
+      "learning_rate": 1.6698127428345936e-05,
+      "loss": 1.7825420379638672,
+      "mean_token_accuracy": 0.6315708436071873,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.29921497588747,
+      "eval_loss": 1.169131875038147,
+      "eval_mean_token_accuracy": 0.7185706342204233,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 91.7898,
+      "eval_samples_per_second": 15.47,
+      "eval_steps_per_second": 1.939,
+      "step": 20
+    },
+    {
+      "entropy": 1.0346344470977784,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.7916979789733887,
+      "learning_rate": 3.427510366871008e-05,
+      "loss": 0.9480395317077637,
+      "mean_token_accuracy": 0.7543429024517536,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 0.9673695430326997,
+      "eval_loss": 0.8588430285453796,
+      "eval_mean_token_accuracy": 0.7737233002534073,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 89.8219,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 40
+    },
+    {
+      "entropy": 0.8977566614747048,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7972187995910645,
+      "learning_rate": 5.185207990907422e-05,
+      "loss": 0.8029604911804199,
+      "mean_token_accuracy": 0.7818016350269318,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8436836982041263,
+      "eval_loss": 0.7893033623695374,
+      "eval_mean_token_accuracy": 0.7858914984076211,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 89.9349,
+      "eval_samples_per_second": 15.789,
+      "eval_steps_per_second": 1.979,
+      "step": 60
+    },
+    {
+      "entropy": 0.8650321319699288,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.2128729820251465,
+      "learning_rate": 6.942905614943836e-05,
+      "loss": 0.77417893409729,
+      "mean_token_accuracy": 0.7875479467213153,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8209857180547179,
+      "eval_loss": 0.7572280168533325,
+      "eval_mean_token_accuracy": 0.7925970383574453,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 89.9459,
+      "eval_samples_per_second": 15.787,
+      "eval_steps_per_second": 1.979,
+      "step": 80
+    },
+    {
+      "entropy": 0.8096660938113928,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.1254287958145142,
+      "learning_rate": 8.700603238980251e-05,
+      "loss": 0.7170331001281738,
+      "mean_token_accuracy": 0.7994555257260799,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.792780208788561,
+      "eval_loss": 0.7442984580993652,
+      "eval_mean_token_accuracy": 0.7954704573984896,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 90.294,
+      "eval_samples_per_second": 15.726,
+      "eval_steps_per_second": 1.971,
+      "step": 100
+    },
+    {
+      "entropy": 0.7979829199612141,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2140507698059082,
+      "learning_rate": 0.00010458300863016666,
+      "loss": 0.7141861915588379,
+      "mean_token_accuracy": 0.8000014744699001,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.7938692486018277,
+      "eval_loss": 0.7279797196388245,
+      "eval_mean_token_accuracy": 0.7989798734027348,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 89.7189,
+      "eval_samples_per_second": 15.827,
+      "eval_steps_per_second": 1.984,
+      "step": 120
+    },
+    {
+      "entropy": 0.7916347607970238,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.0506458282470703,
+      "learning_rate": 0.0001221599848705308,
+      "loss": 0.7117055416107178,
+      "mean_token_accuracy": 0.7970046654343605,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.8052781063519167,
+      "eval_loss": 0.722576916217804,
+      "eval_mean_token_accuracy": 0.7982530523551984,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 90.0376,
+      "eval_samples_per_second": 15.771,
+      "eval_steps_per_second": 1.977,
+      "step": 140
+    },
+    {
+      "entropy": 0.7880666613578796,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8409207463264465,
+      "learning_rate": 0.00013973696111089492,
+      "loss": 0.7052523136138916,
+      "mean_token_accuracy": 0.8000123649835587,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7787443134891853,
+      "eval_loss": 0.7137772440910339,
+      "eval_mean_token_accuracy": 0.8014209287220173,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 90.2562,
+      "eval_samples_per_second": 15.733,
+      "eval_steps_per_second": 1.972,
+      "step": 160
+    },
+    {
+      "entropy": 0.7696007996797561,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.0182477235794067,
+      "learning_rate": 0.00015731393735125907,
+      "loss": 0.7008797645568847,
+      "mean_token_accuracy": 0.8007259473204613,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.752259682068664,
+      "eval_loss": 0.7171286344528198,
+      "eval_mean_token_accuracy": 0.80052122574174,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 90.0797,
+      "eval_samples_per_second": 15.764,
+      "eval_steps_per_second": 1.976,
+      "step": 180
+    },
+    {
+      "entropy": 0.7828126326203346,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9553632736206055,
+      "learning_rate": 0.0001748909135916232,
+      "loss": 0.7046853542327881,
+      "mean_token_accuracy": 0.7988562889397144,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7389638169427936,
+      "eval_loss": 0.7187935709953308,
+      "eval_mean_token_accuracy": 0.7995899999409579,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 89.9223,
+      "eval_samples_per_second": 15.791,
+      "eval_steps_per_second": 1.979,
+      "step": 200
+    },
+    {
+      "entropy": 0.7909286297857762,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 1.134459376335144,
+      "learning_rate": 0.00019246788983198735,
+      "loss": 0.6989312171936035,
+      "mean_token_accuracy": 0.7991565138101577,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7496965199374082,
+      "eval_loss": 0.7106321454048157,
+      "eval_mean_token_accuracy": 0.8046183442131857,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 89.6899,
+      "eval_samples_per_second": 15.832,
+      "eval_steps_per_second": 1.985,
+      "step": 220
+    },
+    {
+      "entropy": 0.7651963323354721,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.9557585716247559,
+      "learning_rate": 0.0002100448660723515,
+      "loss": 0.6984559059143066,
+      "mean_token_accuracy": 0.8019823372364044,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7615563079212488,
+      "eval_loss": 0.7098539471626282,
+      "eval_mean_token_accuracy": 0.8047233508544022,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 90.3752,
+      "eval_samples_per_second": 15.712,
+      "eval_steps_per_second": 1.97,
+      "step": 240
+    },
+    {
+      "entropy": 0.787507726252079,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.9279311299324036,
+      "learning_rate": 0.00022762184231271568,
+      "loss": 0.708138370513916,
+      "mean_token_accuracy": 0.8013308703899383,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7856640896100676,
+      "eval_loss": 0.7191570401191711,
+      "eval_mean_token_accuracy": 0.8024677249153008,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 93.7336,
+      "eval_samples_per_second": 15.149,
+      "eval_steps_per_second": 1.899,
+      "step": 260
+    },
+    {
+      "entropy": 0.768489234894514,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.7992776036262512,
+      "learning_rate": 0.0002451988185530798,
+      "loss": 0.7035849571228028,
+      "mean_token_accuracy": 0.8031105332076549,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7682498740346244,
+      "eval_loss": 0.7149022221565247,
+      "eval_mean_token_accuracy": 0.7973497995499814,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 89.7309,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 280
+    },
+    {
+      "entropy": 0.7720955617725849,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.1931514739990234,
+      "learning_rate": 0.0002627757947934439,
+      "loss": 0.6965402603149414,
+      "mean_token_accuracy": 0.8040348663926125,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7214708304807042,
+      "eval_loss": 0.7350823879241943,
+      "eval_mean_token_accuracy": 0.7942911925610532,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 89.6451,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 300
+    },
+    {
+      "entropy": 0.7930095426738262,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8994467258453369,
+      "learning_rate": 0.0002803527710338081,
+      "loss": 0.7331492900848389,
+      "mean_token_accuracy": 0.796185302734375,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.8142535488926963,
+      "eval_loss": 0.7237834334373474,
+      "eval_mean_token_accuracy": 0.799495273761535,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 90.6759,
+      "eval_samples_per_second": 15.66,
+      "eval_steps_per_second": 1.963,
+      "step": 320
+    },
+    {
+      "entropy": 0.7825102441012859,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 1.8224239349365234,
+      "learning_rate": 0.00029792974727417223,
+      "loss": 0.7331175804138184,
+      "mean_token_accuracy": 0.7977667152881622,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7327135443017724,
+      "eval_loss": 0.7388784289360046,
+      "eval_mean_token_accuracy": 0.8007041252730938,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 89.7692,
+      "eval_samples_per_second": 15.818,
+      "eval_steps_per_second": 1.983,
+      "step": 340
+    },
+    {
+      "entropy": 0.819331557303667,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 1.0407779216766357,
+      "learning_rate": 0.0003155067235145364,
+      "loss": 0.7456695556640625,
+      "mean_token_accuracy": 0.7936980701982975,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.8065585420372781,
+      "eval_loss": 0.735519528388977,
+      "eval_mean_token_accuracy": 0.7974206296245704,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 89.4336,
+      "eval_samples_per_second": 15.878,
+      "eval_steps_per_second": 1.99,
+      "step": 360
+    },
+    {
+      "entropy": 0.8114653021097183,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.1209490299224854,
+      "learning_rate": 0.0003330836997549005,
+      "loss": 0.7340899467468261,
+      "mean_token_accuracy": 0.7914272703230381,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.8025527285056168,
+      "eval_loss": 0.7295576930046082,
+      "eval_mean_token_accuracy": 0.7990576628218876,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 89.9579,
+      "eval_samples_per_second": 15.785,
+      "eval_steps_per_second": 1.979,
+      "step": 380
+    },
+    {
+      "entropy": 0.8042497783899307,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 1.2437798976898193,
+      "learning_rate": 0.0003506606759952647,
+      "loss": 0.7381744384765625,
+      "mean_token_accuracy": 0.7967362694442273,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7717011336530193,
+      "eval_loss": 0.7436173558235168,
+      "eval_mean_token_accuracy": 0.7975849482450592,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 89.8411,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 400
+    },
+    {
+      "entropy": 0.7987750010056929,
+      "epoch": 1.012070006035003,
+      "grad_norm": 1.2390918731689453,
+      "learning_rate": 0.000364721224843344,
+      "loss": 0.7317790031433106,
+      "mean_token_accuracy": 0.7986087551364651,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7333170604170038,
+      "eval_loss": 0.7456948757171631,
+      "eval_mean_token_accuracy": 0.7964897577682238,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 89.7288,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 420
+    },
+    {
+      "entropy": 0.7372976846992969,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.1552740335464478,
+      "learning_rate": 0.00036468510102269575,
+      "loss": 0.6734249114990234,
+      "mean_token_accuracy": 0.805834824591875,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.7130355526891987,
+      "eval_loss": 0.7555333375930786,
+      "eval_mean_token_accuracy": 0.7961652781186479,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 89.8713,
+      "eval_samples_per_second": 15.8,
+      "eval_steps_per_second": 1.981,
+      "step": 440
+    },
+    {
+      "entropy": 0.7465668715536594,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 1.1991469860076904,
+      "learning_rate": 0.0003645973816745066,
+      "loss": 0.6871879577636719,
+      "mean_token_accuracy": 0.8042450234293937,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7772979542110743,
+      "eval_loss": 0.7561826705932617,
+      "eval_mean_token_accuracy": 0.7953876427720102,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 89.8395,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 460
+    },
+    {
+      "entropy": 0.7477855801582336,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 1.325486660003662,
+      "learning_rate": 0.00036445809162231435,
+      "loss": 0.6836830139160156,
+      "mean_token_accuracy": 0.8021619468927383,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.7098110676481483,
+      "eval_loss": 0.758007287979126,
+      "eval_mean_token_accuracy": 0.7909953189030122,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 89.6101,
+      "eval_samples_per_second": 15.846,
+      "eval_steps_per_second": 1.986,
+      "step": 480
+    },
+    {
+      "entropy": 0.7632691070437432,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 1.0122262239456177,
+      "learning_rate": 0.0003642672702835562,
+      "loss": 0.7071213245391845,
+      "mean_token_accuracy": 0.8000680930912495,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7645382720432924,
+      "eval_loss": 0.7492606043815613,
+      "eval_mean_token_accuracy": 0.7976858676149604,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 89.6092,
+      "eval_samples_per_second": 15.847,
+      "eval_steps_per_second": 1.986,
+      "step": 500
+    },
+    {
+      "entropy": 0.7623726457357407,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.367774486541748,
+      "learning_rate": 0.00036402497165841384,
+      "loss": 0.6965175151824952,
+      "mean_token_accuracy": 0.8016501650214195,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7868083715438843,
+      "eval_loss": 0.7481449842453003,
+      "eval_mean_token_accuracy": 0.799330582779445,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 89.8241,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 520
+    },
+    {
+      "entropy": 0.7535207405686378,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 1.213614821434021,
+      "learning_rate": 0.00036373126431453207,
+      "loss": 0.7078310489654541,
+      "mean_token_accuracy": 0.7998907208442688,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7829091950748743,
+      "eval_loss": 0.7416072487831116,
+      "eval_mean_token_accuracy": 0.7969146250339036,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 89.4607,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 540
+    },
+    {
+      "entropy": 0.7527363449335098,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 1.0929982662200928,
+      "learning_rate": 0.00036338623136761495,
+      "loss": 0.7058369636535644,
+      "mean_token_accuracy": 0.7997087433934211,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.728419938114252,
+      "eval_loss": 0.744873046875,
+      "eval_mean_token_accuracy": 0.7966634296299366,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 89.6597,
+      "eval_samples_per_second": 15.838,
+      "eval_steps_per_second": 1.985,
+      "step": 560
+    },
+    {
+      "entropy": 0.7406099259853363,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.409364104270935,
+      "learning_rate": 0.00036298997045790515,
+      "loss": 0.6902852058410645,
+      "mean_token_accuracy": 0.805386807769537,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.6991632758231645,
+      "eval_loss": 0.7480319738388062,
+      "eval_mean_token_accuracy": 0.7970068006033308,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 89.8329,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 580
+    },
+    {
+      "entropy": 0.7459449704736472,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.7053111791610718,
+      "learning_rate": 0.00036254259372255277,
+      "loss": 0.7044640064239502,
+      "mean_token_accuracy": 0.8005018755793571,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7454566660891758,
+      "eval_loss": 0.7551484704017639,
+      "eval_mean_token_accuracy": 0.7897093731365846,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 89.8313,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 600
+    },
+    {
+      "entropy": 0.7621391348540782,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 1.174443244934082,
+      "learning_rate": 0.000362044227763882,
+      "loss": 0.7111573696136475,
+      "mean_token_accuracy": 0.798752411454916,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7664387949397055,
+      "eval_loss": 0.7365804314613342,
+      "eval_mean_token_accuracy": 0.7963101200843126,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 89.9623,
+      "eval_samples_per_second": 15.784,
+      "eval_steps_per_second": 1.979,
+      "step": 620
+    },
+    {
+      "entropy": 0.7424944408237935,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.2898184061050415,
+      "learning_rate": 0.000361495013613564,
+      "loss": 0.6913110733032226,
+      "mean_token_accuracy": 0.8009015507996082,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7255479529332579,
+      "eval_loss": 0.7345991730690002,
+      "eval_mean_token_accuracy": 0.8013592945056015,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 89.817,
+      "eval_samples_per_second": 15.81,
+      "eval_steps_per_second": 1.982,
+      "step": 640
+    },
+    {
+      "entropy": 0.7303367637097835,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.1310476064682007,
+      "learning_rate": 0.00036089510669270666,
+      "loss": 0.6875617027282714,
+      "mean_token_accuracy": 0.8035397171974182,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7101308248016271,
+      "eval_loss": 0.7394917011260986,
+      "eval_mean_token_accuracy": 0.7990792942850777,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 89.7431,
+      "eval_samples_per_second": 15.823,
+      "eval_steps_per_second": 1.983,
+      "step": 660
+    },
+    {
+      "entropy": 0.7387678384780884,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.9645389914512634,
+      "learning_rate": 0.0003602446767678725,
+      "loss": 0.6911204338073731,
+      "mean_token_accuracy": 0.8056333385407924,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7281073737010527,
+      "eval_loss": 0.7376708984375,
+      "eval_mean_token_accuracy": 0.7997356415464637,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 90.0338,
+      "eval_samples_per_second": 15.772,
+      "eval_steps_per_second": 1.977,
+      "step": 680
+    },
+    {
+      "entropy": 0.7449554048478604,
+      "epoch": 1.687990343995172,
+      "grad_norm": 1.9069629907608032,
+      "learning_rate": 0.0003595439079030364,
+      "loss": 0.7042286396026611,
+      "mean_token_accuracy": 0.8018639378249646,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7022633817088738,
+      "eval_loss": 0.7320939898490906,
+      "eval_mean_token_accuracy": 0.8006051541044471,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 89.8948,
+      "eval_samples_per_second": 15.796,
+      "eval_steps_per_second": 1.98,
+      "step": 700
+    },
+    {
+      "entropy": 0.723249051719904,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.8333877325057983,
+      "learning_rate": 0.00035879299840749777,
+      "loss": 0.7043714046478271,
+      "mean_token_accuracy": 0.8028345607221127,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.672436946898364,
+      "eval_loss": 0.713233470916748,
+      "eval_mean_token_accuracy": 0.8047760577684038,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 89.6394,
+      "eval_samples_per_second": 15.841,
+      "eval_steps_per_second": 1.986,
+      "step": 720
+    },
+    {
+      "entropy": 0.730014206469059,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.4580037593841553,
+      "learning_rate": 0.00035799216077976137,
+      "loss": 0.6985176563262939,
+      "mean_token_accuracy": 0.8015547141432762,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.6908873795123582,
+      "eval_loss": 0.7198202013969421,
+      "eval_mean_token_accuracy": 0.8024131682481659,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 89.3771,
+      "eval_samples_per_second": 15.888,
+      "eval_steps_per_second": 1.992,
+      "step": 740
+    },
+    {
+      "entropy": 0.7219605796039105,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.51682710647583,
+      "learning_rate": 0.000357141621647403,
+      "loss": 0.6844424724578857,
+      "mean_token_accuracy": 0.8032813094556331,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7113401109582922,
+      "eval_loss": 0.7195309400558472,
+      "eval_mean_token_accuracy": 0.8019187035185568,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 89.4658,
+      "eval_samples_per_second": 15.872,
+      "eval_steps_per_second": 1.99,
+      "step": 760
+    },
+    {
+      "entropy": 0.729826743900776,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 1.0838000774383545,
+      "learning_rate": 0.0003562416217029361,
+      "loss": 0.685858678817749,
+      "mean_token_accuracy": 0.8050567395985126,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.649370267484965,
+      "eval_loss": 0.7177029252052307,
+      "eval_mean_token_accuracy": 0.8050174043419656,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 89.2287,
+      "eval_samples_per_second": 15.914,
+      "eval_steps_per_second": 1.995,
+      "step": 780
+    },
+    {
+      "entropy": 0.6799364153295755,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 1.1364924907684326,
+      "learning_rate": 0.00035529241563569903,
+      "loss": 0.6714428901672364,
+      "mean_token_accuracy": 0.8087341658771038,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6603799047095052,
+      "eval_loss": 0.7201809287071228,
+      "eval_mean_token_accuracy": 0.802863868769635,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 89.2718,
+      "eval_samples_per_second": 15.906,
+      "eval_steps_per_second": 1.994,
+      "step": 800
+    },
+    {
+      "entropy": 0.7078610375523567,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 1.2191942930221558,
+      "learning_rate": 0.0003542942720597808,
+      "loss": 0.6996590614318847,
+      "mean_token_accuracy": 0.8049512051045895,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.7240545264120852,
+      "eval_loss": 0.7301309108734131,
+      "eval_mean_token_accuracy": 0.8015973166133581,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 89.3438,
+      "eval_samples_per_second": 15.894,
+      "eval_steps_per_second": 1.992,
+      "step": 820
+    },
+    {
+      "entropy": 0.6557391556826505,
+      "epoch": 2.024140012070006,
+      "grad_norm": 1.36203134059906,
+      "learning_rate": 0.0003532474734380064,
+      "loss": 0.6327113628387451,
+      "mean_token_accuracy": 0.8142829847026181,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6573808648613062,
+      "eval_loss": 0.7334365844726562,
+      "eval_mean_token_accuracy": 0.8043633893634496,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 89.3359,
+      "eval_samples_per_second": 15.895,
+      "eval_steps_per_second": 1.992,
+      "step": 840
+    },
+    {
+      "entropy": 0.6409128502011299,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 1.5804955959320068,
+      "learning_rate": 0.0003521523160020035,
+      "loss": 0.6023559093475341,
+      "mean_token_accuracy": 0.8222251623868942,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6983234788594621,
+      "eval_loss": 0.7232135534286499,
+      "eval_mean_token_accuracy": 0.8010654268639811,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 89.2615,
+      "eval_samples_per_second": 15.908,
+      "eval_steps_per_second": 1.994,
+      "step": 860
+    },
+    {
+      "entropy": 0.6151149723678827,
+      "epoch": 2.12070006035003,
+      "grad_norm": 1.1975436210632324,
+      "learning_rate": 0.00035100910966837193,
+      "loss": 0.5803564548492431,
+      "mean_token_accuracy": 0.8256325736641884,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6636555573243773,
+      "eval_loss": 0.7205662727355957,
+      "eval_mean_token_accuracy": 0.8036768369460374,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 89.2957,
+      "eval_samples_per_second": 15.902,
+      "eval_steps_per_second": 1.993,
+      "step": 880
+    },
+    {
+      "entropy": 0.6201727617532015,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9907131195068359,
+      "learning_rate": 0.0003498181779509813,
+      "loss": 0.6101872444152832,
+      "mean_token_accuracy": 0.8196512959897518,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.6473788086617931,
+      "eval_loss": 0.7062397003173828,
+      "eval_mean_token_accuracy": 0.8078362184963869,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 88.6198,
+      "eval_samples_per_second": 16.024,
+      "eval_steps_per_second": 2.009,
+      "step": 900
+    },
+    {
+      "entropy": 0.6097332935780286,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.9055633544921875,
+      "learning_rate": 0.00034857985786942026,
+      "loss": 0.5956094264984131,
+      "mean_token_accuracy": 0.8240575887262821,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.612629868341296,
+      "eval_loss": 0.7106770277023315,
+      "eval_mean_token_accuracy": 0.805324277181304,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 89.4383,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 920
+    },
+    {
+      "entropy": 0.6182701248675585,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.9513736367225647,
+      "learning_rate": 0.00034729449985362404,
+      "loss": 0.6019269466400147,
+      "mean_token_accuracy": 0.8187106013298034,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6437647456533453,
+      "eval_loss": 0.7173502445220947,
+      "eval_mean_token_accuracy": 0.8056439573175451,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.9568,
+      "eval_samples_per_second": 15.963,
+      "eval_steps_per_second": 2.001,
+      "step": 940
+    },
+    {
+      "entropy": 0.6508570514619351,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.9606207013130188,
+      "learning_rate": 0.0003459624676447067,
+      "loss": 0.6296409606933594,
+      "mean_token_accuracy": 0.8167732007801533,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6288247135248077,
+      "eval_loss": 0.7160675525665283,
+      "eval_mean_token_accuracy": 0.8066010689467527,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 89.0044,
+      "eval_samples_per_second": 15.954,
+      "eval_steps_per_second": 2.0,
+      "step": 960
+    },
+    {
+      "entropy": 0.6380081418901682,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.9677181839942932,
+      "learning_rate": 0.000344584138192027,
+      "loss": 0.6210546493530273,
+      "mean_token_accuracy": 0.8186135642230511,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6407805031604981,
+      "eval_loss": 0.7055043578147888,
+      "eval_mean_token_accuracy": 0.8070103157772107,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 89.5606,
+      "eval_samples_per_second": 15.855,
+      "eval_steps_per_second": 1.987,
+      "step": 980
+    },
+    {
+      "entropy": 0.6310251638293266,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 2.1046407222747803,
+      "learning_rate": 0.000343159901546516,
+      "loss": 0.6165830135345459,
+      "mean_token_accuracy": 0.8207855716347694,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6461507470420237,
+      "eval_loss": 0.7292136549949646,
+      "eval_mean_token_accuracy": 0.8030418261383356,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 89.315,
+      "eval_samples_per_second": 15.899,
+      "eval_steps_per_second": 1.993,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6256617344915867,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 1.1768349409103394,
+      "learning_rate": 0.0003416901607502972,
+      "loss": 0.6142709255218506,
+      "mean_token_accuracy": 0.8220669947564602,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6150523993406403,
+      "eval_loss": 0.7019563317298889,
+      "eval_mean_token_accuracy": 0.8094323099998946,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 89.419,
+      "eval_samples_per_second": 15.88,
+      "eval_steps_per_second": 1.991,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6126735735684633,
+      "epoch": 2.506940253470127,
+      "grad_norm": 1.8065496683120728,
+      "learning_rate": 0.00034017533172263055,
+      "loss": 0.6056458473205566,
+      "mean_token_accuracy": 0.8203762136399746,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6544345456562685,
+      "eval_loss": 0.7086517214775085,
+      "eval_mean_token_accuracy": 0.806709805901131,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 88.981,
+      "eval_samples_per_second": 15.958,
+      "eval_steps_per_second": 2.0,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6311754353344441,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 1.0576369762420654,
+      "learning_rate": 0.00033861584314221236,
+      "loss": 0.6176392555236816,
+      "mean_token_accuracy": 0.8177410490810871,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6273799200406235,
+      "eval_loss": 0.708163321018219,
+      "eval_mean_token_accuracy": 0.8032017099053672,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 89.7668,
+      "eval_samples_per_second": 15.819,
+      "eval_steps_per_second": 1.983,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6200249589979648,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.997065007686615,
+      "learning_rate": 0.0003370121363258637,
+      "loss": 0.6203888893127442,
+      "mean_token_accuracy": 0.8196275025606156,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6457576811983344,
+      "eval_loss": 0.6972290277481079,
+      "eval_mean_token_accuracy": 0.8095980849158898,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 89.2027,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 1080
+    },
+    {
+      "entropy": 0.626810473203659,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.865475058555603,
+      "learning_rate": 0.0003353646651036438,
+      "loss": 0.6102585792541504,
+      "mean_token_accuracy": 0.819708751142025,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.6412022431914726,
+      "eval_loss": 0.7036139965057373,
+      "eval_mean_token_accuracy": 0.8070317135098275,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 89.4733,
+      "eval_samples_per_second": 15.871,
+      "eval_steps_per_second": 1.989,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6581619590520859,
+      "epoch": 2.700060350030175,
+      "grad_norm": 1.00307035446167,
+      "learning_rate": 0.00033367389569042064,
+      "loss": 0.6397720336914062,
+      "mean_token_accuracy": 0.81413309648633,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6288736402318719,
+      "eval_loss": 0.6966825723648071,
+      "eval_mean_token_accuracy": 0.8094039669867312,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 88.7341,
+      "eval_samples_per_second": 16.003,
+      "eval_steps_per_second": 2.006,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6399412982165813,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.8589329719543457,
+      "learning_rate": 0.0003319403065539384,
+      "loss": 0.6347338199615479,
+      "mean_token_accuracy": 0.8139534957706929,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.683491913455256,
+      "eval_loss": 0.6984149217605591,
+      "eval_mean_token_accuracy": 0.8083989469522841,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 89.3929,
+      "eval_samples_per_second": 15.885,
+      "eval_steps_per_second": 1.991,
+      "step": 1140
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.141886791596626e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1160/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1160/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1160/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1160/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1160/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..7ce4b072fcbe5b636867e13937c56fedee3b68b4
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1160/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.02728394274790723,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1160/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1160/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1160/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1160/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1160/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1160/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1160/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1160/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..a8dce921e9db896b938d62e71a0cb7ccf6d7bab0
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1160/trainer_state.json
@@ -0,0 +1,1252 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 2.796620398310199,
+  "eval_steps": 20,
+  "global_step": 1160,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.024795526266098,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.945375919342041,
+      "learning_rate": 1.6698127428345936e-05,
+      "loss": 1.7825420379638672,
+      "mean_token_accuracy": 0.6315708436071873,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.29921497588747,
+      "eval_loss": 1.169131875038147,
+      "eval_mean_token_accuracy": 0.7185706342204233,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 91.7898,
+      "eval_samples_per_second": 15.47,
+      "eval_steps_per_second": 1.939,
+      "step": 20
+    },
+    {
+      "entropy": 1.0346344470977784,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.7916979789733887,
+      "learning_rate": 3.427510366871008e-05,
+      "loss": 0.9480395317077637,
+      "mean_token_accuracy": 0.7543429024517536,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 0.9673695430326997,
+      "eval_loss": 0.8588430285453796,
+      "eval_mean_token_accuracy": 0.7737233002534073,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 89.8219,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 40
+    },
+    {
+      "entropy": 0.8977566614747048,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7972187995910645,
+      "learning_rate": 5.185207990907422e-05,
+      "loss": 0.8029604911804199,
+      "mean_token_accuracy": 0.7818016350269318,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8436836982041263,
+      "eval_loss": 0.7893033623695374,
+      "eval_mean_token_accuracy": 0.7858914984076211,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 89.9349,
+      "eval_samples_per_second": 15.789,
+      "eval_steps_per_second": 1.979,
+      "step": 60
+    },
+    {
+      "entropy": 0.8650321319699288,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.2128729820251465,
+      "learning_rate": 6.942905614943836e-05,
+      "loss": 0.77417893409729,
+      "mean_token_accuracy": 0.7875479467213153,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8209857180547179,
+      "eval_loss": 0.7572280168533325,
+      "eval_mean_token_accuracy": 0.7925970383574453,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 89.9459,
+      "eval_samples_per_second": 15.787,
+      "eval_steps_per_second": 1.979,
+      "step": 80
+    },
+    {
+      "entropy": 0.8096660938113928,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.1254287958145142,
+      "learning_rate": 8.700603238980251e-05,
+      "loss": 0.7170331001281738,
+      "mean_token_accuracy": 0.7994555257260799,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.792780208788561,
+      "eval_loss": 0.7442984580993652,
+      "eval_mean_token_accuracy": 0.7954704573984896,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 90.294,
+      "eval_samples_per_second": 15.726,
+      "eval_steps_per_second": 1.971,
+      "step": 100
+    },
+    {
+      "entropy": 0.7979829199612141,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2140507698059082,
+      "learning_rate": 0.00010458300863016666,
+      "loss": 0.7141861915588379,
+      "mean_token_accuracy": 0.8000014744699001,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.7938692486018277,
+      "eval_loss": 0.7279797196388245,
+      "eval_mean_token_accuracy": 0.7989798734027348,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 89.7189,
+      "eval_samples_per_second": 15.827,
+      "eval_steps_per_second": 1.984,
+      "step": 120
+    },
+    {
+      "entropy": 0.7916347607970238,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.0506458282470703,
+      "learning_rate": 0.0001221599848705308,
+      "loss": 0.7117055416107178,
+      "mean_token_accuracy": 0.7970046654343605,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.8052781063519167,
+      "eval_loss": 0.722576916217804,
+      "eval_mean_token_accuracy": 0.7982530523551984,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 90.0376,
+      "eval_samples_per_second": 15.771,
+      "eval_steps_per_second": 1.977,
+      "step": 140
+    },
+    {
+      "entropy": 0.7880666613578796,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8409207463264465,
+      "learning_rate": 0.00013973696111089492,
+      "loss": 0.7052523136138916,
+      "mean_token_accuracy": 0.8000123649835587,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7787443134891853,
+      "eval_loss": 0.7137772440910339,
+      "eval_mean_token_accuracy": 0.8014209287220173,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 90.2562,
+      "eval_samples_per_second": 15.733,
+      "eval_steps_per_second": 1.972,
+      "step": 160
+    },
+    {
+      "entropy": 0.7696007996797561,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.0182477235794067,
+      "learning_rate": 0.00015731393735125907,
+      "loss": 0.7008797645568847,
+      "mean_token_accuracy": 0.8007259473204613,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.752259682068664,
+      "eval_loss": 0.7171286344528198,
+      "eval_mean_token_accuracy": 0.80052122574174,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 90.0797,
+      "eval_samples_per_second": 15.764,
+      "eval_steps_per_second": 1.976,
+      "step": 180
+    },
+    {
+      "entropy": 0.7828126326203346,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9553632736206055,
+      "learning_rate": 0.0001748909135916232,
+      "loss": 0.7046853542327881,
+      "mean_token_accuracy": 0.7988562889397144,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7389638169427936,
+      "eval_loss": 0.7187935709953308,
+      "eval_mean_token_accuracy": 0.7995899999409579,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 89.9223,
+      "eval_samples_per_second": 15.791,
+      "eval_steps_per_second": 1.979,
+      "step": 200
+    },
+    {
+      "entropy": 0.7909286297857762,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 1.134459376335144,
+      "learning_rate": 0.00019246788983198735,
+      "loss": 0.6989312171936035,
+      "mean_token_accuracy": 0.7991565138101577,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7496965199374082,
+      "eval_loss": 0.7106321454048157,
+      "eval_mean_token_accuracy": 0.8046183442131857,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 89.6899,
+      "eval_samples_per_second": 15.832,
+      "eval_steps_per_second": 1.985,
+      "step": 220
+    },
+    {
+      "entropy": 0.7651963323354721,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.9557585716247559,
+      "learning_rate": 0.0002100448660723515,
+      "loss": 0.6984559059143066,
+      "mean_token_accuracy": 0.8019823372364044,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7615563079212488,
+      "eval_loss": 0.7098539471626282,
+      "eval_mean_token_accuracy": 0.8047233508544022,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 90.3752,
+      "eval_samples_per_second": 15.712,
+      "eval_steps_per_second": 1.97,
+      "step": 240
+    },
+    {
+      "entropy": 0.787507726252079,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.9279311299324036,
+      "learning_rate": 0.00022762184231271568,
+      "loss": 0.708138370513916,
+      "mean_token_accuracy": 0.8013308703899383,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7856640896100676,
+      "eval_loss": 0.7191570401191711,
+      "eval_mean_token_accuracy": 0.8024677249153008,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 93.7336,
+      "eval_samples_per_second": 15.149,
+      "eval_steps_per_second": 1.899,
+      "step": 260
+    },
+    {
+      "entropy": 0.768489234894514,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.7992776036262512,
+      "learning_rate": 0.0002451988185530798,
+      "loss": 0.7035849571228028,
+      "mean_token_accuracy": 0.8031105332076549,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7682498740346244,
+      "eval_loss": 0.7149022221565247,
+      "eval_mean_token_accuracy": 0.7973497995499814,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 89.7309,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 280
+    },
+    {
+      "entropy": 0.7720955617725849,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.1931514739990234,
+      "learning_rate": 0.0002627757947934439,
+      "loss": 0.6965402603149414,
+      "mean_token_accuracy": 0.8040348663926125,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7214708304807042,
+      "eval_loss": 0.7350823879241943,
+      "eval_mean_token_accuracy": 0.7942911925610532,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 89.6451,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 300
+    },
+    {
+      "entropy": 0.7930095426738262,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8994467258453369,
+      "learning_rate": 0.0002803527710338081,
+      "loss": 0.7331492900848389,
+      "mean_token_accuracy": 0.796185302734375,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.8142535488926963,
+      "eval_loss": 0.7237834334373474,
+      "eval_mean_token_accuracy": 0.799495273761535,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 90.6759,
+      "eval_samples_per_second": 15.66,
+      "eval_steps_per_second": 1.963,
+      "step": 320
+    },
+    {
+      "entropy": 0.7825102441012859,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 1.8224239349365234,
+      "learning_rate": 0.00029792974727417223,
+      "loss": 0.7331175804138184,
+      "mean_token_accuracy": 0.7977667152881622,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7327135443017724,
+      "eval_loss": 0.7388784289360046,
+      "eval_mean_token_accuracy": 0.8007041252730938,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 89.7692,
+      "eval_samples_per_second": 15.818,
+      "eval_steps_per_second": 1.983,
+      "step": 340
+    },
+    {
+      "entropy": 0.819331557303667,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 1.0407779216766357,
+      "learning_rate": 0.0003155067235145364,
+      "loss": 0.7456695556640625,
+      "mean_token_accuracy": 0.7936980701982975,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.8065585420372781,
+      "eval_loss": 0.735519528388977,
+      "eval_mean_token_accuracy": 0.7974206296245704,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 89.4336,
+      "eval_samples_per_second": 15.878,
+      "eval_steps_per_second": 1.99,
+      "step": 360
+    },
+    {
+      "entropy": 0.8114653021097183,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.1209490299224854,
+      "learning_rate": 0.0003330836997549005,
+      "loss": 0.7340899467468261,
+      "mean_token_accuracy": 0.7914272703230381,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.8025527285056168,
+      "eval_loss": 0.7295576930046082,
+      "eval_mean_token_accuracy": 0.7990576628218876,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 89.9579,
+      "eval_samples_per_second": 15.785,
+      "eval_steps_per_second": 1.979,
+      "step": 380
+    },
+    {
+      "entropy": 0.8042497783899307,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 1.2437798976898193,
+      "learning_rate": 0.0003506606759952647,
+      "loss": 0.7381744384765625,
+      "mean_token_accuracy": 0.7967362694442273,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7717011336530193,
+      "eval_loss": 0.7436173558235168,
+      "eval_mean_token_accuracy": 0.7975849482450592,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 89.8411,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 400
+    },
+    {
+      "entropy": 0.7987750010056929,
+      "epoch": 1.012070006035003,
+      "grad_norm": 1.2390918731689453,
+      "learning_rate": 0.000364721224843344,
+      "loss": 0.7317790031433106,
+      "mean_token_accuracy": 0.7986087551364651,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7333170604170038,
+      "eval_loss": 0.7456948757171631,
+      "eval_mean_token_accuracy": 0.7964897577682238,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 89.7288,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 420
+    },
+    {
+      "entropy": 0.7372976846992969,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.1552740335464478,
+      "learning_rate": 0.00036468510102269575,
+      "loss": 0.6734249114990234,
+      "mean_token_accuracy": 0.805834824591875,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.7130355526891987,
+      "eval_loss": 0.7555333375930786,
+      "eval_mean_token_accuracy": 0.7961652781186479,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 89.8713,
+      "eval_samples_per_second": 15.8,
+      "eval_steps_per_second": 1.981,
+      "step": 440
+    },
+    {
+      "entropy": 0.7465668715536594,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 1.1991469860076904,
+      "learning_rate": 0.0003645973816745066,
+      "loss": 0.6871879577636719,
+      "mean_token_accuracy": 0.8042450234293937,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7772979542110743,
+      "eval_loss": 0.7561826705932617,
+      "eval_mean_token_accuracy": 0.7953876427720102,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 89.8395,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 460
+    },
+    {
+      "entropy": 0.7477855801582336,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 1.325486660003662,
+      "learning_rate": 0.00036445809162231435,
+      "loss": 0.6836830139160156,
+      "mean_token_accuracy": 0.8021619468927383,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.7098110676481483,
+      "eval_loss": 0.758007287979126,
+      "eval_mean_token_accuracy": 0.7909953189030122,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 89.6101,
+      "eval_samples_per_second": 15.846,
+      "eval_steps_per_second": 1.986,
+      "step": 480
+    },
+    {
+      "entropy": 0.7632691070437432,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 1.0122262239456177,
+      "learning_rate": 0.0003642672702835562,
+      "loss": 0.7071213245391845,
+      "mean_token_accuracy": 0.8000680930912495,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7645382720432924,
+      "eval_loss": 0.7492606043815613,
+      "eval_mean_token_accuracy": 0.7976858676149604,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 89.6092,
+      "eval_samples_per_second": 15.847,
+      "eval_steps_per_second": 1.986,
+      "step": 500
+    },
+    {
+      "entropy": 0.7623726457357407,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.367774486541748,
+      "learning_rate": 0.00036402497165841384,
+      "loss": 0.6965175151824952,
+      "mean_token_accuracy": 0.8016501650214195,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7868083715438843,
+      "eval_loss": 0.7481449842453003,
+      "eval_mean_token_accuracy": 0.799330582779445,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 89.8241,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 520
+    },
+    {
+      "entropy": 0.7535207405686378,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 1.213614821434021,
+      "learning_rate": 0.00036373126431453207,
+      "loss": 0.7078310489654541,
+      "mean_token_accuracy": 0.7998907208442688,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7829091950748743,
+      "eval_loss": 0.7416072487831116,
+      "eval_mean_token_accuracy": 0.7969146250339036,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 89.4607,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 540
+    },
+    {
+      "entropy": 0.7527363449335098,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 1.0929982662200928,
+      "learning_rate": 0.00036338623136761495,
+      "loss": 0.7058369636535644,
+      "mean_token_accuracy": 0.7997087433934211,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.728419938114252,
+      "eval_loss": 0.744873046875,
+      "eval_mean_token_accuracy": 0.7966634296299366,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 89.6597,
+      "eval_samples_per_second": 15.838,
+      "eval_steps_per_second": 1.985,
+      "step": 560
+    },
+    {
+      "entropy": 0.7406099259853363,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.409364104270935,
+      "learning_rate": 0.00036298997045790515,
+      "loss": 0.6902852058410645,
+      "mean_token_accuracy": 0.805386807769537,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.6991632758231645,
+      "eval_loss": 0.7480319738388062,
+      "eval_mean_token_accuracy": 0.7970068006033308,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 89.8329,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 580
+    },
+    {
+      "entropy": 0.7459449704736472,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.7053111791610718,
+      "learning_rate": 0.00036254259372255277,
+      "loss": 0.7044640064239502,
+      "mean_token_accuracy": 0.8005018755793571,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7454566660891758,
+      "eval_loss": 0.7551484704017639,
+      "eval_mean_token_accuracy": 0.7897093731365846,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 89.8313,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 600
+    },
+    {
+      "entropy": 0.7621391348540782,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 1.174443244934082,
+      "learning_rate": 0.000362044227763882,
+      "loss": 0.7111573696136475,
+      "mean_token_accuracy": 0.798752411454916,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7664387949397055,
+      "eval_loss": 0.7365804314613342,
+      "eval_mean_token_accuracy": 0.7963101200843126,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 89.9623,
+      "eval_samples_per_second": 15.784,
+      "eval_steps_per_second": 1.979,
+      "step": 620
+    },
+    {
+      "entropy": 0.7424944408237935,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.2898184061050415,
+      "learning_rate": 0.000361495013613564,
+      "loss": 0.6913110733032226,
+      "mean_token_accuracy": 0.8009015507996082,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7255479529332579,
+      "eval_loss": 0.7345991730690002,
+      "eval_mean_token_accuracy": 0.8013592945056015,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 89.817,
+      "eval_samples_per_second": 15.81,
+      "eval_steps_per_second": 1.982,
+      "step": 640
+    },
+    {
+      "entropy": 0.7303367637097835,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.1310476064682007,
+      "learning_rate": 0.00036089510669270666,
+      "loss": 0.6875617027282714,
+      "mean_token_accuracy": 0.8035397171974182,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7101308248016271,
+      "eval_loss": 0.7394917011260986,
+      "eval_mean_token_accuracy": 0.7990792942850777,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 89.7431,
+      "eval_samples_per_second": 15.823,
+      "eval_steps_per_second": 1.983,
+      "step": 660
+    },
+    {
+      "entropy": 0.7387678384780884,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.9645389914512634,
+      "learning_rate": 0.0003602446767678725,
+      "loss": 0.6911204338073731,
+      "mean_token_accuracy": 0.8056333385407924,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7281073737010527,
+      "eval_loss": 0.7376708984375,
+      "eval_mean_token_accuracy": 0.7997356415464637,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 90.0338,
+      "eval_samples_per_second": 15.772,
+      "eval_steps_per_second": 1.977,
+      "step": 680
+    },
+    {
+      "entropy": 0.7449554048478604,
+      "epoch": 1.687990343995172,
+      "grad_norm": 1.9069629907608032,
+      "learning_rate": 0.0003595439079030364,
+      "loss": 0.7042286396026611,
+      "mean_token_accuracy": 0.8018639378249646,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7022633817088738,
+      "eval_loss": 0.7320939898490906,
+      "eval_mean_token_accuracy": 0.8006051541044471,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 89.8948,
+      "eval_samples_per_second": 15.796,
+      "eval_steps_per_second": 1.98,
+      "step": 700
+    },
+    {
+      "entropy": 0.723249051719904,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.8333877325057983,
+      "learning_rate": 0.00035879299840749777,
+      "loss": 0.7043714046478271,
+      "mean_token_accuracy": 0.8028345607221127,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.672436946898364,
+      "eval_loss": 0.713233470916748,
+      "eval_mean_token_accuracy": 0.8047760577684038,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 89.6394,
+      "eval_samples_per_second": 15.841,
+      "eval_steps_per_second": 1.986,
+      "step": 720
+    },
+    {
+      "entropy": 0.730014206469059,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.4580037593841553,
+      "learning_rate": 0.00035799216077976137,
+      "loss": 0.6985176563262939,
+      "mean_token_accuracy": 0.8015547141432762,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.6908873795123582,
+      "eval_loss": 0.7198202013969421,
+      "eval_mean_token_accuracy": 0.8024131682481659,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 89.3771,
+      "eval_samples_per_second": 15.888,
+      "eval_steps_per_second": 1.992,
+      "step": 740
+    },
+    {
+      "entropy": 0.7219605796039105,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.51682710647583,
+      "learning_rate": 0.000357141621647403,
+      "loss": 0.6844424724578857,
+      "mean_token_accuracy": 0.8032813094556331,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7113401109582922,
+      "eval_loss": 0.7195309400558472,
+      "eval_mean_token_accuracy": 0.8019187035185568,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 89.4658,
+      "eval_samples_per_second": 15.872,
+      "eval_steps_per_second": 1.99,
+      "step": 760
+    },
+    {
+      "entropy": 0.729826743900776,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 1.0838000774383545,
+      "learning_rate": 0.0003562416217029361,
+      "loss": 0.685858678817749,
+      "mean_token_accuracy": 0.8050567395985126,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.649370267484965,
+      "eval_loss": 0.7177029252052307,
+      "eval_mean_token_accuracy": 0.8050174043419656,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 89.2287,
+      "eval_samples_per_second": 15.914,
+      "eval_steps_per_second": 1.995,
+      "step": 780
+    },
+    {
+      "entropy": 0.6799364153295755,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 1.1364924907684326,
+      "learning_rate": 0.00035529241563569903,
+      "loss": 0.6714428901672364,
+      "mean_token_accuracy": 0.8087341658771038,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6603799047095052,
+      "eval_loss": 0.7201809287071228,
+      "eval_mean_token_accuracy": 0.802863868769635,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 89.2718,
+      "eval_samples_per_second": 15.906,
+      "eval_steps_per_second": 1.994,
+      "step": 800
+    },
+    {
+      "entropy": 0.7078610375523567,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 1.2191942930221558,
+      "learning_rate": 0.0003542942720597808,
+      "loss": 0.6996590614318847,
+      "mean_token_accuracy": 0.8049512051045895,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.7240545264120852,
+      "eval_loss": 0.7301309108734131,
+      "eval_mean_token_accuracy": 0.8015973166133581,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 89.3438,
+      "eval_samples_per_second": 15.894,
+      "eval_steps_per_second": 1.992,
+      "step": 820
+    },
+    {
+      "entropy": 0.6557391556826505,
+      "epoch": 2.024140012070006,
+      "grad_norm": 1.36203134059906,
+      "learning_rate": 0.0003532474734380064,
+      "loss": 0.6327113628387451,
+      "mean_token_accuracy": 0.8142829847026181,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6573808648613062,
+      "eval_loss": 0.7334365844726562,
+      "eval_mean_token_accuracy": 0.8043633893634496,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 89.3359,
+      "eval_samples_per_second": 15.895,
+      "eval_steps_per_second": 1.992,
+      "step": 840
+    },
+    {
+      "entropy": 0.6409128502011299,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 1.5804955959320068,
+      "learning_rate": 0.0003521523160020035,
+      "loss": 0.6023559093475341,
+      "mean_token_accuracy": 0.8222251623868942,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6983234788594621,
+      "eval_loss": 0.7232135534286499,
+      "eval_mean_token_accuracy": 0.8010654268639811,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 89.2615,
+      "eval_samples_per_second": 15.908,
+      "eval_steps_per_second": 1.994,
+      "step": 860
+    },
+    {
+      "entropy": 0.6151149723678827,
+      "epoch": 2.12070006035003,
+      "grad_norm": 1.1975436210632324,
+      "learning_rate": 0.00035100910966837193,
+      "loss": 0.5803564548492431,
+      "mean_token_accuracy": 0.8256325736641884,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6636555573243773,
+      "eval_loss": 0.7205662727355957,
+      "eval_mean_token_accuracy": 0.8036768369460374,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 89.2957,
+      "eval_samples_per_second": 15.902,
+      "eval_steps_per_second": 1.993,
+      "step": 880
+    },
+    {
+      "entropy": 0.6201727617532015,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9907131195068359,
+      "learning_rate": 0.0003498181779509813,
+      "loss": 0.6101872444152832,
+      "mean_token_accuracy": 0.8196512959897518,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.6473788086617931,
+      "eval_loss": 0.7062397003173828,
+      "eval_mean_token_accuracy": 0.8078362184963869,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 88.6198,
+      "eval_samples_per_second": 16.024,
+      "eval_steps_per_second": 2.009,
+      "step": 900
+    },
+    {
+      "entropy": 0.6097332935780286,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.9055633544921875,
+      "learning_rate": 0.00034857985786942026,
+      "loss": 0.5956094264984131,
+      "mean_token_accuracy": 0.8240575887262821,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.612629868341296,
+      "eval_loss": 0.7106770277023315,
+      "eval_mean_token_accuracy": 0.805324277181304,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 89.4383,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 920
+    },
+    {
+      "entropy": 0.6182701248675585,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.9513736367225647,
+      "learning_rate": 0.00034729449985362404,
+      "loss": 0.6019269466400147,
+      "mean_token_accuracy": 0.8187106013298034,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6437647456533453,
+      "eval_loss": 0.7173502445220947,
+      "eval_mean_token_accuracy": 0.8056439573175451,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.9568,
+      "eval_samples_per_second": 15.963,
+      "eval_steps_per_second": 2.001,
+      "step": 940
+    },
+    {
+      "entropy": 0.6508570514619351,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.9606207013130188,
+      "learning_rate": 0.0003459624676447067,
+      "loss": 0.6296409606933594,
+      "mean_token_accuracy": 0.8167732007801533,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6288247135248077,
+      "eval_loss": 0.7160675525665283,
+      "eval_mean_token_accuracy": 0.8066010689467527,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 89.0044,
+      "eval_samples_per_second": 15.954,
+      "eval_steps_per_second": 2.0,
+      "step": 960
+    },
+    {
+      "entropy": 0.6380081418901682,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.9677181839942932,
+      "learning_rate": 0.000344584138192027,
+      "loss": 0.6210546493530273,
+      "mean_token_accuracy": 0.8186135642230511,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6407805031604981,
+      "eval_loss": 0.7055043578147888,
+      "eval_mean_token_accuracy": 0.8070103157772107,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 89.5606,
+      "eval_samples_per_second": 15.855,
+      "eval_steps_per_second": 1.987,
+      "step": 980
+    },
+    {
+      "entropy": 0.6310251638293266,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 2.1046407222747803,
+      "learning_rate": 0.000343159901546516,
+      "loss": 0.6165830135345459,
+      "mean_token_accuracy": 0.8207855716347694,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6461507470420237,
+      "eval_loss": 0.7292136549949646,
+      "eval_mean_token_accuracy": 0.8030418261383356,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 89.315,
+      "eval_samples_per_second": 15.899,
+      "eval_steps_per_second": 1.993,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6256617344915867,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 1.1768349409103394,
+      "learning_rate": 0.0003416901607502972,
+      "loss": 0.6142709255218506,
+      "mean_token_accuracy": 0.8220669947564602,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6150523993406403,
+      "eval_loss": 0.7019563317298889,
+      "eval_mean_token_accuracy": 0.8094323099998946,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 89.419,
+      "eval_samples_per_second": 15.88,
+      "eval_steps_per_second": 1.991,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6126735735684633,
+      "epoch": 2.506940253470127,
+      "grad_norm": 1.8065496683120728,
+      "learning_rate": 0.00034017533172263055,
+      "loss": 0.6056458473205566,
+      "mean_token_accuracy": 0.8203762136399746,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6544345456562685,
+      "eval_loss": 0.7086517214775085,
+      "eval_mean_token_accuracy": 0.806709805901131,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 88.981,
+      "eval_samples_per_second": 15.958,
+      "eval_steps_per_second": 2.0,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6311754353344441,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 1.0576369762420654,
+      "learning_rate": 0.00033861584314221236,
+      "loss": 0.6176392555236816,
+      "mean_token_accuracy": 0.8177410490810871,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6273799200406235,
+      "eval_loss": 0.708163321018219,
+      "eval_mean_token_accuracy": 0.8032017099053672,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 89.7668,
+      "eval_samples_per_second": 15.819,
+      "eval_steps_per_second": 1.983,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6200249589979648,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.997065007686615,
+      "learning_rate": 0.0003370121363258637,
+      "loss": 0.6203888893127442,
+      "mean_token_accuracy": 0.8196275025606156,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6457576811983344,
+      "eval_loss": 0.6972290277481079,
+      "eval_mean_token_accuracy": 0.8095980849158898,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 89.2027,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 1080
+    },
+    {
+      "entropy": 0.626810473203659,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.865475058555603,
+      "learning_rate": 0.0003353646651036438,
+      "loss": 0.6102585792541504,
+      "mean_token_accuracy": 0.819708751142025,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.6412022431914726,
+      "eval_loss": 0.7036139965057373,
+      "eval_mean_token_accuracy": 0.8070317135098275,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 89.4733,
+      "eval_samples_per_second": 15.871,
+      "eval_steps_per_second": 1.989,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6581619590520859,
+      "epoch": 2.700060350030175,
+      "grad_norm": 1.00307035446167,
+      "learning_rate": 0.00033367389569042064,
+      "loss": 0.6397720336914062,
+      "mean_token_accuracy": 0.81413309648633,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6288736402318719,
+      "eval_loss": 0.6966825723648071,
+      "eval_mean_token_accuracy": 0.8094039669867312,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 88.7341,
+      "eval_samples_per_second": 16.003,
+      "eval_steps_per_second": 2.006,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6399412982165813,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.8589329719543457,
+      "learning_rate": 0.0003319403065539384,
+      "loss": 0.6347338199615479,
+      "mean_token_accuracy": 0.8139534957706929,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.683491913455256,
+      "eval_loss": 0.6984149217605591,
+      "eval_mean_token_accuracy": 0.8083989469522841,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 89.3929,
+      "eval_samples_per_second": 15.885,
+      "eval_steps_per_second": 1.991,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6264835461974144,
+      "epoch": 2.796620398310199,
+      "grad_norm": 0.7807853817939758,
+      "learning_rate": 0.0003301643882794163,
+      "loss": 0.6162629127502441,
+      "mean_token_accuracy": 0.8216292977333068,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6578792159476977,
+      "eval_loss": 0.6921441555023193,
+      "eval_mean_token_accuracy": 0.8094299177775223,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 89.2356,
+      "eval_samples_per_second": 15.913,
+      "eval_steps_per_second": 1.995,
+      "step": 1160
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.1627016981849907e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1180/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1180/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1180/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1180/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1180/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..7ce4b072fcbe5b636867e13937c56fedee3b68b4
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1180/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.02728394274790723,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1180/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1180/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1180/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1180/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1180/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1180/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1180/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1180/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..09d11725f6f4868c08785f0f31f647d59e525eea
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1180/trainer_state.json
@@ -0,0 +1,1273 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 2.8449004224502112,
+  "eval_steps": 20,
+  "global_step": 1180,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.024795526266098,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.945375919342041,
+      "learning_rate": 1.6698127428345936e-05,
+      "loss": 1.7825420379638672,
+      "mean_token_accuracy": 0.6315708436071873,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.29921497588747,
+      "eval_loss": 1.169131875038147,
+      "eval_mean_token_accuracy": 0.7185706342204233,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 91.7898,
+      "eval_samples_per_second": 15.47,
+      "eval_steps_per_second": 1.939,
+      "step": 20
+    },
+    {
+      "entropy": 1.0346344470977784,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.7916979789733887,
+      "learning_rate": 3.427510366871008e-05,
+      "loss": 0.9480395317077637,
+      "mean_token_accuracy": 0.7543429024517536,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 0.9673695430326997,
+      "eval_loss": 0.8588430285453796,
+      "eval_mean_token_accuracy": 0.7737233002534073,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 89.8219,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 40
+    },
+    {
+      "entropy": 0.8977566614747048,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7972187995910645,
+      "learning_rate": 5.185207990907422e-05,
+      "loss": 0.8029604911804199,
+      "mean_token_accuracy": 0.7818016350269318,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8436836982041263,
+      "eval_loss": 0.7893033623695374,
+      "eval_mean_token_accuracy": 0.7858914984076211,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 89.9349,
+      "eval_samples_per_second": 15.789,
+      "eval_steps_per_second": 1.979,
+      "step": 60
+    },
+    {
+      "entropy": 0.8650321319699288,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.2128729820251465,
+      "learning_rate": 6.942905614943836e-05,
+      "loss": 0.77417893409729,
+      "mean_token_accuracy": 0.7875479467213153,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8209857180547179,
+      "eval_loss": 0.7572280168533325,
+      "eval_mean_token_accuracy": 0.7925970383574453,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 89.9459,
+      "eval_samples_per_second": 15.787,
+      "eval_steps_per_second": 1.979,
+      "step": 80
+    },
+    {
+      "entropy": 0.8096660938113928,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.1254287958145142,
+      "learning_rate": 8.700603238980251e-05,
+      "loss": 0.7170331001281738,
+      "mean_token_accuracy": 0.7994555257260799,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.792780208788561,
+      "eval_loss": 0.7442984580993652,
+      "eval_mean_token_accuracy": 0.7954704573984896,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 90.294,
+      "eval_samples_per_second": 15.726,
+      "eval_steps_per_second": 1.971,
+      "step": 100
+    },
+    {
+      "entropy": 0.7979829199612141,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2140507698059082,
+      "learning_rate": 0.00010458300863016666,
+      "loss": 0.7141861915588379,
+      "mean_token_accuracy": 0.8000014744699001,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.7938692486018277,
+      "eval_loss": 0.7279797196388245,
+      "eval_mean_token_accuracy": 0.7989798734027348,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 89.7189,
+      "eval_samples_per_second": 15.827,
+      "eval_steps_per_second": 1.984,
+      "step": 120
+    },
+    {
+      "entropy": 0.7916347607970238,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.0506458282470703,
+      "learning_rate": 0.0001221599848705308,
+      "loss": 0.7117055416107178,
+      "mean_token_accuracy": 0.7970046654343605,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.8052781063519167,
+      "eval_loss": 0.722576916217804,
+      "eval_mean_token_accuracy": 0.7982530523551984,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 90.0376,
+      "eval_samples_per_second": 15.771,
+      "eval_steps_per_second": 1.977,
+      "step": 140
+    },
+    {
+      "entropy": 0.7880666613578796,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8409207463264465,
+      "learning_rate": 0.00013973696111089492,
+      "loss": 0.7052523136138916,
+      "mean_token_accuracy": 0.8000123649835587,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7787443134891853,
+      "eval_loss": 0.7137772440910339,
+      "eval_mean_token_accuracy": 0.8014209287220173,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 90.2562,
+      "eval_samples_per_second": 15.733,
+      "eval_steps_per_second": 1.972,
+      "step": 160
+    },
+    {
+      "entropy": 0.7696007996797561,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.0182477235794067,
+      "learning_rate": 0.00015731393735125907,
+      "loss": 0.7008797645568847,
+      "mean_token_accuracy": 0.8007259473204613,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.752259682068664,
+      "eval_loss": 0.7171286344528198,
+      "eval_mean_token_accuracy": 0.80052122574174,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 90.0797,
+      "eval_samples_per_second": 15.764,
+      "eval_steps_per_second": 1.976,
+      "step": 180
+    },
+    {
+      "entropy": 0.7828126326203346,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9553632736206055,
+      "learning_rate": 0.0001748909135916232,
+      "loss": 0.7046853542327881,
+      "mean_token_accuracy": 0.7988562889397144,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7389638169427936,
+      "eval_loss": 0.7187935709953308,
+      "eval_mean_token_accuracy": 0.7995899999409579,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 89.9223,
+      "eval_samples_per_second": 15.791,
+      "eval_steps_per_second": 1.979,
+      "step": 200
+    },
+    {
+      "entropy": 0.7909286297857762,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 1.134459376335144,
+      "learning_rate": 0.00019246788983198735,
+      "loss": 0.6989312171936035,
+      "mean_token_accuracy": 0.7991565138101577,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7496965199374082,
+      "eval_loss": 0.7106321454048157,
+      "eval_mean_token_accuracy": 0.8046183442131857,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 89.6899,
+      "eval_samples_per_second": 15.832,
+      "eval_steps_per_second": 1.985,
+      "step": 220
+    },
+    {
+      "entropy": 0.7651963323354721,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.9557585716247559,
+      "learning_rate": 0.0002100448660723515,
+      "loss": 0.6984559059143066,
+      "mean_token_accuracy": 0.8019823372364044,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7615563079212488,
+      "eval_loss": 0.7098539471626282,
+      "eval_mean_token_accuracy": 0.8047233508544022,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 90.3752,
+      "eval_samples_per_second": 15.712,
+      "eval_steps_per_second": 1.97,
+      "step": 240
+    },
+    {
+      "entropy": 0.787507726252079,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.9279311299324036,
+      "learning_rate": 0.00022762184231271568,
+      "loss": 0.708138370513916,
+      "mean_token_accuracy": 0.8013308703899383,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7856640896100676,
+      "eval_loss": 0.7191570401191711,
+      "eval_mean_token_accuracy": 0.8024677249153008,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 93.7336,
+      "eval_samples_per_second": 15.149,
+      "eval_steps_per_second": 1.899,
+      "step": 260
+    },
+    {
+      "entropy": 0.768489234894514,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.7992776036262512,
+      "learning_rate": 0.0002451988185530798,
+      "loss": 0.7035849571228028,
+      "mean_token_accuracy": 0.8031105332076549,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7682498740346244,
+      "eval_loss": 0.7149022221565247,
+      "eval_mean_token_accuracy": 0.7973497995499814,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 89.7309,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 280
+    },
+    {
+      "entropy": 0.7720955617725849,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.1931514739990234,
+      "learning_rate": 0.0002627757947934439,
+      "loss": 0.6965402603149414,
+      "mean_token_accuracy": 0.8040348663926125,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7214708304807042,
+      "eval_loss": 0.7350823879241943,
+      "eval_mean_token_accuracy": 0.7942911925610532,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 89.6451,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 300
+    },
+    {
+      "entropy": 0.7930095426738262,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8994467258453369,
+      "learning_rate": 0.0002803527710338081,
+      "loss": 0.7331492900848389,
+      "mean_token_accuracy": 0.796185302734375,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.8142535488926963,
+      "eval_loss": 0.7237834334373474,
+      "eval_mean_token_accuracy": 0.799495273761535,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 90.6759,
+      "eval_samples_per_second": 15.66,
+      "eval_steps_per_second": 1.963,
+      "step": 320
+    },
+    {
+      "entropy": 0.7825102441012859,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 1.8224239349365234,
+      "learning_rate": 0.00029792974727417223,
+      "loss": 0.7331175804138184,
+      "mean_token_accuracy": 0.7977667152881622,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7327135443017724,
+      "eval_loss": 0.7388784289360046,
+      "eval_mean_token_accuracy": 0.8007041252730938,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 89.7692,
+      "eval_samples_per_second": 15.818,
+      "eval_steps_per_second": 1.983,
+      "step": 340
+    },
+    {
+      "entropy": 0.819331557303667,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 1.0407779216766357,
+      "learning_rate": 0.0003155067235145364,
+      "loss": 0.7456695556640625,
+      "mean_token_accuracy": 0.7936980701982975,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.8065585420372781,
+      "eval_loss": 0.735519528388977,
+      "eval_mean_token_accuracy": 0.7974206296245704,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 89.4336,
+      "eval_samples_per_second": 15.878,
+      "eval_steps_per_second": 1.99,
+      "step": 360
+    },
+    {
+      "entropy": 0.8114653021097183,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.1209490299224854,
+      "learning_rate": 0.0003330836997549005,
+      "loss": 0.7340899467468261,
+      "mean_token_accuracy": 0.7914272703230381,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.8025527285056168,
+      "eval_loss": 0.7295576930046082,
+      "eval_mean_token_accuracy": 0.7990576628218876,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 89.9579,
+      "eval_samples_per_second": 15.785,
+      "eval_steps_per_second": 1.979,
+      "step": 380
+    },
+    {
+      "entropy": 0.8042497783899307,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 1.2437798976898193,
+      "learning_rate": 0.0003506606759952647,
+      "loss": 0.7381744384765625,
+      "mean_token_accuracy": 0.7967362694442273,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7717011336530193,
+      "eval_loss": 0.7436173558235168,
+      "eval_mean_token_accuracy": 0.7975849482450592,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 89.8411,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 400
+    },
+    {
+      "entropy": 0.7987750010056929,
+      "epoch": 1.012070006035003,
+      "grad_norm": 1.2390918731689453,
+      "learning_rate": 0.000364721224843344,
+      "loss": 0.7317790031433106,
+      "mean_token_accuracy": 0.7986087551364651,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7333170604170038,
+      "eval_loss": 0.7456948757171631,
+      "eval_mean_token_accuracy": 0.7964897577682238,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 89.7288,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 420
+    },
+    {
+      "entropy": 0.7372976846992969,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.1552740335464478,
+      "learning_rate": 0.00036468510102269575,
+      "loss": 0.6734249114990234,
+      "mean_token_accuracy": 0.805834824591875,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.7130355526891987,
+      "eval_loss": 0.7555333375930786,
+      "eval_mean_token_accuracy": 0.7961652781186479,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 89.8713,
+      "eval_samples_per_second": 15.8,
+      "eval_steps_per_second": 1.981,
+      "step": 440
+    },
+    {
+      "entropy": 0.7465668715536594,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 1.1991469860076904,
+      "learning_rate": 0.0003645973816745066,
+      "loss": 0.6871879577636719,
+      "mean_token_accuracy": 0.8042450234293937,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7772979542110743,
+      "eval_loss": 0.7561826705932617,
+      "eval_mean_token_accuracy": 0.7953876427720102,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 89.8395,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 460
+    },
+    {
+      "entropy": 0.7477855801582336,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 1.325486660003662,
+      "learning_rate": 0.00036445809162231435,
+      "loss": 0.6836830139160156,
+      "mean_token_accuracy": 0.8021619468927383,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.7098110676481483,
+      "eval_loss": 0.758007287979126,
+      "eval_mean_token_accuracy": 0.7909953189030122,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 89.6101,
+      "eval_samples_per_second": 15.846,
+      "eval_steps_per_second": 1.986,
+      "step": 480
+    },
+    {
+      "entropy": 0.7632691070437432,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 1.0122262239456177,
+      "learning_rate": 0.0003642672702835562,
+      "loss": 0.7071213245391845,
+      "mean_token_accuracy": 0.8000680930912495,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7645382720432924,
+      "eval_loss": 0.7492606043815613,
+      "eval_mean_token_accuracy": 0.7976858676149604,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 89.6092,
+      "eval_samples_per_second": 15.847,
+      "eval_steps_per_second": 1.986,
+      "step": 500
+    },
+    {
+      "entropy": 0.7623726457357407,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.367774486541748,
+      "learning_rate": 0.00036402497165841384,
+      "loss": 0.6965175151824952,
+      "mean_token_accuracy": 0.8016501650214195,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7868083715438843,
+      "eval_loss": 0.7481449842453003,
+      "eval_mean_token_accuracy": 0.799330582779445,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 89.8241,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 520
+    },
+    {
+      "entropy": 0.7535207405686378,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 1.213614821434021,
+      "learning_rate": 0.00036373126431453207,
+      "loss": 0.7078310489654541,
+      "mean_token_accuracy": 0.7998907208442688,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7829091950748743,
+      "eval_loss": 0.7416072487831116,
+      "eval_mean_token_accuracy": 0.7969146250339036,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 89.4607,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 540
+    },
+    {
+      "entropy": 0.7527363449335098,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 1.0929982662200928,
+      "learning_rate": 0.00036338623136761495,
+      "loss": 0.7058369636535644,
+      "mean_token_accuracy": 0.7997087433934211,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.728419938114252,
+      "eval_loss": 0.744873046875,
+      "eval_mean_token_accuracy": 0.7966634296299366,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 89.6597,
+      "eval_samples_per_second": 15.838,
+      "eval_steps_per_second": 1.985,
+      "step": 560
+    },
+    {
+      "entropy": 0.7406099259853363,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.409364104270935,
+      "learning_rate": 0.00036298997045790515,
+      "loss": 0.6902852058410645,
+      "mean_token_accuracy": 0.805386807769537,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.6991632758231645,
+      "eval_loss": 0.7480319738388062,
+      "eval_mean_token_accuracy": 0.7970068006033308,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 89.8329,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 580
+    },
+    {
+      "entropy": 0.7459449704736472,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.7053111791610718,
+      "learning_rate": 0.00036254259372255277,
+      "loss": 0.7044640064239502,
+      "mean_token_accuracy": 0.8005018755793571,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7454566660891758,
+      "eval_loss": 0.7551484704017639,
+      "eval_mean_token_accuracy": 0.7897093731365846,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 89.8313,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 600
+    },
+    {
+      "entropy": 0.7621391348540782,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 1.174443244934082,
+      "learning_rate": 0.000362044227763882,
+      "loss": 0.7111573696136475,
+      "mean_token_accuracy": 0.798752411454916,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7664387949397055,
+      "eval_loss": 0.7365804314613342,
+      "eval_mean_token_accuracy": 0.7963101200843126,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 89.9623,
+      "eval_samples_per_second": 15.784,
+      "eval_steps_per_second": 1.979,
+      "step": 620
+    },
+    {
+      "entropy": 0.7424944408237935,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.2898184061050415,
+      "learning_rate": 0.000361495013613564,
+      "loss": 0.6913110733032226,
+      "mean_token_accuracy": 0.8009015507996082,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7255479529332579,
+      "eval_loss": 0.7345991730690002,
+      "eval_mean_token_accuracy": 0.8013592945056015,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 89.817,
+      "eval_samples_per_second": 15.81,
+      "eval_steps_per_second": 1.982,
+      "step": 640
+    },
+    {
+      "entropy": 0.7303367637097835,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.1310476064682007,
+      "learning_rate": 0.00036089510669270666,
+      "loss": 0.6875617027282714,
+      "mean_token_accuracy": 0.8035397171974182,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7101308248016271,
+      "eval_loss": 0.7394917011260986,
+      "eval_mean_token_accuracy": 0.7990792942850777,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 89.7431,
+      "eval_samples_per_second": 15.823,
+      "eval_steps_per_second": 1.983,
+      "step": 660
+    },
+    {
+      "entropy": 0.7387678384780884,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.9645389914512634,
+      "learning_rate": 0.0003602446767678725,
+      "loss": 0.6911204338073731,
+      "mean_token_accuracy": 0.8056333385407924,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7281073737010527,
+      "eval_loss": 0.7376708984375,
+      "eval_mean_token_accuracy": 0.7997356415464637,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 90.0338,
+      "eval_samples_per_second": 15.772,
+      "eval_steps_per_second": 1.977,
+      "step": 680
+    },
+    {
+      "entropy": 0.7449554048478604,
+      "epoch": 1.687990343995172,
+      "grad_norm": 1.9069629907608032,
+      "learning_rate": 0.0003595439079030364,
+      "loss": 0.7042286396026611,
+      "mean_token_accuracy": 0.8018639378249646,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7022633817088738,
+      "eval_loss": 0.7320939898490906,
+      "eval_mean_token_accuracy": 0.8006051541044471,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 89.8948,
+      "eval_samples_per_second": 15.796,
+      "eval_steps_per_second": 1.98,
+      "step": 700
+    },
+    {
+      "entropy": 0.723249051719904,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.8333877325057983,
+      "learning_rate": 0.00035879299840749777,
+      "loss": 0.7043714046478271,
+      "mean_token_accuracy": 0.8028345607221127,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.672436946898364,
+      "eval_loss": 0.713233470916748,
+      "eval_mean_token_accuracy": 0.8047760577684038,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 89.6394,
+      "eval_samples_per_second": 15.841,
+      "eval_steps_per_second": 1.986,
+      "step": 720
+    },
+    {
+      "entropy": 0.730014206469059,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.4580037593841553,
+      "learning_rate": 0.00035799216077976137,
+      "loss": 0.6985176563262939,
+      "mean_token_accuracy": 0.8015547141432762,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.6908873795123582,
+      "eval_loss": 0.7198202013969421,
+      "eval_mean_token_accuracy": 0.8024131682481659,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 89.3771,
+      "eval_samples_per_second": 15.888,
+      "eval_steps_per_second": 1.992,
+      "step": 740
+    },
+    {
+      "entropy": 0.7219605796039105,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.51682710647583,
+      "learning_rate": 0.000357141621647403,
+      "loss": 0.6844424724578857,
+      "mean_token_accuracy": 0.8032813094556331,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7113401109582922,
+      "eval_loss": 0.7195309400558472,
+      "eval_mean_token_accuracy": 0.8019187035185568,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 89.4658,
+      "eval_samples_per_second": 15.872,
+      "eval_steps_per_second": 1.99,
+      "step": 760
+    },
+    {
+      "entropy": 0.729826743900776,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 1.0838000774383545,
+      "learning_rate": 0.0003562416217029361,
+      "loss": 0.685858678817749,
+      "mean_token_accuracy": 0.8050567395985126,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.649370267484965,
+      "eval_loss": 0.7177029252052307,
+      "eval_mean_token_accuracy": 0.8050174043419656,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 89.2287,
+      "eval_samples_per_second": 15.914,
+      "eval_steps_per_second": 1.995,
+      "step": 780
+    },
+    {
+      "entropy": 0.6799364153295755,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 1.1364924907684326,
+      "learning_rate": 0.00035529241563569903,
+      "loss": 0.6714428901672364,
+      "mean_token_accuracy": 0.8087341658771038,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6603799047095052,
+      "eval_loss": 0.7201809287071228,
+      "eval_mean_token_accuracy": 0.802863868769635,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 89.2718,
+      "eval_samples_per_second": 15.906,
+      "eval_steps_per_second": 1.994,
+      "step": 800
+    },
+    {
+      "entropy": 0.7078610375523567,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 1.2191942930221558,
+      "learning_rate": 0.0003542942720597808,
+      "loss": 0.6996590614318847,
+      "mean_token_accuracy": 0.8049512051045895,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.7240545264120852,
+      "eval_loss": 0.7301309108734131,
+      "eval_mean_token_accuracy": 0.8015973166133581,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 89.3438,
+      "eval_samples_per_second": 15.894,
+      "eval_steps_per_second": 1.992,
+      "step": 820
+    },
+    {
+      "entropy": 0.6557391556826505,
+      "epoch": 2.024140012070006,
+      "grad_norm": 1.36203134059906,
+      "learning_rate": 0.0003532474734380064,
+      "loss": 0.6327113628387451,
+      "mean_token_accuracy": 0.8142829847026181,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6573808648613062,
+      "eval_loss": 0.7334365844726562,
+      "eval_mean_token_accuracy": 0.8043633893634496,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 89.3359,
+      "eval_samples_per_second": 15.895,
+      "eval_steps_per_second": 1.992,
+      "step": 840
+    },
+    {
+      "entropy": 0.6409128502011299,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 1.5804955959320068,
+      "learning_rate": 0.0003521523160020035,
+      "loss": 0.6023559093475341,
+      "mean_token_accuracy": 0.8222251623868942,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6983234788594621,
+      "eval_loss": 0.7232135534286499,
+      "eval_mean_token_accuracy": 0.8010654268639811,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 89.2615,
+      "eval_samples_per_second": 15.908,
+      "eval_steps_per_second": 1.994,
+      "step": 860
+    },
+    {
+      "entropy": 0.6151149723678827,
+      "epoch": 2.12070006035003,
+      "grad_norm": 1.1975436210632324,
+      "learning_rate": 0.00035100910966837193,
+      "loss": 0.5803564548492431,
+      "mean_token_accuracy": 0.8256325736641884,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6636555573243773,
+      "eval_loss": 0.7205662727355957,
+      "eval_mean_token_accuracy": 0.8036768369460374,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 89.2957,
+      "eval_samples_per_second": 15.902,
+      "eval_steps_per_second": 1.993,
+      "step": 880
+    },
+    {
+      "entropy": 0.6201727617532015,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9907131195068359,
+      "learning_rate": 0.0003498181779509813,
+      "loss": 0.6101872444152832,
+      "mean_token_accuracy": 0.8196512959897518,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.6473788086617931,
+      "eval_loss": 0.7062397003173828,
+      "eval_mean_token_accuracy": 0.8078362184963869,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 88.6198,
+      "eval_samples_per_second": 16.024,
+      "eval_steps_per_second": 2.009,
+      "step": 900
+    },
+    {
+      "entropy": 0.6097332935780286,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.9055633544921875,
+      "learning_rate": 0.00034857985786942026,
+      "loss": 0.5956094264984131,
+      "mean_token_accuracy": 0.8240575887262821,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.612629868341296,
+      "eval_loss": 0.7106770277023315,
+      "eval_mean_token_accuracy": 0.805324277181304,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 89.4383,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 920
+    },
+    {
+      "entropy": 0.6182701248675585,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.9513736367225647,
+      "learning_rate": 0.00034729449985362404,
+      "loss": 0.6019269466400147,
+      "mean_token_accuracy": 0.8187106013298034,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6437647456533453,
+      "eval_loss": 0.7173502445220947,
+      "eval_mean_token_accuracy": 0.8056439573175451,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.9568,
+      "eval_samples_per_second": 15.963,
+      "eval_steps_per_second": 2.001,
+      "step": 940
+    },
+    {
+      "entropy": 0.6508570514619351,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.9606207013130188,
+      "learning_rate": 0.0003459624676447067,
+      "loss": 0.6296409606933594,
+      "mean_token_accuracy": 0.8167732007801533,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6288247135248077,
+      "eval_loss": 0.7160675525665283,
+      "eval_mean_token_accuracy": 0.8066010689467527,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 89.0044,
+      "eval_samples_per_second": 15.954,
+      "eval_steps_per_second": 2.0,
+      "step": 960
+    },
+    {
+      "entropy": 0.6380081418901682,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.9677181839942932,
+      "learning_rate": 0.000344584138192027,
+      "loss": 0.6210546493530273,
+      "mean_token_accuracy": 0.8186135642230511,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6407805031604981,
+      "eval_loss": 0.7055043578147888,
+      "eval_mean_token_accuracy": 0.8070103157772107,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 89.5606,
+      "eval_samples_per_second": 15.855,
+      "eval_steps_per_second": 1.987,
+      "step": 980
+    },
+    {
+      "entropy": 0.6310251638293266,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 2.1046407222747803,
+      "learning_rate": 0.000343159901546516,
+      "loss": 0.6165830135345459,
+      "mean_token_accuracy": 0.8207855716347694,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6461507470420237,
+      "eval_loss": 0.7292136549949646,
+      "eval_mean_token_accuracy": 0.8030418261383356,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 89.315,
+      "eval_samples_per_second": 15.899,
+      "eval_steps_per_second": 1.993,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6256617344915867,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 1.1768349409103394,
+      "learning_rate": 0.0003416901607502972,
+      "loss": 0.6142709255218506,
+      "mean_token_accuracy": 0.8220669947564602,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6150523993406403,
+      "eval_loss": 0.7019563317298889,
+      "eval_mean_token_accuracy": 0.8094323099998946,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 89.419,
+      "eval_samples_per_second": 15.88,
+      "eval_steps_per_second": 1.991,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6126735735684633,
+      "epoch": 2.506940253470127,
+      "grad_norm": 1.8065496683120728,
+      "learning_rate": 0.00034017533172263055,
+      "loss": 0.6056458473205566,
+      "mean_token_accuracy": 0.8203762136399746,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6544345456562685,
+      "eval_loss": 0.7086517214775085,
+      "eval_mean_token_accuracy": 0.806709805901131,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 88.981,
+      "eval_samples_per_second": 15.958,
+      "eval_steps_per_second": 2.0,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6311754353344441,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 1.0576369762420654,
+      "learning_rate": 0.00033861584314221236,
+      "loss": 0.6176392555236816,
+      "mean_token_accuracy": 0.8177410490810871,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6273799200406235,
+      "eval_loss": 0.708163321018219,
+      "eval_mean_token_accuracy": 0.8032017099053672,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 89.7668,
+      "eval_samples_per_second": 15.819,
+      "eval_steps_per_second": 1.983,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6200249589979648,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.997065007686615,
+      "learning_rate": 0.0003370121363258637,
+      "loss": 0.6203888893127442,
+      "mean_token_accuracy": 0.8196275025606156,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6457576811983344,
+      "eval_loss": 0.6972290277481079,
+      "eval_mean_token_accuracy": 0.8095980849158898,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 89.2027,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 1080
+    },
+    {
+      "entropy": 0.626810473203659,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.865475058555603,
+      "learning_rate": 0.0003353646651036438,
+      "loss": 0.6102585792541504,
+      "mean_token_accuracy": 0.819708751142025,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.6412022431914726,
+      "eval_loss": 0.7036139965057373,
+      "eval_mean_token_accuracy": 0.8070317135098275,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 89.4733,
+      "eval_samples_per_second": 15.871,
+      "eval_steps_per_second": 1.989,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6581619590520859,
+      "epoch": 2.700060350030175,
+      "grad_norm": 1.00307035446167,
+      "learning_rate": 0.00033367389569042064,
+      "loss": 0.6397720336914062,
+      "mean_token_accuracy": 0.81413309648633,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6288736402318719,
+      "eval_loss": 0.6966825723648071,
+      "eval_mean_token_accuracy": 0.8094039669867312,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 88.7341,
+      "eval_samples_per_second": 16.003,
+      "eval_steps_per_second": 2.006,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6399412982165813,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.8589329719543457,
+      "learning_rate": 0.0003319403065539384,
+      "loss": 0.6347338199615479,
+      "mean_token_accuracy": 0.8139534957706929,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.683491913455256,
+      "eval_loss": 0.6984149217605591,
+      "eval_mean_token_accuracy": 0.8083989469522841,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 89.3929,
+      "eval_samples_per_second": 15.885,
+      "eval_steps_per_second": 1.991,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6264835461974144,
+      "epoch": 2.796620398310199,
+      "grad_norm": 0.7807853817939758,
+      "learning_rate": 0.0003301643882794163,
+      "loss": 0.6162629127502441,
+      "mean_token_accuracy": 0.8216292977333068,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6578792159476977,
+      "eval_loss": 0.6921441555023193,
+      "eval_mean_token_accuracy": 0.8094299177775223,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 89.2356,
+      "eval_samples_per_second": 15.913,
+      "eval_steps_per_second": 1.995,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6269129011780024,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 1.1675593852996826,
+      "learning_rate": 0.0003283466434307189,
+      "loss": 0.6235532760620117,
+      "mean_token_accuracy": 0.8188897788524627,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6533248447970058,
+      "eval_loss": 0.696849524974823,
+      "eval_mean_token_accuracy": 0.8102703713968898,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2602,
+      "eval_samples_per_second": 16.089,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.1821567693494067e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-120/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-120/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-120/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-120/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-120/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..7ce4b072fcbe5b636867e13937c56fedee3b68b4
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-120/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.02728394274790723,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-120/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-120/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-120/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-120/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-120/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-120/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-120/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-120/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..f131523f6622afe9f5febb16b447632b7c852cbd
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-120/trainer_state.json
@@ -0,0 +1,160 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.28968014484007243,
+  "eval_steps": 20,
+  "global_step": 120,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.024795526266098,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.945375919342041,
+      "learning_rate": 1.6698127428345936e-05,
+      "loss": 1.7825420379638672,
+      "mean_token_accuracy": 0.6315708436071873,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.29921497588747,
+      "eval_loss": 1.169131875038147,
+      "eval_mean_token_accuracy": 0.7185706342204233,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 91.7898,
+      "eval_samples_per_second": 15.47,
+      "eval_steps_per_second": 1.939,
+      "step": 20
+    },
+    {
+      "entropy": 1.0346344470977784,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.7916979789733887,
+      "learning_rate": 3.427510366871008e-05,
+      "loss": 0.9480395317077637,
+      "mean_token_accuracy": 0.7543429024517536,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 0.9673695430326997,
+      "eval_loss": 0.8588430285453796,
+      "eval_mean_token_accuracy": 0.7737233002534073,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 89.8219,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 40
+    },
+    {
+      "entropy": 0.8977566614747048,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7972187995910645,
+      "learning_rate": 5.185207990907422e-05,
+      "loss": 0.8029604911804199,
+      "mean_token_accuracy": 0.7818016350269318,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8436836982041263,
+      "eval_loss": 0.7893033623695374,
+      "eval_mean_token_accuracy": 0.7858914984076211,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 89.9349,
+      "eval_samples_per_second": 15.789,
+      "eval_steps_per_second": 1.979,
+      "step": 60
+    },
+    {
+      "entropy": 0.8650321319699288,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.2128729820251465,
+      "learning_rate": 6.942905614943836e-05,
+      "loss": 0.77417893409729,
+      "mean_token_accuracy": 0.7875479467213153,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8209857180547179,
+      "eval_loss": 0.7572280168533325,
+      "eval_mean_token_accuracy": 0.7925970383574453,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 89.9459,
+      "eval_samples_per_second": 15.787,
+      "eval_steps_per_second": 1.979,
+      "step": 80
+    },
+    {
+      "entropy": 0.8096660938113928,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.1254287958145142,
+      "learning_rate": 8.700603238980251e-05,
+      "loss": 0.7170331001281738,
+      "mean_token_accuracy": 0.7994555257260799,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.792780208788561,
+      "eval_loss": 0.7442984580993652,
+      "eval_mean_token_accuracy": 0.7954704573984896,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 90.294,
+      "eval_samples_per_second": 15.726,
+      "eval_steps_per_second": 1.971,
+      "step": 100
+    },
+    {
+      "entropy": 0.7979829199612141,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2140507698059082,
+      "learning_rate": 0.00010458300863016666,
+      "loss": 0.7141861915588379,
+      "mean_token_accuracy": 0.8000014744699001,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.7938692486018277,
+      "eval_loss": 0.7279797196388245,
+      "eval_mean_token_accuracy": 0.7989798734027348,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 89.7189,
+      "eval_samples_per_second": 15.827,
+      "eval_steps_per_second": 1.984,
+      "step": 120
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.2278999724859392e+16,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1200/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1200/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1200/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1200/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1200/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..7ce4b072fcbe5b636867e13937c56fedee3b68b4
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1200/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.02728394274790723,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1200/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1200/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1200/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1200/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1200/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1200/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1200/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1200/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..e4f56b396f4cf722f14498333e1513140cf50226
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1200/trainer_state.json
@@ -0,0 +1,1294 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 2.8931804465902236,
+  "eval_steps": 20,
+  "global_step": 1200,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.024795526266098,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.945375919342041,
+      "learning_rate": 1.6698127428345936e-05,
+      "loss": 1.7825420379638672,
+      "mean_token_accuracy": 0.6315708436071873,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.29921497588747,
+      "eval_loss": 1.169131875038147,
+      "eval_mean_token_accuracy": 0.7185706342204233,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 91.7898,
+      "eval_samples_per_second": 15.47,
+      "eval_steps_per_second": 1.939,
+      "step": 20
+    },
+    {
+      "entropy": 1.0346344470977784,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.7916979789733887,
+      "learning_rate": 3.427510366871008e-05,
+      "loss": 0.9480395317077637,
+      "mean_token_accuracy": 0.7543429024517536,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 0.9673695430326997,
+      "eval_loss": 0.8588430285453796,
+      "eval_mean_token_accuracy": 0.7737233002534073,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 89.8219,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 40
+    },
+    {
+      "entropy": 0.8977566614747048,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7972187995910645,
+      "learning_rate": 5.185207990907422e-05,
+      "loss": 0.8029604911804199,
+      "mean_token_accuracy": 0.7818016350269318,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8436836982041263,
+      "eval_loss": 0.7893033623695374,
+      "eval_mean_token_accuracy": 0.7858914984076211,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 89.9349,
+      "eval_samples_per_second": 15.789,
+      "eval_steps_per_second": 1.979,
+      "step": 60
+    },
+    {
+      "entropy": 0.8650321319699288,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.2128729820251465,
+      "learning_rate": 6.942905614943836e-05,
+      "loss": 0.77417893409729,
+      "mean_token_accuracy": 0.7875479467213153,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8209857180547179,
+      "eval_loss": 0.7572280168533325,
+      "eval_mean_token_accuracy": 0.7925970383574453,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 89.9459,
+      "eval_samples_per_second": 15.787,
+      "eval_steps_per_second": 1.979,
+      "step": 80
+    },
+    {
+      "entropy": 0.8096660938113928,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.1254287958145142,
+      "learning_rate": 8.700603238980251e-05,
+      "loss": 0.7170331001281738,
+      "mean_token_accuracy": 0.7994555257260799,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.792780208788561,
+      "eval_loss": 0.7442984580993652,
+      "eval_mean_token_accuracy": 0.7954704573984896,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 90.294,
+      "eval_samples_per_second": 15.726,
+      "eval_steps_per_second": 1.971,
+      "step": 100
+    },
+    {
+      "entropy": 0.7979829199612141,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2140507698059082,
+      "learning_rate": 0.00010458300863016666,
+      "loss": 0.7141861915588379,
+      "mean_token_accuracy": 0.8000014744699001,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.7938692486018277,
+      "eval_loss": 0.7279797196388245,
+      "eval_mean_token_accuracy": 0.7989798734027348,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 89.7189,
+      "eval_samples_per_second": 15.827,
+      "eval_steps_per_second": 1.984,
+      "step": 120
+    },
+    {
+      "entropy": 0.7916347607970238,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.0506458282470703,
+      "learning_rate": 0.0001221599848705308,
+      "loss": 0.7117055416107178,
+      "mean_token_accuracy": 0.7970046654343605,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.8052781063519167,
+      "eval_loss": 0.722576916217804,
+      "eval_mean_token_accuracy": 0.7982530523551984,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 90.0376,
+      "eval_samples_per_second": 15.771,
+      "eval_steps_per_second": 1.977,
+      "step": 140
+    },
+    {
+      "entropy": 0.7880666613578796,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8409207463264465,
+      "learning_rate": 0.00013973696111089492,
+      "loss": 0.7052523136138916,
+      "mean_token_accuracy": 0.8000123649835587,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7787443134891853,
+      "eval_loss": 0.7137772440910339,
+      "eval_mean_token_accuracy": 0.8014209287220173,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 90.2562,
+      "eval_samples_per_second": 15.733,
+      "eval_steps_per_second": 1.972,
+      "step": 160
+    },
+    {
+      "entropy": 0.7696007996797561,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.0182477235794067,
+      "learning_rate": 0.00015731393735125907,
+      "loss": 0.7008797645568847,
+      "mean_token_accuracy": 0.8007259473204613,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.752259682068664,
+      "eval_loss": 0.7171286344528198,
+      "eval_mean_token_accuracy": 0.80052122574174,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 90.0797,
+      "eval_samples_per_second": 15.764,
+      "eval_steps_per_second": 1.976,
+      "step": 180
+    },
+    {
+      "entropy": 0.7828126326203346,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9553632736206055,
+      "learning_rate": 0.0001748909135916232,
+      "loss": 0.7046853542327881,
+      "mean_token_accuracy": 0.7988562889397144,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7389638169427936,
+      "eval_loss": 0.7187935709953308,
+      "eval_mean_token_accuracy": 0.7995899999409579,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 89.9223,
+      "eval_samples_per_second": 15.791,
+      "eval_steps_per_second": 1.979,
+      "step": 200
+    },
+    {
+      "entropy": 0.7909286297857762,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 1.134459376335144,
+      "learning_rate": 0.00019246788983198735,
+      "loss": 0.6989312171936035,
+      "mean_token_accuracy": 0.7991565138101577,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7496965199374082,
+      "eval_loss": 0.7106321454048157,
+      "eval_mean_token_accuracy": 0.8046183442131857,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 89.6899,
+      "eval_samples_per_second": 15.832,
+      "eval_steps_per_second": 1.985,
+      "step": 220
+    },
+    {
+      "entropy": 0.7651963323354721,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.9557585716247559,
+      "learning_rate": 0.0002100448660723515,
+      "loss": 0.6984559059143066,
+      "mean_token_accuracy": 0.8019823372364044,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7615563079212488,
+      "eval_loss": 0.7098539471626282,
+      "eval_mean_token_accuracy": 0.8047233508544022,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 90.3752,
+      "eval_samples_per_second": 15.712,
+      "eval_steps_per_second": 1.97,
+      "step": 240
+    },
+    {
+      "entropy": 0.787507726252079,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.9279311299324036,
+      "learning_rate": 0.00022762184231271568,
+      "loss": 0.708138370513916,
+      "mean_token_accuracy": 0.8013308703899383,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7856640896100676,
+      "eval_loss": 0.7191570401191711,
+      "eval_mean_token_accuracy": 0.8024677249153008,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 93.7336,
+      "eval_samples_per_second": 15.149,
+      "eval_steps_per_second": 1.899,
+      "step": 260
+    },
+    {
+      "entropy": 0.768489234894514,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.7992776036262512,
+      "learning_rate": 0.0002451988185530798,
+      "loss": 0.7035849571228028,
+      "mean_token_accuracy": 0.8031105332076549,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7682498740346244,
+      "eval_loss": 0.7149022221565247,
+      "eval_mean_token_accuracy": 0.7973497995499814,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 89.7309,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 280
+    },
+    {
+      "entropy": 0.7720955617725849,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.1931514739990234,
+      "learning_rate": 0.0002627757947934439,
+      "loss": 0.6965402603149414,
+      "mean_token_accuracy": 0.8040348663926125,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7214708304807042,
+      "eval_loss": 0.7350823879241943,
+      "eval_mean_token_accuracy": 0.7942911925610532,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 89.6451,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 300
+    },
+    {
+      "entropy": 0.7930095426738262,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8994467258453369,
+      "learning_rate": 0.0002803527710338081,
+      "loss": 0.7331492900848389,
+      "mean_token_accuracy": 0.796185302734375,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.8142535488926963,
+      "eval_loss": 0.7237834334373474,
+      "eval_mean_token_accuracy": 0.799495273761535,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 90.6759,
+      "eval_samples_per_second": 15.66,
+      "eval_steps_per_second": 1.963,
+      "step": 320
+    },
+    {
+      "entropy": 0.7825102441012859,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 1.8224239349365234,
+      "learning_rate": 0.00029792974727417223,
+      "loss": 0.7331175804138184,
+      "mean_token_accuracy": 0.7977667152881622,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7327135443017724,
+      "eval_loss": 0.7388784289360046,
+      "eval_mean_token_accuracy": 0.8007041252730938,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 89.7692,
+      "eval_samples_per_second": 15.818,
+      "eval_steps_per_second": 1.983,
+      "step": 340
+    },
+    {
+      "entropy": 0.819331557303667,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 1.0407779216766357,
+      "learning_rate": 0.0003155067235145364,
+      "loss": 0.7456695556640625,
+      "mean_token_accuracy": 0.7936980701982975,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.8065585420372781,
+      "eval_loss": 0.735519528388977,
+      "eval_mean_token_accuracy": 0.7974206296245704,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 89.4336,
+      "eval_samples_per_second": 15.878,
+      "eval_steps_per_second": 1.99,
+      "step": 360
+    },
+    {
+      "entropy": 0.8114653021097183,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.1209490299224854,
+      "learning_rate": 0.0003330836997549005,
+      "loss": 0.7340899467468261,
+      "mean_token_accuracy": 0.7914272703230381,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.8025527285056168,
+      "eval_loss": 0.7295576930046082,
+      "eval_mean_token_accuracy": 0.7990576628218876,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 89.9579,
+      "eval_samples_per_second": 15.785,
+      "eval_steps_per_second": 1.979,
+      "step": 380
+    },
+    {
+      "entropy": 0.8042497783899307,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 1.2437798976898193,
+      "learning_rate": 0.0003506606759952647,
+      "loss": 0.7381744384765625,
+      "mean_token_accuracy": 0.7967362694442273,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7717011336530193,
+      "eval_loss": 0.7436173558235168,
+      "eval_mean_token_accuracy": 0.7975849482450592,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 89.8411,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 400
+    },
+    {
+      "entropy": 0.7987750010056929,
+      "epoch": 1.012070006035003,
+      "grad_norm": 1.2390918731689453,
+      "learning_rate": 0.000364721224843344,
+      "loss": 0.7317790031433106,
+      "mean_token_accuracy": 0.7986087551364651,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7333170604170038,
+      "eval_loss": 0.7456948757171631,
+      "eval_mean_token_accuracy": 0.7964897577682238,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 89.7288,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 420
+    },
+    {
+      "entropy": 0.7372976846992969,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.1552740335464478,
+      "learning_rate": 0.00036468510102269575,
+      "loss": 0.6734249114990234,
+      "mean_token_accuracy": 0.805834824591875,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.7130355526891987,
+      "eval_loss": 0.7555333375930786,
+      "eval_mean_token_accuracy": 0.7961652781186479,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 89.8713,
+      "eval_samples_per_second": 15.8,
+      "eval_steps_per_second": 1.981,
+      "step": 440
+    },
+    {
+      "entropy": 0.7465668715536594,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 1.1991469860076904,
+      "learning_rate": 0.0003645973816745066,
+      "loss": 0.6871879577636719,
+      "mean_token_accuracy": 0.8042450234293937,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7772979542110743,
+      "eval_loss": 0.7561826705932617,
+      "eval_mean_token_accuracy": 0.7953876427720102,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 89.8395,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 460
+    },
+    {
+      "entropy": 0.7477855801582336,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 1.325486660003662,
+      "learning_rate": 0.00036445809162231435,
+      "loss": 0.6836830139160156,
+      "mean_token_accuracy": 0.8021619468927383,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.7098110676481483,
+      "eval_loss": 0.758007287979126,
+      "eval_mean_token_accuracy": 0.7909953189030122,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 89.6101,
+      "eval_samples_per_second": 15.846,
+      "eval_steps_per_second": 1.986,
+      "step": 480
+    },
+    {
+      "entropy": 0.7632691070437432,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 1.0122262239456177,
+      "learning_rate": 0.0003642672702835562,
+      "loss": 0.7071213245391845,
+      "mean_token_accuracy": 0.8000680930912495,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7645382720432924,
+      "eval_loss": 0.7492606043815613,
+      "eval_mean_token_accuracy": 0.7976858676149604,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 89.6092,
+      "eval_samples_per_second": 15.847,
+      "eval_steps_per_second": 1.986,
+      "step": 500
+    },
+    {
+      "entropy": 0.7623726457357407,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.367774486541748,
+      "learning_rate": 0.00036402497165841384,
+      "loss": 0.6965175151824952,
+      "mean_token_accuracy": 0.8016501650214195,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7868083715438843,
+      "eval_loss": 0.7481449842453003,
+      "eval_mean_token_accuracy": 0.799330582779445,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 89.8241,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 520
+    },
+    {
+      "entropy": 0.7535207405686378,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 1.213614821434021,
+      "learning_rate": 0.00036373126431453207,
+      "loss": 0.7078310489654541,
+      "mean_token_accuracy": 0.7998907208442688,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7829091950748743,
+      "eval_loss": 0.7416072487831116,
+      "eval_mean_token_accuracy": 0.7969146250339036,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 89.4607,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 540
+    },
+    {
+      "entropy": 0.7527363449335098,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 1.0929982662200928,
+      "learning_rate": 0.00036338623136761495,
+      "loss": 0.7058369636535644,
+      "mean_token_accuracy": 0.7997087433934211,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.728419938114252,
+      "eval_loss": 0.744873046875,
+      "eval_mean_token_accuracy": 0.7966634296299366,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 89.6597,
+      "eval_samples_per_second": 15.838,
+      "eval_steps_per_second": 1.985,
+      "step": 560
+    },
+    {
+      "entropy": 0.7406099259853363,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.409364104270935,
+      "learning_rate": 0.00036298997045790515,
+      "loss": 0.6902852058410645,
+      "mean_token_accuracy": 0.805386807769537,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.6991632758231645,
+      "eval_loss": 0.7480319738388062,
+      "eval_mean_token_accuracy": 0.7970068006033308,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 89.8329,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 580
+    },
+    {
+      "entropy": 0.7459449704736472,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.7053111791610718,
+      "learning_rate": 0.00036254259372255277,
+      "loss": 0.7044640064239502,
+      "mean_token_accuracy": 0.8005018755793571,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7454566660891758,
+      "eval_loss": 0.7551484704017639,
+      "eval_mean_token_accuracy": 0.7897093731365846,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 89.8313,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 600
+    },
+    {
+      "entropy": 0.7621391348540782,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 1.174443244934082,
+      "learning_rate": 0.000362044227763882,
+      "loss": 0.7111573696136475,
+      "mean_token_accuracy": 0.798752411454916,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7664387949397055,
+      "eval_loss": 0.7365804314613342,
+      "eval_mean_token_accuracy": 0.7963101200843126,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 89.9623,
+      "eval_samples_per_second": 15.784,
+      "eval_steps_per_second": 1.979,
+      "step": 620
+    },
+    {
+      "entropy": 0.7424944408237935,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.2898184061050415,
+      "learning_rate": 0.000361495013613564,
+      "loss": 0.6913110733032226,
+      "mean_token_accuracy": 0.8009015507996082,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7255479529332579,
+      "eval_loss": 0.7345991730690002,
+      "eval_mean_token_accuracy": 0.8013592945056015,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 89.817,
+      "eval_samples_per_second": 15.81,
+      "eval_steps_per_second": 1.982,
+      "step": 640
+    },
+    {
+      "entropy": 0.7303367637097835,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.1310476064682007,
+      "learning_rate": 0.00036089510669270666,
+      "loss": 0.6875617027282714,
+      "mean_token_accuracy": 0.8035397171974182,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7101308248016271,
+      "eval_loss": 0.7394917011260986,
+      "eval_mean_token_accuracy": 0.7990792942850777,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 89.7431,
+      "eval_samples_per_second": 15.823,
+      "eval_steps_per_second": 1.983,
+      "step": 660
+    },
+    {
+      "entropy": 0.7387678384780884,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.9645389914512634,
+      "learning_rate": 0.0003602446767678725,
+      "loss": 0.6911204338073731,
+      "mean_token_accuracy": 0.8056333385407924,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7281073737010527,
+      "eval_loss": 0.7376708984375,
+      "eval_mean_token_accuracy": 0.7997356415464637,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 90.0338,
+      "eval_samples_per_second": 15.772,
+      "eval_steps_per_second": 1.977,
+      "step": 680
+    },
+    {
+      "entropy": 0.7449554048478604,
+      "epoch": 1.687990343995172,
+      "grad_norm": 1.9069629907608032,
+      "learning_rate": 0.0003595439079030364,
+      "loss": 0.7042286396026611,
+      "mean_token_accuracy": 0.8018639378249646,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7022633817088738,
+      "eval_loss": 0.7320939898490906,
+      "eval_mean_token_accuracy": 0.8006051541044471,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 89.8948,
+      "eval_samples_per_second": 15.796,
+      "eval_steps_per_second": 1.98,
+      "step": 700
+    },
+    {
+      "entropy": 0.723249051719904,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.8333877325057983,
+      "learning_rate": 0.00035879299840749777,
+      "loss": 0.7043714046478271,
+      "mean_token_accuracy": 0.8028345607221127,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.672436946898364,
+      "eval_loss": 0.713233470916748,
+      "eval_mean_token_accuracy": 0.8047760577684038,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 89.6394,
+      "eval_samples_per_second": 15.841,
+      "eval_steps_per_second": 1.986,
+      "step": 720
+    },
+    {
+      "entropy": 0.730014206469059,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.4580037593841553,
+      "learning_rate": 0.00035799216077976137,
+      "loss": 0.6985176563262939,
+      "mean_token_accuracy": 0.8015547141432762,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.6908873795123582,
+      "eval_loss": 0.7198202013969421,
+      "eval_mean_token_accuracy": 0.8024131682481659,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 89.3771,
+      "eval_samples_per_second": 15.888,
+      "eval_steps_per_second": 1.992,
+      "step": 740
+    },
+    {
+      "entropy": 0.7219605796039105,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.51682710647583,
+      "learning_rate": 0.000357141621647403,
+      "loss": 0.6844424724578857,
+      "mean_token_accuracy": 0.8032813094556331,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7113401109582922,
+      "eval_loss": 0.7195309400558472,
+      "eval_mean_token_accuracy": 0.8019187035185568,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 89.4658,
+      "eval_samples_per_second": 15.872,
+      "eval_steps_per_second": 1.99,
+      "step": 760
+    },
+    {
+      "entropy": 0.729826743900776,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 1.0838000774383545,
+      "learning_rate": 0.0003562416217029361,
+      "loss": 0.685858678817749,
+      "mean_token_accuracy": 0.8050567395985126,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.649370267484965,
+      "eval_loss": 0.7177029252052307,
+      "eval_mean_token_accuracy": 0.8050174043419656,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 89.2287,
+      "eval_samples_per_second": 15.914,
+      "eval_steps_per_second": 1.995,
+      "step": 780
+    },
+    {
+      "entropy": 0.6799364153295755,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 1.1364924907684326,
+      "learning_rate": 0.00035529241563569903,
+      "loss": 0.6714428901672364,
+      "mean_token_accuracy": 0.8087341658771038,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6603799047095052,
+      "eval_loss": 0.7201809287071228,
+      "eval_mean_token_accuracy": 0.802863868769635,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 89.2718,
+      "eval_samples_per_second": 15.906,
+      "eval_steps_per_second": 1.994,
+      "step": 800
+    },
+    {
+      "entropy": 0.7078610375523567,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 1.2191942930221558,
+      "learning_rate": 0.0003542942720597808,
+      "loss": 0.6996590614318847,
+      "mean_token_accuracy": 0.8049512051045895,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.7240545264120852,
+      "eval_loss": 0.7301309108734131,
+      "eval_mean_token_accuracy": 0.8015973166133581,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 89.3438,
+      "eval_samples_per_second": 15.894,
+      "eval_steps_per_second": 1.992,
+      "step": 820
+    },
+    {
+      "entropy": 0.6557391556826505,
+      "epoch": 2.024140012070006,
+      "grad_norm": 1.36203134059906,
+      "learning_rate": 0.0003532474734380064,
+      "loss": 0.6327113628387451,
+      "mean_token_accuracy": 0.8142829847026181,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6573808648613062,
+      "eval_loss": 0.7334365844726562,
+      "eval_mean_token_accuracy": 0.8043633893634496,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 89.3359,
+      "eval_samples_per_second": 15.895,
+      "eval_steps_per_second": 1.992,
+      "step": 840
+    },
+    {
+      "entropy": 0.6409128502011299,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 1.5804955959320068,
+      "learning_rate": 0.0003521523160020035,
+      "loss": 0.6023559093475341,
+      "mean_token_accuracy": 0.8222251623868942,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6983234788594621,
+      "eval_loss": 0.7232135534286499,
+      "eval_mean_token_accuracy": 0.8010654268639811,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 89.2615,
+      "eval_samples_per_second": 15.908,
+      "eval_steps_per_second": 1.994,
+      "step": 860
+    },
+    {
+      "entropy": 0.6151149723678827,
+      "epoch": 2.12070006035003,
+      "grad_norm": 1.1975436210632324,
+      "learning_rate": 0.00035100910966837193,
+      "loss": 0.5803564548492431,
+      "mean_token_accuracy": 0.8256325736641884,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6636555573243773,
+      "eval_loss": 0.7205662727355957,
+      "eval_mean_token_accuracy": 0.8036768369460374,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 89.2957,
+      "eval_samples_per_second": 15.902,
+      "eval_steps_per_second": 1.993,
+      "step": 880
+    },
+    {
+      "entropy": 0.6201727617532015,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9907131195068359,
+      "learning_rate": 0.0003498181779509813,
+      "loss": 0.6101872444152832,
+      "mean_token_accuracy": 0.8196512959897518,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.6473788086617931,
+      "eval_loss": 0.7062397003173828,
+      "eval_mean_token_accuracy": 0.8078362184963869,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 88.6198,
+      "eval_samples_per_second": 16.024,
+      "eval_steps_per_second": 2.009,
+      "step": 900
+    },
+    {
+      "entropy": 0.6097332935780286,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.9055633544921875,
+      "learning_rate": 0.00034857985786942026,
+      "loss": 0.5956094264984131,
+      "mean_token_accuracy": 0.8240575887262821,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.612629868341296,
+      "eval_loss": 0.7106770277023315,
+      "eval_mean_token_accuracy": 0.805324277181304,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 89.4383,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 920
+    },
+    {
+      "entropy": 0.6182701248675585,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.9513736367225647,
+      "learning_rate": 0.00034729449985362404,
+      "loss": 0.6019269466400147,
+      "mean_token_accuracy": 0.8187106013298034,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6437647456533453,
+      "eval_loss": 0.7173502445220947,
+      "eval_mean_token_accuracy": 0.8056439573175451,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.9568,
+      "eval_samples_per_second": 15.963,
+      "eval_steps_per_second": 2.001,
+      "step": 940
+    },
+    {
+      "entropy": 0.6508570514619351,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.9606207013130188,
+      "learning_rate": 0.0003459624676447067,
+      "loss": 0.6296409606933594,
+      "mean_token_accuracy": 0.8167732007801533,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6288247135248077,
+      "eval_loss": 0.7160675525665283,
+      "eval_mean_token_accuracy": 0.8066010689467527,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 89.0044,
+      "eval_samples_per_second": 15.954,
+      "eval_steps_per_second": 2.0,
+      "step": 960
+    },
+    {
+      "entropy": 0.6380081418901682,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.9677181839942932,
+      "learning_rate": 0.000344584138192027,
+      "loss": 0.6210546493530273,
+      "mean_token_accuracy": 0.8186135642230511,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6407805031604981,
+      "eval_loss": 0.7055043578147888,
+      "eval_mean_token_accuracy": 0.8070103157772107,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 89.5606,
+      "eval_samples_per_second": 15.855,
+      "eval_steps_per_second": 1.987,
+      "step": 980
+    },
+    {
+      "entropy": 0.6310251638293266,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 2.1046407222747803,
+      "learning_rate": 0.000343159901546516,
+      "loss": 0.6165830135345459,
+      "mean_token_accuracy": 0.8207855716347694,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6461507470420237,
+      "eval_loss": 0.7292136549949646,
+      "eval_mean_token_accuracy": 0.8030418261383356,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 89.315,
+      "eval_samples_per_second": 15.899,
+      "eval_steps_per_second": 1.993,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6256617344915867,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 1.1768349409103394,
+      "learning_rate": 0.0003416901607502972,
+      "loss": 0.6142709255218506,
+      "mean_token_accuracy": 0.8220669947564602,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6150523993406403,
+      "eval_loss": 0.7019563317298889,
+      "eval_mean_token_accuracy": 0.8094323099998946,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 89.419,
+      "eval_samples_per_second": 15.88,
+      "eval_steps_per_second": 1.991,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6126735735684633,
+      "epoch": 2.506940253470127,
+      "grad_norm": 1.8065496683120728,
+      "learning_rate": 0.00034017533172263055,
+      "loss": 0.6056458473205566,
+      "mean_token_accuracy": 0.8203762136399746,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6544345456562685,
+      "eval_loss": 0.7086517214775085,
+      "eval_mean_token_accuracy": 0.806709805901131,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 88.981,
+      "eval_samples_per_second": 15.958,
+      "eval_steps_per_second": 2.0,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6311754353344441,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 1.0576369762420654,
+      "learning_rate": 0.00033861584314221236,
+      "loss": 0.6176392555236816,
+      "mean_token_accuracy": 0.8177410490810871,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6273799200406235,
+      "eval_loss": 0.708163321018219,
+      "eval_mean_token_accuracy": 0.8032017099053672,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 89.7668,
+      "eval_samples_per_second": 15.819,
+      "eval_steps_per_second": 1.983,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6200249589979648,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.997065007686615,
+      "learning_rate": 0.0003370121363258637,
+      "loss": 0.6203888893127442,
+      "mean_token_accuracy": 0.8196275025606156,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6457576811983344,
+      "eval_loss": 0.6972290277481079,
+      "eval_mean_token_accuracy": 0.8095980849158898,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 89.2027,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 1080
+    },
+    {
+      "entropy": 0.626810473203659,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.865475058555603,
+      "learning_rate": 0.0003353646651036438,
+      "loss": 0.6102585792541504,
+      "mean_token_accuracy": 0.819708751142025,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.6412022431914726,
+      "eval_loss": 0.7036139965057373,
+      "eval_mean_token_accuracy": 0.8070317135098275,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 89.4733,
+      "eval_samples_per_second": 15.871,
+      "eval_steps_per_second": 1.989,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6581619590520859,
+      "epoch": 2.700060350030175,
+      "grad_norm": 1.00307035446167,
+      "learning_rate": 0.00033367389569042064,
+      "loss": 0.6397720336914062,
+      "mean_token_accuracy": 0.81413309648633,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6288736402318719,
+      "eval_loss": 0.6966825723648071,
+      "eval_mean_token_accuracy": 0.8094039669867312,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 88.7341,
+      "eval_samples_per_second": 16.003,
+      "eval_steps_per_second": 2.006,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6399412982165813,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.8589329719543457,
+      "learning_rate": 0.0003319403065539384,
+      "loss": 0.6347338199615479,
+      "mean_token_accuracy": 0.8139534957706929,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.683491913455256,
+      "eval_loss": 0.6984149217605591,
+      "eval_mean_token_accuracy": 0.8083989469522841,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 89.3929,
+      "eval_samples_per_second": 15.885,
+      "eval_steps_per_second": 1.991,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6264835461974144,
+      "epoch": 2.796620398310199,
+      "grad_norm": 0.7807853817939758,
+      "learning_rate": 0.0003301643882794163,
+      "loss": 0.6162629127502441,
+      "mean_token_accuracy": 0.8216292977333068,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6578792159476977,
+      "eval_loss": 0.6921441555023193,
+      "eval_mean_token_accuracy": 0.8094299177775223,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 89.2356,
+      "eval_samples_per_second": 15.913,
+      "eval_steps_per_second": 1.995,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6269129011780024,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 1.1675593852996826,
+      "learning_rate": 0.0003283466434307189,
+      "loss": 0.6235532760620117,
+      "mean_token_accuracy": 0.8188897788524627,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6533248447970058,
+      "eval_loss": 0.696849524974823,
+      "eval_mean_token_accuracy": 0.8102703713968898,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2602,
+      "eval_samples_per_second": 16.089,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.6259873781353236,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.9596486687660217,
+      "learning_rate": 0.00032648758640813655,
+      "loss": 0.6118728160858155,
+      "mean_token_accuracy": 0.8171859130263328,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6212078754821521,
+      "eval_loss": 0.6942870020866394,
+      "eval_mean_token_accuracy": 0.810484343365337,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 89.4375,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 1200
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.202393633684951e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1220/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1220/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1220/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1220/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1220/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..7ce4b072fcbe5b636867e13937c56fedee3b68b4
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1220/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.02728394274790723,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1220/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1220/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1220/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1220/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1220/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1220/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1220/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1220/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..0f861783a8513552d6db099d26e9e700ed1ef3a0
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1220/trainer_state.json
@@ -0,0 +1,1315 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 2.9414604707302354,
+  "eval_steps": 20,
+  "global_step": 1220,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.024795526266098,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.945375919342041,
+      "learning_rate": 1.6698127428345936e-05,
+      "loss": 1.7825420379638672,
+      "mean_token_accuracy": 0.6315708436071873,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.29921497588747,
+      "eval_loss": 1.169131875038147,
+      "eval_mean_token_accuracy": 0.7185706342204233,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 91.7898,
+      "eval_samples_per_second": 15.47,
+      "eval_steps_per_second": 1.939,
+      "step": 20
+    },
+    {
+      "entropy": 1.0346344470977784,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.7916979789733887,
+      "learning_rate": 3.427510366871008e-05,
+      "loss": 0.9480395317077637,
+      "mean_token_accuracy": 0.7543429024517536,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 0.9673695430326997,
+      "eval_loss": 0.8588430285453796,
+      "eval_mean_token_accuracy": 0.7737233002534073,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 89.8219,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 40
+    },
+    {
+      "entropy": 0.8977566614747048,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7972187995910645,
+      "learning_rate": 5.185207990907422e-05,
+      "loss": 0.8029604911804199,
+      "mean_token_accuracy": 0.7818016350269318,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8436836982041263,
+      "eval_loss": 0.7893033623695374,
+      "eval_mean_token_accuracy": 0.7858914984076211,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 89.9349,
+      "eval_samples_per_second": 15.789,
+      "eval_steps_per_second": 1.979,
+      "step": 60
+    },
+    {
+      "entropy": 0.8650321319699288,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.2128729820251465,
+      "learning_rate": 6.942905614943836e-05,
+      "loss": 0.77417893409729,
+      "mean_token_accuracy": 0.7875479467213153,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8209857180547179,
+      "eval_loss": 0.7572280168533325,
+      "eval_mean_token_accuracy": 0.7925970383574453,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 89.9459,
+      "eval_samples_per_second": 15.787,
+      "eval_steps_per_second": 1.979,
+      "step": 80
+    },
+    {
+      "entropy": 0.8096660938113928,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.1254287958145142,
+      "learning_rate": 8.700603238980251e-05,
+      "loss": 0.7170331001281738,
+      "mean_token_accuracy": 0.7994555257260799,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.792780208788561,
+      "eval_loss": 0.7442984580993652,
+      "eval_mean_token_accuracy": 0.7954704573984896,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 90.294,
+      "eval_samples_per_second": 15.726,
+      "eval_steps_per_second": 1.971,
+      "step": 100
+    },
+    {
+      "entropy": 0.7979829199612141,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2140507698059082,
+      "learning_rate": 0.00010458300863016666,
+      "loss": 0.7141861915588379,
+      "mean_token_accuracy": 0.8000014744699001,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.7938692486018277,
+      "eval_loss": 0.7279797196388245,
+      "eval_mean_token_accuracy": 0.7989798734027348,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 89.7189,
+      "eval_samples_per_second": 15.827,
+      "eval_steps_per_second": 1.984,
+      "step": 120
+    },
+    {
+      "entropy": 0.7916347607970238,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.0506458282470703,
+      "learning_rate": 0.0001221599848705308,
+      "loss": 0.7117055416107178,
+      "mean_token_accuracy": 0.7970046654343605,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.8052781063519167,
+      "eval_loss": 0.722576916217804,
+      "eval_mean_token_accuracy": 0.7982530523551984,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 90.0376,
+      "eval_samples_per_second": 15.771,
+      "eval_steps_per_second": 1.977,
+      "step": 140
+    },
+    {
+      "entropy": 0.7880666613578796,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8409207463264465,
+      "learning_rate": 0.00013973696111089492,
+      "loss": 0.7052523136138916,
+      "mean_token_accuracy": 0.8000123649835587,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7787443134891853,
+      "eval_loss": 0.7137772440910339,
+      "eval_mean_token_accuracy": 0.8014209287220173,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 90.2562,
+      "eval_samples_per_second": 15.733,
+      "eval_steps_per_second": 1.972,
+      "step": 160
+    },
+    {
+      "entropy": 0.7696007996797561,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.0182477235794067,
+      "learning_rate": 0.00015731393735125907,
+      "loss": 0.7008797645568847,
+      "mean_token_accuracy": 0.8007259473204613,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.752259682068664,
+      "eval_loss": 0.7171286344528198,
+      "eval_mean_token_accuracy": 0.80052122574174,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 90.0797,
+      "eval_samples_per_second": 15.764,
+      "eval_steps_per_second": 1.976,
+      "step": 180
+    },
+    {
+      "entropy": 0.7828126326203346,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9553632736206055,
+      "learning_rate": 0.0001748909135916232,
+      "loss": 0.7046853542327881,
+      "mean_token_accuracy": 0.7988562889397144,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7389638169427936,
+      "eval_loss": 0.7187935709953308,
+      "eval_mean_token_accuracy": 0.7995899999409579,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 89.9223,
+      "eval_samples_per_second": 15.791,
+      "eval_steps_per_second": 1.979,
+      "step": 200
+    },
+    {
+      "entropy": 0.7909286297857762,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 1.134459376335144,
+      "learning_rate": 0.00019246788983198735,
+      "loss": 0.6989312171936035,
+      "mean_token_accuracy": 0.7991565138101577,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7496965199374082,
+      "eval_loss": 0.7106321454048157,
+      "eval_mean_token_accuracy": 0.8046183442131857,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 89.6899,
+      "eval_samples_per_second": 15.832,
+      "eval_steps_per_second": 1.985,
+      "step": 220
+    },
+    {
+      "entropy": 0.7651963323354721,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.9557585716247559,
+      "learning_rate": 0.0002100448660723515,
+      "loss": 0.6984559059143066,
+      "mean_token_accuracy": 0.8019823372364044,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7615563079212488,
+      "eval_loss": 0.7098539471626282,
+      "eval_mean_token_accuracy": 0.8047233508544022,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 90.3752,
+      "eval_samples_per_second": 15.712,
+      "eval_steps_per_second": 1.97,
+      "step": 240
+    },
+    {
+      "entropy": 0.787507726252079,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.9279311299324036,
+      "learning_rate": 0.00022762184231271568,
+      "loss": 0.708138370513916,
+      "mean_token_accuracy": 0.8013308703899383,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7856640896100676,
+      "eval_loss": 0.7191570401191711,
+      "eval_mean_token_accuracy": 0.8024677249153008,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 93.7336,
+      "eval_samples_per_second": 15.149,
+      "eval_steps_per_second": 1.899,
+      "step": 260
+    },
+    {
+      "entropy": 0.768489234894514,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.7992776036262512,
+      "learning_rate": 0.0002451988185530798,
+      "loss": 0.7035849571228028,
+      "mean_token_accuracy": 0.8031105332076549,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7682498740346244,
+      "eval_loss": 0.7149022221565247,
+      "eval_mean_token_accuracy": 0.7973497995499814,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 89.7309,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 280
+    },
+    {
+      "entropy": 0.7720955617725849,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.1931514739990234,
+      "learning_rate": 0.0002627757947934439,
+      "loss": 0.6965402603149414,
+      "mean_token_accuracy": 0.8040348663926125,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7214708304807042,
+      "eval_loss": 0.7350823879241943,
+      "eval_mean_token_accuracy": 0.7942911925610532,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 89.6451,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 300
+    },
+    {
+      "entropy": 0.7930095426738262,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8994467258453369,
+      "learning_rate": 0.0002803527710338081,
+      "loss": 0.7331492900848389,
+      "mean_token_accuracy": 0.796185302734375,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.8142535488926963,
+      "eval_loss": 0.7237834334373474,
+      "eval_mean_token_accuracy": 0.799495273761535,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 90.6759,
+      "eval_samples_per_second": 15.66,
+      "eval_steps_per_second": 1.963,
+      "step": 320
+    },
+    {
+      "entropy": 0.7825102441012859,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 1.8224239349365234,
+      "learning_rate": 0.00029792974727417223,
+      "loss": 0.7331175804138184,
+      "mean_token_accuracy": 0.7977667152881622,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7327135443017724,
+      "eval_loss": 0.7388784289360046,
+      "eval_mean_token_accuracy": 0.8007041252730938,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 89.7692,
+      "eval_samples_per_second": 15.818,
+      "eval_steps_per_second": 1.983,
+      "step": 340
+    },
+    {
+      "entropy": 0.819331557303667,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 1.0407779216766357,
+      "learning_rate": 0.0003155067235145364,
+      "loss": 0.7456695556640625,
+      "mean_token_accuracy": 0.7936980701982975,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.8065585420372781,
+      "eval_loss": 0.735519528388977,
+      "eval_mean_token_accuracy": 0.7974206296245704,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 89.4336,
+      "eval_samples_per_second": 15.878,
+      "eval_steps_per_second": 1.99,
+      "step": 360
+    },
+    {
+      "entropy": 0.8114653021097183,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.1209490299224854,
+      "learning_rate": 0.0003330836997549005,
+      "loss": 0.7340899467468261,
+      "mean_token_accuracy": 0.7914272703230381,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.8025527285056168,
+      "eval_loss": 0.7295576930046082,
+      "eval_mean_token_accuracy": 0.7990576628218876,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 89.9579,
+      "eval_samples_per_second": 15.785,
+      "eval_steps_per_second": 1.979,
+      "step": 380
+    },
+    {
+      "entropy": 0.8042497783899307,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 1.2437798976898193,
+      "learning_rate": 0.0003506606759952647,
+      "loss": 0.7381744384765625,
+      "mean_token_accuracy": 0.7967362694442273,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7717011336530193,
+      "eval_loss": 0.7436173558235168,
+      "eval_mean_token_accuracy": 0.7975849482450592,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 89.8411,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 400
+    },
+    {
+      "entropy": 0.7987750010056929,
+      "epoch": 1.012070006035003,
+      "grad_norm": 1.2390918731689453,
+      "learning_rate": 0.000364721224843344,
+      "loss": 0.7317790031433106,
+      "mean_token_accuracy": 0.7986087551364651,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7333170604170038,
+      "eval_loss": 0.7456948757171631,
+      "eval_mean_token_accuracy": 0.7964897577682238,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 89.7288,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 420
+    },
+    {
+      "entropy": 0.7372976846992969,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.1552740335464478,
+      "learning_rate": 0.00036468510102269575,
+      "loss": 0.6734249114990234,
+      "mean_token_accuracy": 0.805834824591875,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.7130355526891987,
+      "eval_loss": 0.7555333375930786,
+      "eval_mean_token_accuracy": 0.7961652781186479,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 89.8713,
+      "eval_samples_per_second": 15.8,
+      "eval_steps_per_second": 1.981,
+      "step": 440
+    },
+    {
+      "entropy": 0.7465668715536594,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 1.1991469860076904,
+      "learning_rate": 0.0003645973816745066,
+      "loss": 0.6871879577636719,
+      "mean_token_accuracy": 0.8042450234293937,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7772979542110743,
+      "eval_loss": 0.7561826705932617,
+      "eval_mean_token_accuracy": 0.7953876427720102,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 89.8395,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 460
+    },
+    {
+      "entropy": 0.7477855801582336,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 1.325486660003662,
+      "learning_rate": 0.00036445809162231435,
+      "loss": 0.6836830139160156,
+      "mean_token_accuracy": 0.8021619468927383,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.7098110676481483,
+      "eval_loss": 0.758007287979126,
+      "eval_mean_token_accuracy": 0.7909953189030122,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 89.6101,
+      "eval_samples_per_second": 15.846,
+      "eval_steps_per_second": 1.986,
+      "step": 480
+    },
+    {
+      "entropy": 0.7632691070437432,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 1.0122262239456177,
+      "learning_rate": 0.0003642672702835562,
+      "loss": 0.7071213245391845,
+      "mean_token_accuracy": 0.8000680930912495,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7645382720432924,
+      "eval_loss": 0.7492606043815613,
+      "eval_mean_token_accuracy": 0.7976858676149604,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 89.6092,
+      "eval_samples_per_second": 15.847,
+      "eval_steps_per_second": 1.986,
+      "step": 500
+    },
+    {
+      "entropy": 0.7623726457357407,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.367774486541748,
+      "learning_rate": 0.00036402497165841384,
+      "loss": 0.6965175151824952,
+      "mean_token_accuracy": 0.8016501650214195,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7868083715438843,
+      "eval_loss": 0.7481449842453003,
+      "eval_mean_token_accuracy": 0.799330582779445,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 89.8241,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 520
+    },
+    {
+      "entropy": 0.7535207405686378,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 1.213614821434021,
+      "learning_rate": 0.00036373126431453207,
+      "loss": 0.7078310489654541,
+      "mean_token_accuracy": 0.7998907208442688,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7829091950748743,
+      "eval_loss": 0.7416072487831116,
+      "eval_mean_token_accuracy": 0.7969146250339036,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 89.4607,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 540
+    },
+    {
+      "entropy": 0.7527363449335098,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 1.0929982662200928,
+      "learning_rate": 0.00036338623136761495,
+      "loss": 0.7058369636535644,
+      "mean_token_accuracy": 0.7997087433934211,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.728419938114252,
+      "eval_loss": 0.744873046875,
+      "eval_mean_token_accuracy": 0.7966634296299366,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 89.6597,
+      "eval_samples_per_second": 15.838,
+      "eval_steps_per_second": 1.985,
+      "step": 560
+    },
+    {
+      "entropy": 0.7406099259853363,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.409364104270935,
+      "learning_rate": 0.00036298997045790515,
+      "loss": 0.6902852058410645,
+      "mean_token_accuracy": 0.805386807769537,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.6991632758231645,
+      "eval_loss": 0.7480319738388062,
+      "eval_mean_token_accuracy": 0.7970068006033308,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 89.8329,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 580
+    },
+    {
+      "entropy": 0.7459449704736472,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.7053111791610718,
+      "learning_rate": 0.00036254259372255277,
+      "loss": 0.7044640064239502,
+      "mean_token_accuracy": 0.8005018755793571,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7454566660891758,
+      "eval_loss": 0.7551484704017639,
+      "eval_mean_token_accuracy": 0.7897093731365846,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 89.8313,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 600
+    },
+    {
+      "entropy": 0.7621391348540782,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 1.174443244934082,
+      "learning_rate": 0.000362044227763882,
+      "loss": 0.7111573696136475,
+      "mean_token_accuracy": 0.798752411454916,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7664387949397055,
+      "eval_loss": 0.7365804314613342,
+      "eval_mean_token_accuracy": 0.7963101200843126,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 89.9623,
+      "eval_samples_per_second": 15.784,
+      "eval_steps_per_second": 1.979,
+      "step": 620
+    },
+    {
+      "entropy": 0.7424944408237935,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.2898184061050415,
+      "learning_rate": 0.000361495013613564,
+      "loss": 0.6913110733032226,
+      "mean_token_accuracy": 0.8009015507996082,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7255479529332579,
+      "eval_loss": 0.7345991730690002,
+      "eval_mean_token_accuracy": 0.8013592945056015,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 89.817,
+      "eval_samples_per_second": 15.81,
+      "eval_steps_per_second": 1.982,
+      "step": 640
+    },
+    {
+      "entropy": 0.7303367637097835,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.1310476064682007,
+      "learning_rate": 0.00036089510669270666,
+      "loss": 0.6875617027282714,
+      "mean_token_accuracy": 0.8035397171974182,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7101308248016271,
+      "eval_loss": 0.7394917011260986,
+      "eval_mean_token_accuracy": 0.7990792942850777,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 89.7431,
+      "eval_samples_per_second": 15.823,
+      "eval_steps_per_second": 1.983,
+      "step": 660
+    },
+    {
+      "entropy": 0.7387678384780884,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.9645389914512634,
+      "learning_rate": 0.0003602446767678725,
+      "loss": 0.6911204338073731,
+      "mean_token_accuracy": 0.8056333385407924,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7281073737010527,
+      "eval_loss": 0.7376708984375,
+      "eval_mean_token_accuracy": 0.7997356415464637,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 90.0338,
+      "eval_samples_per_second": 15.772,
+      "eval_steps_per_second": 1.977,
+      "step": 680
+    },
+    {
+      "entropy": 0.7449554048478604,
+      "epoch": 1.687990343995172,
+      "grad_norm": 1.9069629907608032,
+      "learning_rate": 0.0003595439079030364,
+      "loss": 0.7042286396026611,
+      "mean_token_accuracy": 0.8018639378249646,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7022633817088738,
+      "eval_loss": 0.7320939898490906,
+      "eval_mean_token_accuracy": 0.8006051541044471,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 89.8948,
+      "eval_samples_per_second": 15.796,
+      "eval_steps_per_second": 1.98,
+      "step": 700
+    },
+    {
+      "entropy": 0.723249051719904,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.8333877325057983,
+      "learning_rate": 0.00035879299840749777,
+      "loss": 0.7043714046478271,
+      "mean_token_accuracy": 0.8028345607221127,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.672436946898364,
+      "eval_loss": 0.713233470916748,
+      "eval_mean_token_accuracy": 0.8047760577684038,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 89.6394,
+      "eval_samples_per_second": 15.841,
+      "eval_steps_per_second": 1.986,
+      "step": 720
+    },
+    {
+      "entropy": 0.730014206469059,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.4580037593841553,
+      "learning_rate": 0.00035799216077976137,
+      "loss": 0.6985176563262939,
+      "mean_token_accuracy": 0.8015547141432762,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.6908873795123582,
+      "eval_loss": 0.7198202013969421,
+      "eval_mean_token_accuracy": 0.8024131682481659,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 89.3771,
+      "eval_samples_per_second": 15.888,
+      "eval_steps_per_second": 1.992,
+      "step": 740
+    },
+    {
+      "entropy": 0.7219605796039105,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.51682710647583,
+      "learning_rate": 0.000357141621647403,
+      "loss": 0.6844424724578857,
+      "mean_token_accuracy": 0.8032813094556331,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7113401109582922,
+      "eval_loss": 0.7195309400558472,
+      "eval_mean_token_accuracy": 0.8019187035185568,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 89.4658,
+      "eval_samples_per_second": 15.872,
+      "eval_steps_per_second": 1.99,
+      "step": 760
+    },
+    {
+      "entropy": 0.729826743900776,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 1.0838000774383545,
+      "learning_rate": 0.0003562416217029361,
+      "loss": 0.685858678817749,
+      "mean_token_accuracy": 0.8050567395985126,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.649370267484965,
+      "eval_loss": 0.7177029252052307,
+      "eval_mean_token_accuracy": 0.8050174043419656,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 89.2287,
+      "eval_samples_per_second": 15.914,
+      "eval_steps_per_second": 1.995,
+      "step": 780
+    },
+    {
+      "entropy": 0.6799364153295755,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 1.1364924907684326,
+      "learning_rate": 0.00035529241563569903,
+      "loss": 0.6714428901672364,
+      "mean_token_accuracy": 0.8087341658771038,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6603799047095052,
+      "eval_loss": 0.7201809287071228,
+      "eval_mean_token_accuracy": 0.802863868769635,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 89.2718,
+      "eval_samples_per_second": 15.906,
+      "eval_steps_per_second": 1.994,
+      "step": 800
+    },
+    {
+      "entropy": 0.7078610375523567,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 1.2191942930221558,
+      "learning_rate": 0.0003542942720597808,
+      "loss": 0.6996590614318847,
+      "mean_token_accuracy": 0.8049512051045895,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.7240545264120852,
+      "eval_loss": 0.7301309108734131,
+      "eval_mean_token_accuracy": 0.8015973166133581,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 89.3438,
+      "eval_samples_per_second": 15.894,
+      "eval_steps_per_second": 1.992,
+      "step": 820
+    },
+    {
+      "entropy": 0.6557391556826505,
+      "epoch": 2.024140012070006,
+      "grad_norm": 1.36203134059906,
+      "learning_rate": 0.0003532474734380064,
+      "loss": 0.6327113628387451,
+      "mean_token_accuracy": 0.8142829847026181,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6573808648613062,
+      "eval_loss": 0.7334365844726562,
+      "eval_mean_token_accuracy": 0.8043633893634496,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 89.3359,
+      "eval_samples_per_second": 15.895,
+      "eval_steps_per_second": 1.992,
+      "step": 840
+    },
+    {
+      "entropy": 0.6409128502011299,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 1.5804955959320068,
+      "learning_rate": 0.0003521523160020035,
+      "loss": 0.6023559093475341,
+      "mean_token_accuracy": 0.8222251623868942,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6983234788594621,
+      "eval_loss": 0.7232135534286499,
+      "eval_mean_token_accuracy": 0.8010654268639811,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 89.2615,
+      "eval_samples_per_second": 15.908,
+      "eval_steps_per_second": 1.994,
+      "step": 860
+    },
+    {
+      "entropy": 0.6151149723678827,
+      "epoch": 2.12070006035003,
+      "grad_norm": 1.1975436210632324,
+      "learning_rate": 0.00035100910966837193,
+      "loss": 0.5803564548492431,
+      "mean_token_accuracy": 0.8256325736641884,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6636555573243773,
+      "eval_loss": 0.7205662727355957,
+      "eval_mean_token_accuracy": 0.8036768369460374,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 89.2957,
+      "eval_samples_per_second": 15.902,
+      "eval_steps_per_second": 1.993,
+      "step": 880
+    },
+    {
+      "entropy": 0.6201727617532015,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9907131195068359,
+      "learning_rate": 0.0003498181779509813,
+      "loss": 0.6101872444152832,
+      "mean_token_accuracy": 0.8196512959897518,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.6473788086617931,
+      "eval_loss": 0.7062397003173828,
+      "eval_mean_token_accuracy": 0.8078362184963869,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 88.6198,
+      "eval_samples_per_second": 16.024,
+      "eval_steps_per_second": 2.009,
+      "step": 900
+    },
+    {
+      "entropy": 0.6097332935780286,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.9055633544921875,
+      "learning_rate": 0.00034857985786942026,
+      "loss": 0.5956094264984131,
+      "mean_token_accuracy": 0.8240575887262821,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.612629868341296,
+      "eval_loss": 0.7106770277023315,
+      "eval_mean_token_accuracy": 0.805324277181304,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 89.4383,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 920
+    },
+    {
+      "entropy": 0.6182701248675585,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.9513736367225647,
+      "learning_rate": 0.00034729449985362404,
+      "loss": 0.6019269466400147,
+      "mean_token_accuracy": 0.8187106013298034,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6437647456533453,
+      "eval_loss": 0.7173502445220947,
+      "eval_mean_token_accuracy": 0.8056439573175451,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.9568,
+      "eval_samples_per_second": 15.963,
+      "eval_steps_per_second": 2.001,
+      "step": 940
+    },
+    {
+      "entropy": 0.6508570514619351,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.9606207013130188,
+      "learning_rate": 0.0003459624676447067,
+      "loss": 0.6296409606933594,
+      "mean_token_accuracy": 0.8167732007801533,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6288247135248077,
+      "eval_loss": 0.7160675525665283,
+      "eval_mean_token_accuracy": 0.8066010689467527,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 89.0044,
+      "eval_samples_per_second": 15.954,
+      "eval_steps_per_second": 2.0,
+      "step": 960
+    },
+    {
+      "entropy": 0.6380081418901682,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.9677181839942932,
+      "learning_rate": 0.000344584138192027,
+      "loss": 0.6210546493530273,
+      "mean_token_accuracy": 0.8186135642230511,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6407805031604981,
+      "eval_loss": 0.7055043578147888,
+      "eval_mean_token_accuracy": 0.8070103157772107,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 89.5606,
+      "eval_samples_per_second": 15.855,
+      "eval_steps_per_second": 1.987,
+      "step": 980
+    },
+    {
+      "entropy": 0.6310251638293266,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 2.1046407222747803,
+      "learning_rate": 0.000343159901546516,
+      "loss": 0.6165830135345459,
+      "mean_token_accuracy": 0.8207855716347694,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6461507470420237,
+      "eval_loss": 0.7292136549949646,
+      "eval_mean_token_accuracy": 0.8030418261383356,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 89.315,
+      "eval_samples_per_second": 15.899,
+      "eval_steps_per_second": 1.993,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6256617344915867,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 1.1768349409103394,
+      "learning_rate": 0.0003416901607502972,
+      "loss": 0.6142709255218506,
+      "mean_token_accuracy": 0.8220669947564602,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6150523993406403,
+      "eval_loss": 0.7019563317298889,
+      "eval_mean_token_accuracy": 0.8094323099998946,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 89.419,
+      "eval_samples_per_second": 15.88,
+      "eval_steps_per_second": 1.991,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6126735735684633,
+      "epoch": 2.506940253470127,
+      "grad_norm": 1.8065496683120728,
+      "learning_rate": 0.00034017533172263055,
+      "loss": 0.6056458473205566,
+      "mean_token_accuracy": 0.8203762136399746,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6544345456562685,
+      "eval_loss": 0.7086517214775085,
+      "eval_mean_token_accuracy": 0.806709805901131,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 88.981,
+      "eval_samples_per_second": 15.958,
+      "eval_steps_per_second": 2.0,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6311754353344441,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 1.0576369762420654,
+      "learning_rate": 0.00033861584314221236,
+      "loss": 0.6176392555236816,
+      "mean_token_accuracy": 0.8177410490810871,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6273799200406235,
+      "eval_loss": 0.708163321018219,
+      "eval_mean_token_accuracy": 0.8032017099053672,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 89.7668,
+      "eval_samples_per_second": 15.819,
+      "eval_steps_per_second": 1.983,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6200249589979648,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.997065007686615,
+      "learning_rate": 0.0003370121363258637,
+      "loss": 0.6203888893127442,
+      "mean_token_accuracy": 0.8196275025606156,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6457576811983344,
+      "eval_loss": 0.6972290277481079,
+      "eval_mean_token_accuracy": 0.8095980849158898,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 89.2027,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 1080
+    },
+    {
+      "entropy": 0.626810473203659,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.865475058555603,
+      "learning_rate": 0.0003353646651036438,
+      "loss": 0.6102585792541504,
+      "mean_token_accuracy": 0.819708751142025,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.6412022431914726,
+      "eval_loss": 0.7036139965057373,
+      "eval_mean_token_accuracy": 0.8070317135098275,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 89.4733,
+      "eval_samples_per_second": 15.871,
+      "eval_steps_per_second": 1.989,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6581619590520859,
+      "epoch": 2.700060350030175,
+      "grad_norm": 1.00307035446167,
+      "learning_rate": 0.00033367389569042064,
+      "loss": 0.6397720336914062,
+      "mean_token_accuracy": 0.81413309648633,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6288736402318719,
+      "eval_loss": 0.6966825723648071,
+      "eval_mean_token_accuracy": 0.8094039669867312,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 88.7341,
+      "eval_samples_per_second": 16.003,
+      "eval_steps_per_second": 2.006,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6399412982165813,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.8589329719543457,
+      "learning_rate": 0.0003319403065539384,
+      "loss": 0.6347338199615479,
+      "mean_token_accuracy": 0.8139534957706929,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.683491913455256,
+      "eval_loss": 0.6984149217605591,
+      "eval_mean_token_accuracy": 0.8083989469522841,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 89.3929,
+      "eval_samples_per_second": 15.885,
+      "eval_steps_per_second": 1.991,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6264835461974144,
+      "epoch": 2.796620398310199,
+      "grad_norm": 0.7807853817939758,
+      "learning_rate": 0.0003301643882794163,
+      "loss": 0.6162629127502441,
+      "mean_token_accuracy": 0.8216292977333068,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6578792159476977,
+      "eval_loss": 0.6921441555023193,
+      "eval_mean_token_accuracy": 0.8094299177775223,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 89.2356,
+      "eval_samples_per_second": 15.913,
+      "eval_steps_per_second": 1.995,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6269129011780024,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 1.1675593852996826,
+      "learning_rate": 0.0003283466434307189,
+      "loss": 0.6235532760620117,
+      "mean_token_accuracy": 0.8188897788524627,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6533248447970058,
+      "eval_loss": 0.696849524974823,
+      "eval_mean_token_accuracy": 0.8102703713968898,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2602,
+      "eval_samples_per_second": 16.089,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.6259873781353236,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.9596486687660217,
+      "learning_rate": 0.00032648758640813655,
+      "loss": 0.6118728160858155,
+      "mean_token_accuracy": 0.8171859130263328,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6212078754821521,
+      "eval_loss": 0.6942870020866394,
+      "eval_mean_token_accuracy": 0.810484343365337,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 89.4375,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 1200
+    },
+    {
+      "entropy": 0.6142537571489811,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 0.9625058770179749,
+      "learning_rate": 0.00032458774330281615,
+      "loss": 0.6060261249542236,
+      "mean_token_accuracy": 0.8223113484680653,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6553665667437436,
+      "eval_loss": 0.6835269331932068,
+      "eval_mean_token_accuracy": 0.8134723968720168,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 89.5092,
+      "eval_samples_per_second": 15.864,
+      "eval_steps_per_second": 1.989,
+      "step": 1220
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.22275795454208e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1240/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1240/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1240/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1240/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1240/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..7ce4b072fcbe5b636867e13937c56fedee3b68b4
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1240/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.02728394274790723,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1240/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1240/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1240/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1240/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1240/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1240/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1240/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1240/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..0ba6d4f23ebf4c7f726c8e6fb64e39b4db05eec9
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1240/trainer_state.json
@@ -0,0 +1,1336 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 2.9897404948702473,
+  "eval_steps": 20,
+  "global_step": 1240,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.024795526266098,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.945375919342041,
+      "learning_rate": 1.6698127428345936e-05,
+      "loss": 1.7825420379638672,
+      "mean_token_accuracy": 0.6315708436071873,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.29921497588747,
+      "eval_loss": 1.169131875038147,
+      "eval_mean_token_accuracy": 0.7185706342204233,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 91.7898,
+      "eval_samples_per_second": 15.47,
+      "eval_steps_per_second": 1.939,
+      "step": 20
+    },
+    {
+      "entropy": 1.0346344470977784,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.7916979789733887,
+      "learning_rate": 3.427510366871008e-05,
+      "loss": 0.9480395317077637,
+      "mean_token_accuracy": 0.7543429024517536,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 0.9673695430326997,
+      "eval_loss": 0.8588430285453796,
+      "eval_mean_token_accuracy": 0.7737233002534073,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 89.8219,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 40
+    },
+    {
+      "entropy": 0.8977566614747048,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7972187995910645,
+      "learning_rate": 5.185207990907422e-05,
+      "loss": 0.8029604911804199,
+      "mean_token_accuracy": 0.7818016350269318,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8436836982041263,
+      "eval_loss": 0.7893033623695374,
+      "eval_mean_token_accuracy": 0.7858914984076211,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 89.9349,
+      "eval_samples_per_second": 15.789,
+      "eval_steps_per_second": 1.979,
+      "step": 60
+    },
+    {
+      "entropy": 0.8650321319699288,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.2128729820251465,
+      "learning_rate": 6.942905614943836e-05,
+      "loss": 0.77417893409729,
+      "mean_token_accuracy": 0.7875479467213153,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8209857180547179,
+      "eval_loss": 0.7572280168533325,
+      "eval_mean_token_accuracy": 0.7925970383574453,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 89.9459,
+      "eval_samples_per_second": 15.787,
+      "eval_steps_per_second": 1.979,
+      "step": 80
+    },
+    {
+      "entropy": 0.8096660938113928,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.1254287958145142,
+      "learning_rate": 8.700603238980251e-05,
+      "loss": 0.7170331001281738,
+      "mean_token_accuracy": 0.7994555257260799,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.792780208788561,
+      "eval_loss": 0.7442984580993652,
+      "eval_mean_token_accuracy": 0.7954704573984896,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 90.294,
+      "eval_samples_per_second": 15.726,
+      "eval_steps_per_second": 1.971,
+      "step": 100
+    },
+    {
+      "entropy": 0.7979829199612141,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2140507698059082,
+      "learning_rate": 0.00010458300863016666,
+      "loss": 0.7141861915588379,
+      "mean_token_accuracy": 0.8000014744699001,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.7938692486018277,
+      "eval_loss": 0.7279797196388245,
+      "eval_mean_token_accuracy": 0.7989798734027348,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 89.7189,
+      "eval_samples_per_second": 15.827,
+      "eval_steps_per_second": 1.984,
+      "step": 120
+    },
+    {
+      "entropy": 0.7916347607970238,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.0506458282470703,
+      "learning_rate": 0.0001221599848705308,
+      "loss": 0.7117055416107178,
+      "mean_token_accuracy": 0.7970046654343605,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.8052781063519167,
+      "eval_loss": 0.722576916217804,
+      "eval_mean_token_accuracy": 0.7982530523551984,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 90.0376,
+      "eval_samples_per_second": 15.771,
+      "eval_steps_per_second": 1.977,
+      "step": 140
+    },
+    {
+      "entropy": 0.7880666613578796,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8409207463264465,
+      "learning_rate": 0.00013973696111089492,
+      "loss": 0.7052523136138916,
+      "mean_token_accuracy": 0.8000123649835587,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7787443134891853,
+      "eval_loss": 0.7137772440910339,
+      "eval_mean_token_accuracy": 0.8014209287220173,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 90.2562,
+      "eval_samples_per_second": 15.733,
+      "eval_steps_per_second": 1.972,
+      "step": 160
+    },
+    {
+      "entropy": 0.7696007996797561,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.0182477235794067,
+      "learning_rate": 0.00015731393735125907,
+      "loss": 0.7008797645568847,
+      "mean_token_accuracy": 0.8007259473204613,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.752259682068664,
+      "eval_loss": 0.7171286344528198,
+      "eval_mean_token_accuracy": 0.80052122574174,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 90.0797,
+      "eval_samples_per_second": 15.764,
+      "eval_steps_per_second": 1.976,
+      "step": 180
+    },
+    {
+      "entropy": 0.7828126326203346,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9553632736206055,
+      "learning_rate": 0.0001748909135916232,
+      "loss": 0.7046853542327881,
+      "mean_token_accuracy": 0.7988562889397144,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7389638169427936,
+      "eval_loss": 0.7187935709953308,
+      "eval_mean_token_accuracy": 0.7995899999409579,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 89.9223,
+      "eval_samples_per_second": 15.791,
+      "eval_steps_per_second": 1.979,
+      "step": 200
+    },
+    {
+      "entropy": 0.7909286297857762,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 1.134459376335144,
+      "learning_rate": 0.00019246788983198735,
+      "loss": 0.6989312171936035,
+      "mean_token_accuracy": 0.7991565138101577,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7496965199374082,
+      "eval_loss": 0.7106321454048157,
+      "eval_mean_token_accuracy": 0.8046183442131857,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 89.6899,
+      "eval_samples_per_second": 15.832,
+      "eval_steps_per_second": 1.985,
+      "step": 220
+    },
+    {
+      "entropy": 0.7651963323354721,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.9557585716247559,
+      "learning_rate": 0.0002100448660723515,
+      "loss": 0.6984559059143066,
+      "mean_token_accuracy": 0.8019823372364044,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7615563079212488,
+      "eval_loss": 0.7098539471626282,
+      "eval_mean_token_accuracy": 0.8047233508544022,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 90.3752,
+      "eval_samples_per_second": 15.712,
+      "eval_steps_per_second": 1.97,
+      "step": 240
+    },
+    {
+      "entropy": 0.787507726252079,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.9279311299324036,
+      "learning_rate": 0.00022762184231271568,
+      "loss": 0.708138370513916,
+      "mean_token_accuracy": 0.8013308703899383,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7856640896100676,
+      "eval_loss": 0.7191570401191711,
+      "eval_mean_token_accuracy": 0.8024677249153008,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 93.7336,
+      "eval_samples_per_second": 15.149,
+      "eval_steps_per_second": 1.899,
+      "step": 260
+    },
+    {
+      "entropy": 0.768489234894514,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.7992776036262512,
+      "learning_rate": 0.0002451988185530798,
+      "loss": 0.7035849571228028,
+      "mean_token_accuracy": 0.8031105332076549,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7682498740346244,
+      "eval_loss": 0.7149022221565247,
+      "eval_mean_token_accuracy": 0.7973497995499814,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 89.7309,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 280
+    },
+    {
+      "entropy": 0.7720955617725849,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.1931514739990234,
+      "learning_rate": 0.0002627757947934439,
+      "loss": 0.6965402603149414,
+      "mean_token_accuracy": 0.8040348663926125,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7214708304807042,
+      "eval_loss": 0.7350823879241943,
+      "eval_mean_token_accuracy": 0.7942911925610532,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 89.6451,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 300
+    },
+    {
+      "entropy": 0.7930095426738262,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8994467258453369,
+      "learning_rate": 0.0002803527710338081,
+      "loss": 0.7331492900848389,
+      "mean_token_accuracy": 0.796185302734375,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.8142535488926963,
+      "eval_loss": 0.7237834334373474,
+      "eval_mean_token_accuracy": 0.799495273761535,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 90.6759,
+      "eval_samples_per_second": 15.66,
+      "eval_steps_per_second": 1.963,
+      "step": 320
+    },
+    {
+      "entropy": 0.7825102441012859,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 1.8224239349365234,
+      "learning_rate": 0.00029792974727417223,
+      "loss": 0.7331175804138184,
+      "mean_token_accuracy": 0.7977667152881622,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7327135443017724,
+      "eval_loss": 0.7388784289360046,
+      "eval_mean_token_accuracy": 0.8007041252730938,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 89.7692,
+      "eval_samples_per_second": 15.818,
+      "eval_steps_per_second": 1.983,
+      "step": 340
+    },
+    {
+      "entropy": 0.819331557303667,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 1.0407779216766357,
+      "learning_rate": 0.0003155067235145364,
+      "loss": 0.7456695556640625,
+      "mean_token_accuracy": 0.7936980701982975,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.8065585420372781,
+      "eval_loss": 0.735519528388977,
+      "eval_mean_token_accuracy": 0.7974206296245704,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 89.4336,
+      "eval_samples_per_second": 15.878,
+      "eval_steps_per_second": 1.99,
+      "step": 360
+    },
+    {
+      "entropy": 0.8114653021097183,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.1209490299224854,
+      "learning_rate": 0.0003330836997549005,
+      "loss": 0.7340899467468261,
+      "mean_token_accuracy": 0.7914272703230381,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.8025527285056168,
+      "eval_loss": 0.7295576930046082,
+      "eval_mean_token_accuracy": 0.7990576628218876,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 89.9579,
+      "eval_samples_per_second": 15.785,
+      "eval_steps_per_second": 1.979,
+      "step": 380
+    },
+    {
+      "entropy": 0.8042497783899307,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 1.2437798976898193,
+      "learning_rate": 0.0003506606759952647,
+      "loss": 0.7381744384765625,
+      "mean_token_accuracy": 0.7967362694442273,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7717011336530193,
+      "eval_loss": 0.7436173558235168,
+      "eval_mean_token_accuracy": 0.7975849482450592,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 89.8411,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 400
+    },
+    {
+      "entropy": 0.7987750010056929,
+      "epoch": 1.012070006035003,
+      "grad_norm": 1.2390918731689453,
+      "learning_rate": 0.000364721224843344,
+      "loss": 0.7317790031433106,
+      "mean_token_accuracy": 0.7986087551364651,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7333170604170038,
+      "eval_loss": 0.7456948757171631,
+      "eval_mean_token_accuracy": 0.7964897577682238,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 89.7288,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 420
+    },
+    {
+      "entropy": 0.7372976846992969,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.1552740335464478,
+      "learning_rate": 0.00036468510102269575,
+      "loss": 0.6734249114990234,
+      "mean_token_accuracy": 0.805834824591875,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.7130355526891987,
+      "eval_loss": 0.7555333375930786,
+      "eval_mean_token_accuracy": 0.7961652781186479,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 89.8713,
+      "eval_samples_per_second": 15.8,
+      "eval_steps_per_second": 1.981,
+      "step": 440
+    },
+    {
+      "entropy": 0.7465668715536594,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 1.1991469860076904,
+      "learning_rate": 0.0003645973816745066,
+      "loss": 0.6871879577636719,
+      "mean_token_accuracy": 0.8042450234293937,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7772979542110743,
+      "eval_loss": 0.7561826705932617,
+      "eval_mean_token_accuracy": 0.7953876427720102,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 89.8395,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 460
+    },
+    {
+      "entropy": 0.7477855801582336,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 1.325486660003662,
+      "learning_rate": 0.00036445809162231435,
+      "loss": 0.6836830139160156,
+      "mean_token_accuracy": 0.8021619468927383,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.7098110676481483,
+      "eval_loss": 0.758007287979126,
+      "eval_mean_token_accuracy": 0.7909953189030122,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 89.6101,
+      "eval_samples_per_second": 15.846,
+      "eval_steps_per_second": 1.986,
+      "step": 480
+    },
+    {
+      "entropy": 0.7632691070437432,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 1.0122262239456177,
+      "learning_rate": 0.0003642672702835562,
+      "loss": 0.7071213245391845,
+      "mean_token_accuracy": 0.8000680930912495,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7645382720432924,
+      "eval_loss": 0.7492606043815613,
+      "eval_mean_token_accuracy": 0.7976858676149604,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 89.6092,
+      "eval_samples_per_second": 15.847,
+      "eval_steps_per_second": 1.986,
+      "step": 500
+    },
+    {
+      "entropy": 0.7623726457357407,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.367774486541748,
+      "learning_rate": 0.00036402497165841384,
+      "loss": 0.6965175151824952,
+      "mean_token_accuracy": 0.8016501650214195,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7868083715438843,
+      "eval_loss": 0.7481449842453003,
+      "eval_mean_token_accuracy": 0.799330582779445,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 89.8241,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 520
+    },
+    {
+      "entropy": 0.7535207405686378,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 1.213614821434021,
+      "learning_rate": 0.00036373126431453207,
+      "loss": 0.7078310489654541,
+      "mean_token_accuracy": 0.7998907208442688,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7829091950748743,
+      "eval_loss": 0.7416072487831116,
+      "eval_mean_token_accuracy": 0.7969146250339036,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 89.4607,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 540
+    },
+    {
+      "entropy": 0.7527363449335098,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 1.0929982662200928,
+      "learning_rate": 0.00036338623136761495,
+      "loss": 0.7058369636535644,
+      "mean_token_accuracy": 0.7997087433934211,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.728419938114252,
+      "eval_loss": 0.744873046875,
+      "eval_mean_token_accuracy": 0.7966634296299366,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 89.6597,
+      "eval_samples_per_second": 15.838,
+      "eval_steps_per_second": 1.985,
+      "step": 560
+    },
+    {
+      "entropy": 0.7406099259853363,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.409364104270935,
+      "learning_rate": 0.00036298997045790515,
+      "loss": 0.6902852058410645,
+      "mean_token_accuracy": 0.805386807769537,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.6991632758231645,
+      "eval_loss": 0.7480319738388062,
+      "eval_mean_token_accuracy": 0.7970068006033308,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 89.8329,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 580
+    },
+    {
+      "entropy": 0.7459449704736472,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.7053111791610718,
+      "learning_rate": 0.00036254259372255277,
+      "loss": 0.7044640064239502,
+      "mean_token_accuracy": 0.8005018755793571,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7454566660891758,
+      "eval_loss": 0.7551484704017639,
+      "eval_mean_token_accuracy": 0.7897093731365846,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 89.8313,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 600
+    },
+    {
+      "entropy": 0.7621391348540782,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 1.174443244934082,
+      "learning_rate": 0.000362044227763882,
+      "loss": 0.7111573696136475,
+      "mean_token_accuracy": 0.798752411454916,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7664387949397055,
+      "eval_loss": 0.7365804314613342,
+      "eval_mean_token_accuracy": 0.7963101200843126,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 89.9623,
+      "eval_samples_per_second": 15.784,
+      "eval_steps_per_second": 1.979,
+      "step": 620
+    },
+    {
+      "entropy": 0.7424944408237935,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.2898184061050415,
+      "learning_rate": 0.000361495013613564,
+      "loss": 0.6913110733032226,
+      "mean_token_accuracy": 0.8009015507996082,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7255479529332579,
+      "eval_loss": 0.7345991730690002,
+      "eval_mean_token_accuracy": 0.8013592945056015,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 89.817,
+      "eval_samples_per_second": 15.81,
+      "eval_steps_per_second": 1.982,
+      "step": 640
+    },
+    {
+      "entropy": 0.7303367637097835,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.1310476064682007,
+      "learning_rate": 0.00036089510669270666,
+      "loss": 0.6875617027282714,
+      "mean_token_accuracy": 0.8035397171974182,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7101308248016271,
+      "eval_loss": 0.7394917011260986,
+      "eval_mean_token_accuracy": 0.7990792942850777,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 89.7431,
+      "eval_samples_per_second": 15.823,
+      "eval_steps_per_second": 1.983,
+      "step": 660
+    },
+    {
+      "entropy": 0.7387678384780884,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.9645389914512634,
+      "learning_rate": 0.0003602446767678725,
+      "loss": 0.6911204338073731,
+      "mean_token_accuracy": 0.8056333385407924,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7281073737010527,
+      "eval_loss": 0.7376708984375,
+      "eval_mean_token_accuracy": 0.7997356415464637,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 90.0338,
+      "eval_samples_per_second": 15.772,
+      "eval_steps_per_second": 1.977,
+      "step": 680
+    },
+    {
+      "entropy": 0.7449554048478604,
+      "epoch": 1.687990343995172,
+      "grad_norm": 1.9069629907608032,
+      "learning_rate": 0.0003595439079030364,
+      "loss": 0.7042286396026611,
+      "mean_token_accuracy": 0.8018639378249646,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7022633817088738,
+      "eval_loss": 0.7320939898490906,
+      "eval_mean_token_accuracy": 0.8006051541044471,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 89.8948,
+      "eval_samples_per_second": 15.796,
+      "eval_steps_per_second": 1.98,
+      "step": 700
+    },
+    {
+      "entropy": 0.723249051719904,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.8333877325057983,
+      "learning_rate": 0.00035879299840749777,
+      "loss": 0.7043714046478271,
+      "mean_token_accuracy": 0.8028345607221127,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.672436946898364,
+      "eval_loss": 0.713233470916748,
+      "eval_mean_token_accuracy": 0.8047760577684038,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 89.6394,
+      "eval_samples_per_second": 15.841,
+      "eval_steps_per_second": 1.986,
+      "step": 720
+    },
+    {
+      "entropy": 0.730014206469059,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.4580037593841553,
+      "learning_rate": 0.00035799216077976137,
+      "loss": 0.6985176563262939,
+      "mean_token_accuracy": 0.8015547141432762,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.6908873795123582,
+      "eval_loss": 0.7198202013969421,
+      "eval_mean_token_accuracy": 0.8024131682481659,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 89.3771,
+      "eval_samples_per_second": 15.888,
+      "eval_steps_per_second": 1.992,
+      "step": 740
+    },
+    {
+      "entropy": 0.7219605796039105,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.51682710647583,
+      "learning_rate": 0.000357141621647403,
+      "loss": 0.6844424724578857,
+      "mean_token_accuracy": 0.8032813094556331,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7113401109582922,
+      "eval_loss": 0.7195309400558472,
+      "eval_mean_token_accuracy": 0.8019187035185568,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 89.4658,
+      "eval_samples_per_second": 15.872,
+      "eval_steps_per_second": 1.99,
+      "step": 760
+    },
+    {
+      "entropy": 0.729826743900776,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 1.0838000774383545,
+      "learning_rate": 0.0003562416217029361,
+      "loss": 0.685858678817749,
+      "mean_token_accuracy": 0.8050567395985126,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.649370267484965,
+      "eval_loss": 0.7177029252052307,
+      "eval_mean_token_accuracy": 0.8050174043419656,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 89.2287,
+      "eval_samples_per_second": 15.914,
+      "eval_steps_per_second": 1.995,
+      "step": 780
+    },
+    {
+      "entropy": 0.6799364153295755,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 1.1364924907684326,
+      "learning_rate": 0.00035529241563569903,
+      "loss": 0.6714428901672364,
+      "mean_token_accuracy": 0.8087341658771038,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6603799047095052,
+      "eval_loss": 0.7201809287071228,
+      "eval_mean_token_accuracy": 0.802863868769635,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 89.2718,
+      "eval_samples_per_second": 15.906,
+      "eval_steps_per_second": 1.994,
+      "step": 800
+    },
+    {
+      "entropy": 0.7078610375523567,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 1.2191942930221558,
+      "learning_rate": 0.0003542942720597808,
+      "loss": 0.6996590614318847,
+      "mean_token_accuracy": 0.8049512051045895,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.7240545264120852,
+      "eval_loss": 0.7301309108734131,
+      "eval_mean_token_accuracy": 0.8015973166133581,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 89.3438,
+      "eval_samples_per_second": 15.894,
+      "eval_steps_per_second": 1.992,
+      "step": 820
+    },
+    {
+      "entropy": 0.6557391556826505,
+      "epoch": 2.024140012070006,
+      "grad_norm": 1.36203134059906,
+      "learning_rate": 0.0003532474734380064,
+      "loss": 0.6327113628387451,
+      "mean_token_accuracy": 0.8142829847026181,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6573808648613062,
+      "eval_loss": 0.7334365844726562,
+      "eval_mean_token_accuracy": 0.8043633893634496,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 89.3359,
+      "eval_samples_per_second": 15.895,
+      "eval_steps_per_second": 1.992,
+      "step": 840
+    },
+    {
+      "entropy": 0.6409128502011299,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 1.5804955959320068,
+      "learning_rate": 0.0003521523160020035,
+      "loss": 0.6023559093475341,
+      "mean_token_accuracy": 0.8222251623868942,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6983234788594621,
+      "eval_loss": 0.7232135534286499,
+      "eval_mean_token_accuracy": 0.8010654268639811,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 89.2615,
+      "eval_samples_per_second": 15.908,
+      "eval_steps_per_second": 1.994,
+      "step": 860
+    },
+    {
+      "entropy": 0.6151149723678827,
+      "epoch": 2.12070006035003,
+      "grad_norm": 1.1975436210632324,
+      "learning_rate": 0.00035100910966837193,
+      "loss": 0.5803564548492431,
+      "mean_token_accuracy": 0.8256325736641884,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6636555573243773,
+      "eval_loss": 0.7205662727355957,
+      "eval_mean_token_accuracy": 0.8036768369460374,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 89.2957,
+      "eval_samples_per_second": 15.902,
+      "eval_steps_per_second": 1.993,
+      "step": 880
+    },
+    {
+      "entropy": 0.6201727617532015,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9907131195068359,
+      "learning_rate": 0.0003498181779509813,
+      "loss": 0.6101872444152832,
+      "mean_token_accuracy": 0.8196512959897518,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.6473788086617931,
+      "eval_loss": 0.7062397003173828,
+      "eval_mean_token_accuracy": 0.8078362184963869,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 88.6198,
+      "eval_samples_per_second": 16.024,
+      "eval_steps_per_second": 2.009,
+      "step": 900
+    },
+    {
+      "entropy": 0.6097332935780286,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.9055633544921875,
+      "learning_rate": 0.00034857985786942026,
+      "loss": 0.5956094264984131,
+      "mean_token_accuracy": 0.8240575887262821,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.612629868341296,
+      "eval_loss": 0.7106770277023315,
+      "eval_mean_token_accuracy": 0.805324277181304,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 89.4383,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 920
+    },
+    {
+      "entropy": 0.6182701248675585,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.9513736367225647,
+      "learning_rate": 0.00034729449985362404,
+      "loss": 0.6019269466400147,
+      "mean_token_accuracy": 0.8187106013298034,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6437647456533453,
+      "eval_loss": 0.7173502445220947,
+      "eval_mean_token_accuracy": 0.8056439573175451,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.9568,
+      "eval_samples_per_second": 15.963,
+      "eval_steps_per_second": 2.001,
+      "step": 940
+    },
+    {
+      "entropy": 0.6508570514619351,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.9606207013130188,
+      "learning_rate": 0.0003459624676447067,
+      "loss": 0.6296409606933594,
+      "mean_token_accuracy": 0.8167732007801533,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6288247135248077,
+      "eval_loss": 0.7160675525665283,
+      "eval_mean_token_accuracy": 0.8066010689467527,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 89.0044,
+      "eval_samples_per_second": 15.954,
+      "eval_steps_per_second": 2.0,
+      "step": 960
+    },
+    {
+      "entropy": 0.6380081418901682,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.9677181839942932,
+      "learning_rate": 0.000344584138192027,
+      "loss": 0.6210546493530273,
+      "mean_token_accuracy": 0.8186135642230511,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6407805031604981,
+      "eval_loss": 0.7055043578147888,
+      "eval_mean_token_accuracy": 0.8070103157772107,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 89.5606,
+      "eval_samples_per_second": 15.855,
+      "eval_steps_per_second": 1.987,
+      "step": 980
+    },
+    {
+      "entropy": 0.6310251638293266,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 2.1046407222747803,
+      "learning_rate": 0.000343159901546516,
+      "loss": 0.6165830135345459,
+      "mean_token_accuracy": 0.8207855716347694,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6461507470420237,
+      "eval_loss": 0.7292136549949646,
+      "eval_mean_token_accuracy": 0.8030418261383356,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 89.315,
+      "eval_samples_per_second": 15.899,
+      "eval_steps_per_second": 1.993,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6256617344915867,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 1.1768349409103394,
+      "learning_rate": 0.0003416901607502972,
+      "loss": 0.6142709255218506,
+      "mean_token_accuracy": 0.8220669947564602,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6150523993406403,
+      "eval_loss": 0.7019563317298889,
+      "eval_mean_token_accuracy": 0.8094323099998946,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 89.419,
+      "eval_samples_per_second": 15.88,
+      "eval_steps_per_second": 1.991,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6126735735684633,
+      "epoch": 2.506940253470127,
+      "grad_norm": 1.8065496683120728,
+      "learning_rate": 0.00034017533172263055,
+      "loss": 0.6056458473205566,
+      "mean_token_accuracy": 0.8203762136399746,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6544345456562685,
+      "eval_loss": 0.7086517214775085,
+      "eval_mean_token_accuracy": 0.806709805901131,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 88.981,
+      "eval_samples_per_second": 15.958,
+      "eval_steps_per_second": 2.0,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6311754353344441,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 1.0576369762420654,
+      "learning_rate": 0.00033861584314221236,
+      "loss": 0.6176392555236816,
+      "mean_token_accuracy": 0.8177410490810871,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6273799200406235,
+      "eval_loss": 0.708163321018219,
+      "eval_mean_token_accuracy": 0.8032017099053672,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 89.7668,
+      "eval_samples_per_second": 15.819,
+      "eval_steps_per_second": 1.983,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6200249589979648,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.997065007686615,
+      "learning_rate": 0.0003370121363258637,
+      "loss": 0.6203888893127442,
+      "mean_token_accuracy": 0.8196275025606156,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6457576811983344,
+      "eval_loss": 0.6972290277481079,
+      "eval_mean_token_accuracy": 0.8095980849158898,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 89.2027,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 1080
+    },
+    {
+      "entropy": 0.626810473203659,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.865475058555603,
+      "learning_rate": 0.0003353646651036438,
+      "loss": 0.6102585792541504,
+      "mean_token_accuracy": 0.819708751142025,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.6412022431914726,
+      "eval_loss": 0.7036139965057373,
+      "eval_mean_token_accuracy": 0.8070317135098275,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 89.4733,
+      "eval_samples_per_second": 15.871,
+      "eval_steps_per_second": 1.989,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6581619590520859,
+      "epoch": 2.700060350030175,
+      "grad_norm": 1.00307035446167,
+      "learning_rate": 0.00033367389569042064,
+      "loss": 0.6397720336914062,
+      "mean_token_accuracy": 0.81413309648633,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6288736402318719,
+      "eval_loss": 0.6966825723648071,
+      "eval_mean_token_accuracy": 0.8094039669867312,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 88.7341,
+      "eval_samples_per_second": 16.003,
+      "eval_steps_per_second": 2.006,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6399412982165813,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.8589329719543457,
+      "learning_rate": 0.0003319403065539384,
+      "loss": 0.6347338199615479,
+      "mean_token_accuracy": 0.8139534957706929,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.683491913455256,
+      "eval_loss": 0.6984149217605591,
+      "eval_mean_token_accuracy": 0.8083989469522841,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 89.3929,
+      "eval_samples_per_second": 15.885,
+      "eval_steps_per_second": 1.991,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6264835461974144,
+      "epoch": 2.796620398310199,
+      "grad_norm": 0.7807853817939758,
+      "learning_rate": 0.0003301643882794163,
+      "loss": 0.6162629127502441,
+      "mean_token_accuracy": 0.8216292977333068,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6578792159476977,
+      "eval_loss": 0.6921441555023193,
+      "eval_mean_token_accuracy": 0.8094299177775223,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 89.2356,
+      "eval_samples_per_second": 15.913,
+      "eval_steps_per_second": 1.995,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6269129011780024,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 1.1675593852996826,
+      "learning_rate": 0.0003283466434307189,
+      "loss": 0.6235532760620117,
+      "mean_token_accuracy": 0.8188897788524627,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6533248447970058,
+      "eval_loss": 0.696849524974823,
+      "eval_mean_token_accuracy": 0.8102703713968898,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2602,
+      "eval_samples_per_second": 16.089,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.6259873781353236,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.9596486687660217,
+      "learning_rate": 0.00032648758640813655,
+      "loss": 0.6118728160858155,
+      "mean_token_accuracy": 0.8171859130263328,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6212078754821521,
+      "eval_loss": 0.6942870020866394,
+      "eval_mean_token_accuracy": 0.810484343365337,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 89.4375,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 1200
+    },
+    {
+      "entropy": 0.6142537571489811,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 0.9625058770179749,
+      "learning_rate": 0.00032458774330281615,
+      "loss": 0.6060261249542236,
+      "mean_token_accuracy": 0.8223113484680653,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6553665667437436,
+      "eval_loss": 0.6835269331932068,
+      "eval_mean_token_accuracy": 0.8134723968720168,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 89.5092,
+      "eval_samples_per_second": 15.864,
+      "eval_steps_per_second": 1.989,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6301972426474094,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.9657333493232727,
+      "learning_rate": 0.0003226476517478835,
+      "loss": 0.6277759075164795,
+      "mean_token_accuracy": 0.8164266042411328,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6594253118788258,
+      "eval_loss": 0.6820935606956482,
+      "eval_mean_token_accuracy": 0.8132368198941263,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 89.6211,
+      "eval_samples_per_second": 15.844,
+      "eval_steps_per_second": 1.986,
+      "step": 1240
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.2419491368519475e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1260/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1260/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1260/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1260/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1260/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..7ce4b072fcbe5b636867e13937c56fedee3b68b4
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1260/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.02728394274790723,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1260/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1260/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1260/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1260/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1260/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1260/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1260/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1260/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..ed4e1da389ad8d604f86c0c4b5e35136799f708e
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1260/trainer_state.json
@@ -0,0 +1,1357 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 3.036210018105009,
+  "eval_steps": 20,
+  "global_step": 1260,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.024795526266098,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.945375919342041,
+      "learning_rate": 1.6698127428345936e-05,
+      "loss": 1.7825420379638672,
+      "mean_token_accuracy": 0.6315708436071873,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.29921497588747,
+      "eval_loss": 1.169131875038147,
+      "eval_mean_token_accuracy": 0.7185706342204233,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 91.7898,
+      "eval_samples_per_second": 15.47,
+      "eval_steps_per_second": 1.939,
+      "step": 20
+    },
+    {
+      "entropy": 1.0346344470977784,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.7916979789733887,
+      "learning_rate": 3.427510366871008e-05,
+      "loss": 0.9480395317077637,
+      "mean_token_accuracy": 0.7543429024517536,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 0.9673695430326997,
+      "eval_loss": 0.8588430285453796,
+      "eval_mean_token_accuracy": 0.7737233002534073,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 89.8219,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 40
+    },
+    {
+      "entropy": 0.8977566614747048,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7972187995910645,
+      "learning_rate": 5.185207990907422e-05,
+      "loss": 0.8029604911804199,
+      "mean_token_accuracy": 0.7818016350269318,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8436836982041263,
+      "eval_loss": 0.7893033623695374,
+      "eval_mean_token_accuracy": 0.7858914984076211,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 89.9349,
+      "eval_samples_per_second": 15.789,
+      "eval_steps_per_second": 1.979,
+      "step": 60
+    },
+    {
+      "entropy": 0.8650321319699288,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.2128729820251465,
+      "learning_rate": 6.942905614943836e-05,
+      "loss": 0.77417893409729,
+      "mean_token_accuracy": 0.7875479467213153,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8209857180547179,
+      "eval_loss": 0.7572280168533325,
+      "eval_mean_token_accuracy": 0.7925970383574453,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 89.9459,
+      "eval_samples_per_second": 15.787,
+      "eval_steps_per_second": 1.979,
+      "step": 80
+    },
+    {
+      "entropy": 0.8096660938113928,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.1254287958145142,
+      "learning_rate": 8.700603238980251e-05,
+      "loss": 0.7170331001281738,
+      "mean_token_accuracy": 0.7994555257260799,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.792780208788561,
+      "eval_loss": 0.7442984580993652,
+      "eval_mean_token_accuracy": 0.7954704573984896,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 90.294,
+      "eval_samples_per_second": 15.726,
+      "eval_steps_per_second": 1.971,
+      "step": 100
+    },
+    {
+      "entropy": 0.7979829199612141,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2140507698059082,
+      "learning_rate": 0.00010458300863016666,
+      "loss": 0.7141861915588379,
+      "mean_token_accuracy": 0.8000014744699001,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.7938692486018277,
+      "eval_loss": 0.7279797196388245,
+      "eval_mean_token_accuracy": 0.7989798734027348,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 89.7189,
+      "eval_samples_per_second": 15.827,
+      "eval_steps_per_second": 1.984,
+      "step": 120
+    },
+    {
+      "entropy": 0.7916347607970238,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.0506458282470703,
+      "learning_rate": 0.0001221599848705308,
+      "loss": 0.7117055416107178,
+      "mean_token_accuracy": 0.7970046654343605,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.8052781063519167,
+      "eval_loss": 0.722576916217804,
+      "eval_mean_token_accuracy": 0.7982530523551984,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 90.0376,
+      "eval_samples_per_second": 15.771,
+      "eval_steps_per_second": 1.977,
+      "step": 140
+    },
+    {
+      "entropy": 0.7880666613578796,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8409207463264465,
+      "learning_rate": 0.00013973696111089492,
+      "loss": 0.7052523136138916,
+      "mean_token_accuracy": 0.8000123649835587,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7787443134891853,
+      "eval_loss": 0.7137772440910339,
+      "eval_mean_token_accuracy": 0.8014209287220173,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 90.2562,
+      "eval_samples_per_second": 15.733,
+      "eval_steps_per_second": 1.972,
+      "step": 160
+    },
+    {
+      "entropy": 0.7696007996797561,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.0182477235794067,
+      "learning_rate": 0.00015731393735125907,
+      "loss": 0.7008797645568847,
+      "mean_token_accuracy": 0.8007259473204613,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.752259682068664,
+      "eval_loss": 0.7171286344528198,
+      "eval_mean_token_accuracy": 0.80052122574174,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 90.0797,
+      "eval_samples_per_second": 15.764,
+      "eval_steps_per_second": 1.976,
+      "step": 180
+    },
+    {
+      "entropy": 0.7828126326203346,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9553632736206055,
+      "learning_rate": 0.0001748909135916232,
+      "loss": 0.7046853542327881,
+      "mean_token_accuracy": 0.7988562889397144,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7389638169427936,
+      "eval_loss": 0.7187935709953308,
+      "eval_mean_token_accuracy": 0.7995899999409579,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 89.9223,
+      "eval_samples_per_second": 15.791,
+      "eval_steps_per_second": 1.979,
+      "step": 200
+    },
+    {
+      "entropy": 0.7909286297857762,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 1.134459376335144,
+      "learning_rate": 0.00019246788983198735,
+      "loss": 0.6989312171936035,
+      "mean_token_accuracy": 0.7991565138101577,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7496965199374082,
+      "eval_loss": 0.7106321454048157,
+      "eval_mean_token_accuracy": 0.8046183442131857,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 89.6899,
+      "eval_samples_per_second": 15.832,
+      "eval_steps_per_second": 1.985,
+      "step": 220
+    },
+    {
+      "entropy": 0.7651963323354721,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.9557585716247559,
+      "learning_rate": 0.0002100448660723515,
+      "loss": 0.6984559059143066,
+      "mean_token_accuracy": 0.8019823372364044,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7615563079212488,
+      "eval_loss": 0.7098539471626282,
+      "eval_mean_token_accuracy": 0.8047233508544022,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 90.3752,
+      "eval_samples_per_second": 15.712,
+      "eval_steps_per_second": 1.97,
+      "step": 240
+    },
+    {
+      "entropy": 0.787507726252079,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.9279311299324036,
+      "learning_rate": 0.00022762184231271568,
+      "loss": 0.708138370513916,
+      "mean_token_accuracy": 0.8013308703899383,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7856640896100676,
+      "eval_loss": 0.7191570401191711,
+      "eval_mean_token_accuracy": 0.8024677249153008,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 93.7336,
+      "eval_samples_per_second": 15.149,
+      "eval_steps_per_second": 1.899,
+      "step": 260
+    },
+    {
+      "entropy": 0.768489234894514,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.7992776036262512,
+      "learning_rate": 0.0002451988185530798,
+      "loss": 0.7035849571228028,
+      "mean_token_accuracy": 0.8031105332076549,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7682498740346244,
+      "eval_loss": 0.7149022221565247,
+      "eval_mean_token_accuracy": 0.7973497995499814,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 89.7309,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 280
+    },
+    {
+      "entropy": 0.7720955617725849,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.1931514739990234,
+      "learning_rate": 0.0002627757947934439,
+      "loss": 0.6965402603149414,
+      "mean_token_accuracy": 0.8040348663926125,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7214708304807042,
+      "eval_loss": 0.7350823879241943,
+      "eval_mean_token_accuracy": 0.7942911925610532,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 89.6451,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 300
+    },
+    {
+      "entropy": 0.7930095426738262,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8994467258453369,
+      "learning_rate": 0.0002803527710338081,
+      "loss": 0.7331492900848389,
+      "mean_token_accuracy": 0.796185302734375,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.8142535488926963,
+      "eval_loss": 0.7237834334373474,
+      "eval_mean_token_accuracy": 0.799495273761535,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 90.6759,
+      "eval_samples_per_second": 15.66,
+      "eval_steps_per_second": 1.963,
+      "step": 320
+    },
+    {
+      "entropy": 0.7825102441012859,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 1.8224239349365234,
+      "learning_rate": 0.00029792974727417223,
+      "loss": 0.7331175804138184,
+      "mean_token_accuracy": 0.7977667152881622,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7327135443017724,
+      "eval_loss": 0.7388784289360046,
+      "eval_mean_token_accuracy": 0.8007041252730938,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 89.7692,
+      "eval_samples_per_second": 15.818,
+      "eval_steps_per_second": 1.983,
+      "step": 340
+    },
+    {
+      "entropy": 0.819331557303667,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 1.0407779216766357,
+      "learning_rate": 0.0003155067235145364,
+      "loss": 0.7456695556640625,
+      "mean_token_accuracy": 0.7936980701982975,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.8065585420372781,
+      "eval_loss": 0.735519528388977,
+      "eval_mean_token_accuracy": 0.7974206296245704,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 89.4336,
+      "eval_samples_per_second": 15.878,
+      "eval_steps_per_second": 1.99,
+      "step": 360
+    },
+    {
+      "entropy": 0.8114653021097183,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.1209490299224854,
+      "learning_rate": 0.0003330836997549005,
+      "loss": 0.7340899467468261,
+      "mean_token_accuracy": 0.7914272703230381,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.8025527285056168,
+      "eval_loss": 0.7295576930046082,
+      "eval_mean_token_accuracy": 0.7990576628218876,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 89.9579,
+      "eval_samples_per_second": 15.785,
+      "eval_steps_per_second": 1.979,
+      "step": 380
+    },
+    {
+      "entropy": 0.8042497783899307,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 1.2437798976898193,
+      "learning_rate": 0.0003506606759952647,
+      "loss": 0.7381744384765625,
+      "mean_token_accuracy": 0.7967362694442273,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7717011336530193,
+      "eval_loss": 0.7436173558235168,
+      "eval_mean_token_accuracy": 0.7975849482450592,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 89.8411,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 400
+    },
+    {
+      "entropy": 0.7987750010056929,
+      "epoch": 1.012070006035003,
+      "grad_norm": 1.2390918731689453,
+      "learning_rate": 0.000364721224843344,
+      "loss": 0.7317790031433106,
+      "mean_token_accuracy": 0.7986087551364651,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7333170604170038,
+      "eval_loss": 0.7456948757171631,
+      "eval_mean_token_accuracy": 0.7964897577682238,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 89.7288,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 420
+    },
+    {
+      "entropy": 0.7372976846992969,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.1552740335464478,
+      "learning_rate": 0.00036468510102269575,
+      "loss": 0.6734249114990234,
+      "mean_token_accuracy": 0.805834824591875,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.7130355526891987,
+      "eval_loss": 0.7555333375930786,
+      "eval_mean_token_accuracy": 0.7961652781186479,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 89.8713,
+      "eval_samples_per_second": 15.8,
+      "eval_steps_per_second": 1.981,
+      "step": 440
+    },
+    {
+      "entropy": 0.7465668715536594,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 1.1991469860076904,
+      "learning_rate": 0.0003645973816745066,
+      "loss": 0.6871879577636719,
+      "mean_token_accuracy": 0.8042450234293937,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7772979542110743,
+      "eval_loss": 0.7561826705932617,
+      "eval_mean_token_accuracy": 0.7953876427720102,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 89.8395,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 460
+    },
+    {
+      "entropy": 0.7477855801582336,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 1.325486660003662,
+      "learning_rate": 0.00036445809162231435,
+      "loss": 0.6836830139160156,
+      "mean_token_accuracy": 0.8021619468927383,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.7098110676481483,
+      "eval_loss": 0.758007287979126,
+      "eval_mean_token_accuracy": 0.7909953189030122,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 89.6101,
+      "eval_samples_per_second": 15.846,
+      "eval_steps_per_second": 1.986,
+      "step": 480
+    },
+    {
+      "entropy": 0.7632691070437432,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 1.0122262239456177,
+      "learning_rate": 0.0003642672702835562,
+      "loss": 0.7071213245391845,
+      "mean_token_accuracy": 0.8000680930912495,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7645382720432924,
+      "eval_loss": 0.7492606043815613,
+      "eval_mean_token_accuracy": 0.7976858676149604,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 89.6092,
+      "eval_samples_per_second": 15.847,
+      "eval_steps_per_second": 1.986,
+      "step": 500
+    },
+    {
+      "entropy": 0.7623726457357407,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.367774486541748,
+      "learning_rate": 0.00036402497165841384,
+      "loss": 0.6965175151824952,
+      "mean_token_accuracy": 0.8016501650214195,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7868083715438843,
+      "eval_loss": 0.7481449842453003,
+      "eval_mean_token_accuracy": 0.799330582779445,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 89.8241,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 520
+    },
+    {
+      "entropy": 0.7535207405686378,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 1.213614821434021,
+      "learning_rate": 0.00036373126431453207,
+      "loss": 0.7078310489654541,
+      "mean_token_accuracy": 0.7998907208442688,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7829091950748743,
+      "eval_loss": 0.7416072487831116,
+      "eval_mean_token_accuracy": 0.7969146250339036,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 89.4607,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 540
+    },
+    {
+      "entropy": 0.7527363449335098,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 1.0929982662200928,
+      "learning_rate": 0.00036338623136761495,
+      "loss": 0.7058369636535644,
+      "mean_token_accuracy": 0.7997087433934211,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.728419938114252,
+      "eval_loss": 0.744873046875,
+      "eval_mean_token_accuracy": 0.7966634296299366,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 89.6597,
+      "eval_samples_per_second": 15.838,
+      "eval_steps_per_second": 1.985,
+      "step": 560
+    },
+    {
+      "entropy": 0.7406099259853363,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.409364104270935,
+      "learning_rate": 0.00036298997045790515,
+      "loss": 0.6902852058410645,
+      "mean_token_accuracy": 0.805386807769537,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.6991632758231645,
+      "eval_loss": 0.7480319738388062,
+      "eval_mean_token_accuracy": 0.7970068006033308,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 89.8329,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 580
+    },
+    {
+      "entropy": 0.7459449704736472,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.7053111791610718,
+      "learning_rate": 0.00036254259372255277,
+      "loss": 0.7044640064239502,
+      "mean_token_accuracy": 0.8005018755793571,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7454566660891758,
+      "eval_loss": 0.7551484704017639,
+      "eval_mean_token_accuracy": 0.7897093731365846,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 89.8313,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 600
+    },
+    {
+      "entropy": 0.7621391348540782,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 1.174443244934082,
+      "learning_rate": 0.000362044227763882,
+      "loss": 0.7111573696136475,
+      "mean_token_accuracy": 0.798752411454916,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7664387949397055,
+      "eval_loss": 0.7365804314613342,
+      "eval_mean_token_accuracy": 0.7963101200843126,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 89.9623,
+      "eval_samples_per_second": 15.784,
+      "eval_steps_per_second": 1.979,
+      "step": 620
+    },
+    {
+      "entropy": 0.7424944408237935,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.2898184061050415,
+      "learning_rate": 0.000361495013613564,
+      "loss": 0.6913110733032226,
+      "mean_token_accuracy": 0.8009015507996082,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7255479529332579,
+      "eval_loss": 0.7345991730690002,
+      "eval_mean_token_accuracy": 0.8013592945056015,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 89.817,
+      "eval_samples_per_second": 15.81,
+      "eval_steps_per_second": 1.982,
+      "step": 640
+    },
+    {
+      "entropy": 0.7303367637097835,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.1310476064682007,
+      "learning_rate": 0.00036089510669270666,
+      "loss": 0.6875617027282714,
+      "mean_token_accuracy": 0.8035397171974182,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7101308248016271,
+      "eval_loss": 0.7394917011260986,
+      "eval_mean_token_accuracy": 0.7990792942850777,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 89.7431,
+      "eval_samples_per_second": 15.823,
+      "eval_steps_per_second": 1.983,
+      "step": 660
+    },
+    {
+      "entropy": 0.7387678384780884,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.9645389914512634,
+      "learning_rate": 0.0003602446767678725,
+      "loss": 0.6911204338073731,
+      "mean_token_accuracy": 0.8056333385407924,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7281073737010527,
+      "eval_loss": 0.7376708984375,
+      "eval_mean_token_accuracy": 0.7997356415464637,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 90.0338,
+      "eval_samples_per_second": 15.772,
+      "eval_steps_per_second": 1.977,
+      "step": 680
+    },
+    {
+      "entropy": 0.7449554048478604,
+      "epoch": 1.687990343995172,
+      "grad_norm": 1.9069629907608032,
+      "learning_rate": 0.0003595439079030364,
+      "loss": 0.7042286396026611,
+      "mean_token_accuracy": 0.8018639378249646,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7022633817088738,
+      "eval_loss": 0.7320939898490906,
+      "eval_mean_token_accuracy": 0.8006051541044471,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 89.8948,
+      "eval_samples_per_second": 15.796,
+      "eval_steps_per_second": 1.98,
+      "step": 700
+    },
+    {
+      "entropy": 0.723249051719904,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.8333877325057983,
+      "learning_rate": 0.00035879299840749777,
+      "loss": 0.7043714046478271,
+      "mean_token_accuracy": 0.8028345607221127,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.672436946898364,
+      "eval_loss": 0.713233470916748,
+      "eval_mean_token_accuracy": 0.8047760577684038,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 89.6394,
+      "eval_samples_per_second": 15.841,
+      "eval_steps_per_second": 1.986,
+      "step": 720
+    },
+    {
+      "entropy": 0.730014206469059,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.4580037593841553,
+      "learning_rate": 0.00035799216077976137,
+      "loss": 0.6985176563262939,
+      "mean_token_accuracy": 0.8015547141432762,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.6908873795123582,
+      "eval_loss": 0.7198202013969421,
+      "eval_mean_token_accuracy": 0.8024131682481659,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 89.3771,
+      "eval_samples_per_second": 15.888,
+      "eval_steps_per_second": 1.992,
+      "step": 740
+    },
+    {
+      "entropy": 0.7219605796039105,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.51682710647583,
+      "learning_rate": 0.000357141621647403,
+      "loss": 0.6844424724578857,
+      "mean_token_accuracy": 0.8032813094556331,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7113401109582922,
+      "eval_loss": 0.7195309400558472,
+      "eval_mean_token_accuracy": 0.8019187035185568,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 89.4658,
+      "eval_samples_per_second": 15.872,
+      "eval_steps_per_second": 1.99,
+      "step": 760
+    },
+    {
+      "entropy": 0.729826743900776,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 1.0838000774383545,
+      "learning_rate": 0.0003562416217029361,
+      "loss": 0.685858678817749,
+      "mean_token_accuracy": 0.8050567395985126,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.649370267484965,
+      "eval_loss": 0.7177029252052307,
+      "eval_mean_token_accuracy": 0.8050174043419656,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 89.2287,
+      "eval_samples_per_second": 15.914,
+      "eval_steps_per_second": 1.995,
+      "step": 780
+    },
+    {
+      "entropy": 0.6799364153295755,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 1.1364924907684326,
+      "learning_rate": 0.00035529241563569903,
+      "loss": 0.6714428901672364,
+      "mean_token_accuracy": 0.8087341658771038,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6603799047095052,
+      "eval_loss": 0.7201809287071228,
+      "eval_mean_token_accuracy": 0.802863868769635,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 89.2718,
+      "eval_samples_per_second": 15.906,
+      "eval_steps_per_second": 1.994,
+      "step": 800
+    },
+    {
+      "entropy": 0.7078610375523567,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 1.2191942930221558,
+      "learning_rate": 0.0003542942720597808,
+      "loss": 0.6996590614318847,
+      "mean_token_accuracy": 0.8049512051045895,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.7240545264120852,
+      "eval_loss": 0.7301309108734131,
+      "eval_mean_token_accuracy": 0.8015973166133581,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 89.3438,
+      "eval_samples_per_second": 15.894,
+      "eval_steps_per_second": 1.992,
+      "step": 820
+    },
+    {
+      "entropy": 0.6557391556826505,
+      "epoch": 2.024140012070006,
+      "grad_norm": 1.36203134059906,
+      "learning_rate": 0.0003532474734380064,
+      "loss": 0.6327113628387451,
+      "mean_token_accuracy": 0.8142829847026181,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6573808648613062,
+      "eval_loss": 0.7334365844726562,
+      "eval_mean_token_accuracy": 0.8043633893634496,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 89.3359,
+      "eval_samples_per_second": 15.895,
+      "eval_steps_per_second": 1.992,
+      "step": 840
+    },
+    {
+      "entropy": 0.6409128502011299,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 1.5804955959320068,
+      "learning_rate": 0.0003521523160020035,
+      "loss": 0.6023559093475341,
+      "mean_token_accuracy": 0.8222251623868942,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6983234788594621,
+      "eval_loss": 0.7232135534286499,
+      "eval_mean_token_accuracy": 0.8010654268639811,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 89.2615,
+      "eval_samples_per_second": 15.908,
+      "eval_steps_per_second": 1.994,
+      "step": 860
+    },
+    {
+      "entropy": 0.6151149723678827,
+      "epoch": 2.12070006035003,
+      "grad_norm": 1.1975436210632324,
+      "learning_rate": 0.00035100910966837193,
+      "loss": 0.5803564548492431,
+      "mean_token_accuracy": 0.8256325736641884,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6636555573243773,
+      "eval_loss": 0.7205662727355957,
+      "eval_mean_token_accuracy": 0.8036768369460374,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 89.2957,
+      "eval_samples_per_second": 15.902,
+      "eval_steps_per_second": 1.993,
+      "step": 880
+    },
+    {
+      "entropy": 0.6201727617532015,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9907131195068359,
+      "learning_rate": 0.0003498181779509813,
+      "loss": 0.6101872444152832,
+      "mean_token_accuracy": 0.8196512959897518,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.6473788086617931,
+      "eval_loss": 0.7062397003173828,
+      "eval_mean_token_accuracy": 0.8078362184963869,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 88.6198,
+      "eval_samples_per_second": 16.024,
+      "eval_steps_per_second": 2.009,
+      "step": 900
+    },
+    {
+      "entropy": 0.6097332935780286,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.9055633544921875,
+      "learning_rate": 0.00034857985786942026,
+      "loss": 0.5956094264984131,
+      "mean_token_accuracy": 0.8240575887262821,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.612629868341296,
+      "eval_loss": 0.7106770277023315,
+      "eval_mean_token_accuracy": 0.805324277181304,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 89.4383,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 920
+    },
+    {
+      "entropy": 0.6182701248675585,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.9513736367225647,
+      "learning_rate": 0.00034729449985362404,
+      "loss": 0.6019269466400147,
+      "mean_token_accuracy": 0.8187106013298034,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6437647456533453,
+      "eval_loss": 0.7173502445220947,
+      "eval_mean_token_accuracy": 0.8056439573175451,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.9568,
+      "eval_samples_per_second": 15.963,
+      "eval_steps_per_second": 2.001,
+      "step": 940
+    },
+    {
+      "entropy": 0.6508570514619351,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.9606207013130188,
+      "learning_rate": 0.0003459624676447067,
+      "loss": 0.6296409606933594,
+      "mean_token_accuracy": 0.8167732007801533,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6288247135248077,
+      "eval_loss": 0.7160675525665283,
+      "eval_mean_token_accuracy": 0.8066010689467527,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 89.0044,
+      "eval_samples_per_second": 15.954,
+      "eval_steps_per_second": 2.0,
+      "step": 960
+    },
+    {
+      "entropy": 0.6380081418901682,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.9677181839942932,
+      "learning_rate": 0.000344584138192027,
+      "loss": 0.6210546493530273,
+      "mean_token_accuracy": 0.8186135642230511,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6407805031604981,
+      "eval_loss": 0.7055043578147888,
+      "eval_mean_token_accuracy": 0.8070103157772107,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 89.5606,
+      "eval_samples_per_second": 15.855,
+      "eval_steps_per_second": 1.987,
+      "step": 980
+    },
+    {
+      "entropy": 0.6310251638293266,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 2.1046407222747803,
+      "learning_rate": 0.000343159901546516,
+      "loss": 0.6165830135345459,
+      "mean_token_accuracy": 0.8207855716347694,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6461507470420237,
+      "eval_loss": 0.7292136549949646,
+      "eval_mean_token_accuracy": 0.8030418261383356,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 89.315,
+      "eval_samples_per_second": 15.899,
+      "eval_steps_per_second": 1.993,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6256617344915867,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 1.1768349409103394,
+      "learning_rate": 0.0003416901607502972,
+      "loss": 0.6142709255218506,
+      "mean_token_accuracy": 0.8220669947564602,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6150523993406403,
+      "eval_loss": 0.7019563317298889,
+      "eval_mean_token_accuracy": 0.8094323099998946,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 89.419,
+      "eval_samples_per_second": 15.88,
+      "eval_steps_per_second": 1.991,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6126735735684633,
+      "epoch": 2.506940253470127,
+      "grad_norm": 1.8065496683120728,
+      "learning_rate": 0.00034017533172263055,
+      "loss": 0.6056458473205566,
+      "mean_token_accuracy": 0.8203762136399746,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6544345456562685,
+      "eval_loss": 0.7086517214775085,
+      "eval_mean_token_accuracy": 0.806709805901131,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 88.981,
+      "eval_samples_per_second": 15.958,
+      "eval_steps_per_second": 2.0,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6311754353344441,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 1.0576369762420654,
+      "learning_rate": 0.00033861584314221236,
+      "loss": 0.6176392555236816,
+      "mean_token_accuracy": 0.8177410490810871,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6273799200406235,
+      "eval_loss": 0.708163321018219,
+      "eval_mean_token_accuracy": 0.8032017099053672,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 89.7668,
+      "eval_samples_per_second": 15.819,
+      "eval_steps_per_second": 1.983,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6200249589979648,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.997065007686615,
+      "learning_rate": 0.0003370121363258637,
+      "loss": 0.6203888893127442,
+      "mean_token_accuracy": 0.8196275025606156,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6457576811983344,
+      "eval_loss": 0.6972290277481079,
+      "eval_mean_token_accuracy": 0.8095980849158898,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 89.2027,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 1080
+    },
+    {
+      "entropy": 0.626810473203659,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.865475058555603,
+      "learning_rate": 0.0003353646651036438,
+      "loss": 0.6102585792541504,
+      "mean_token_accuracy": 0.819708751142025,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.6412022431914726,
+      "eval_loss": 0.7036139965057373,
+      "eval_mean_token_accuracy": 0.8070317135098275,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 89.4733,
+      "eval_samples_per_second": 15.871,
+      "eval_steps_per_second": 1.989,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6581619590520859,
+      "epoch": 2.700060350030175,
+      "grad_norm": 1.00307035446167,
+      "learning_rate": 0.00033367389569042064,
+      "loss": 0.6397720336914062,
+      "mean_token_accuracy": 0.81413309648633,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6288736402318719,
+      "eval_loss": 0.6966825723648071,
+      "eval_mean_token_accuracy": 0.8094039669867312,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 88.7341,
+      "eval_samples_per_second": 16.003,
+      "eval_steps_per_second": 2.006,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6399412982165813,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.8589329719543457,
+      "learning_rate": 0.0003319403065539384,
+      "loss": 0.6347338199615479,
+      "mean_token_accuracy": 0.8139534957706929,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.683491913455256,
+      "eval_loss": 0.6984149217605591,
+      "eval_mean_token_accuracy": 0.8083989469522841,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 89.3929,
+      "eval_samples_per_second": 15.885,
+      "eval_steps_per_second": 1.991,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6264835461974144,
+      "epoch": 2.796620398310199,
+      "grad_norm": 0.7807853817939758,
+      "learning_rate": 0.0003301643882794163,
+      "loss": 0.6162629127502441,
+      "mean_token_accuracy": 0.8216292977333068,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6578792159476977,
+      "eval_loss": 0.6921441555023193,
+      "eval_mean_token_accuracy": 0.8094299177775223,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 89.2356,
+      "eval_samples_per_second": 15.913,
+      "eval_steps_per_second": 1.995,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6269129011780024,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 1.1675593852996826,
+      "learning_rate": 0.0003283466434307189,
+      "loss": 0.6235532760620117,
+      "mean_token_accuracy": 0.8188897788524627,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6533248447970058,
+      "eval_loss": 0.696849524974823,
+      "eval_mean_token_accuracy": 0.8102703713968898,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2602,
+      "eval_samples_per_second": 16.089,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.6259873781353236,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.9596486687660217,
+      "learning_rate": 0.00032648758640813655,
+      "loss": 0.6118728160858155,
+      "mean_token_accuracy": 0.8171859130263328,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6212078754821521,
+      "eval_loss": 0.6942870020866394,
+      "eval_mean_token_accuracy": 0.810484343365337,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 89.4375,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 1200
+    },
+    {
+      "entropy": 0.6142537571489811,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 0.9625058770179749,
+      "learning_rate": 0.00032458774330281615,
+      "loss": 0.6060261249542236,
+      "mean_token_accuracy": 0.8223113484680653,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6553665667437436,
+      "eval_loss": 0.6835269331932068,
+      "eval_mean_token_accuracy": 0.8134723968720168,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 89.5092,
+      "eval_samples_per_second": 15.864,
+      "eval_steps_per_second": 1.989,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6301972426474094,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.9657333493232727,
+      "learning_rate": 0.0003226476517478835,
+      "loss": 0.6277759075164795,
+      "mean_token_accuracy": 0.8164266042411328,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6594253118788258,
+      "eval_loss": 0.6820935606956482,
+      "eval_mean_token_accuracy": 0.8132368198941263,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 89.6211,
+      "eval_samples_per_second": 15.844,
+      "eval_steps_per_second": 1.986,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5346100655469027,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8541185259819031,
+      "learning_rate": 0.0003206678607662996,
+      "loss": 0.5261692047119141,
+      "mean_token_accuracy": 0.8382289425119177,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5626645662476507,
+      "eval_loss": 0.7028542160987854,
+      "eval_mean_token_accuracy": 0.8133395572056931,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 89.1267,
+      "eval_samples_per_second": 15.932,
+      "eval_steps_per_second": 1.997,
+      "step": 1260
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.2612962738845286e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1280/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1280/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1280/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1280/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1280/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..7ce4b072fcbe5b636867e13937c56fedee3b68b4
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1280/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.02728394274790723,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1280/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1280/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1280/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1280/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1280/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1280/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1280/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1280/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..57a0658bbf82b1e95bf74bf042c63bae5361b7ec
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1280/trainer_state.json
@@ -0,0 +1,1378 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 3.084490042245021,
+  "eval_steps": 20,
+  "global_step": 1280,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.024795526266098,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.945375919342041,
+      "learning_rate": 1.6698127428345936e-05,
+      "loss": 1.7825420379638672,
+      "mean_token_accuracy": 0.6315708436071873,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.29921497588747,
+      "eval_loss": 1.169131875038147,
+      "eval_mean_token_accuracy": 0.7185706342204233,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 91.7898,
+      "eval_samples_per_second": 15.47,
+      "eval_steps_per_second": 1.939,
+      "step": 20
+    },
+    {
+      "entropy": 1.0346344470977784,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.7916979789733887,
+      "learning_rate": 3.427510366871008e-05,
+      "loss": 0.9480395317077637,
+      "mean_token_accuracy": 0.7543429024517536,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 0.9673695430326997,
+      "eval_loss": 0.8588430285453796,
+      "eval_mean_token_accuracy": 0.7737233002534073,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 89.8219,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 40
+    },
+    {
+      "entropy": 0.8977566614747048,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7972187995910645,
+      "learning_rate": 5.185207990907422e-05,
+      "loss": 0.8029604911804199,
+      "mean_token_accuracy": 0.7818016350269318,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8436836982041263,
+      "eval_loss": 0.7893033623695374,
+      "eval_mean_token_accuracy": 0.7858914984076211,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 89.9349,
+      "eval_samples_per_second": 15.789,
+      "eval_steps_per_second": 1.979,
+      "step": 60
+    },
+    {
+      "entropy": 0.8650321319699288,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.2128729820251465,
+      "learning_rate": 6.942905614943836e-05,
+      "loss": 0.77417893409729,
+      "mean_token_accuracy": 0.7875479467213153,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8209857180547179,
+      "eval_loss": 0.7572280168533325,
+      "eval_mean_token_accuracy": 0.7925970383574453,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 89.9459,
+      "eval_samples_per_second": 15.787,
+      "eval_steps_per_second": 1.979,
+      "step": 80
+    },
+    {
+      "entropy": 0.8096660938113928,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.1254287958145142,
+      "learning_rate": 8.700603238980251e-05,
+      "loss": 0.7170331001281738,
+      "mean_token_accuracy": 0.7994555257260799,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.792780208788561,
+      "eval_loss": 0.7442984580993652,
+      "eval_mean_token_accuracy": 0.7954704573984896,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 90.294,
+      "eval_samples_per_second": 15.726,
+      "eval_steps_per_second": 1.971,
+      "step": 100
+    },
+    {
+      "entropy": 0.7979829199612141,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2140507698059082,
+      "learning_rate": 0.00010458300863016666,
+      "loss": 0.7141861915588379,
+      "mean_token_accuracy": 0.8000014744699001,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.7938692486018277,
+      "eval_loss": 0.7279797196388245,
+      "eval_mean_token_accuracy": 0.7989798734027348,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 89.7189,
+      "eval_samples_per_second": 15.827,
+      "eval_steps_per_second": 1.984,
+      "step": 120
+    },
+    {
+      "entropy": 0.7916347607970238,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.0506458282470703,
+      "learning_rate": 0.0001221599848705308,
+      "loss": 0.7117055416107178,
+      "mean_token_accuracy": 0.7970046654343605,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.8052781063519167,
+      "eval_loss": 0.722576916217804,
+      "eval_mean_token_accuracy": 0.7982530523551984,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 90.0376,
+      "eval_samples_per_second": 15.771,
+      "eval_steps_per_second": 1.977,
+      "step": 140
+    },
+    {
+      "entropy": 0.7880666613578796,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8409207463264465,
+      "learning_rate": 0.00013973696111089492,
+      "loss": 0.7052523136138916,
+      "mean_token_accuracy": 0.8000123649835587,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7787443134891853,
+      "eval_loss": 0.7137772440910339,
+      "eval_mean_token_accuracy": 0.8014209287220173,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 90.2562,
+      "eval_samples_per_second": 15.733,
+      "eval_steps_per_second": 1.972,
+      "step": 160
+    },
+    {
+      "entropy": 0.7696007996797561,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.0182477235794067,
+      "learning_rate": 0.00015731393735125907,
+      "loss": 0.7008797645568847,
+      "mean_token_accuracy": 0.8007259473204613,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.752259682068664,
+      "eval_loss": 0.7171286344528198,
+      "eval_mean_token_accuracy": 0.80052122574174,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 90.0797,
+      "eval_samples_per_second": 15.764,
+      "eval_steps_per_second": 1.976,
+      "step": 180
+    },
+    {
+      "entropy": 0.7828126326203346,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9553632736206055,
+      "learning_rate": 0.0001748909135916232,
+      "loss": 0.7046853542327881,
+      "mean_token_accuracy": 0.7988562889397144,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7389638169427936,
+      "eval_loss": 0.7187935709953308,
+      "eval_mean_token_accuracy": 0.7995899999409579,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 89.9223,
+      "eval_samples_per_second": 15.791,
+      "eval_steps_per_second": 1.979,
+      "step": 200
+    },
+    {
+      "entropy": 0.7909286297857762,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 1.134459376335144,
+      "learning_rate": 0.00019246788983198735,
+      "loss": 0.6989312171936035,
+      "mean_token_accuracy": 0.7991565138101577,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7496965199374082,
+      "eval_loss": 0.7106321454048157,
+      "eval_mean_token_accuracy": 0.8046183442131857,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 89.6899,
+      "eval_samples_per_second": 15.832,
+      "eval_steps_per_second": 1.985,
+      "step": 220
+    },
+    {
+      "entropy": 0.7651963323354721,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.9557585716247559,
+      "learning_rate": 0.0002100448660723515,
+      "loss": 0.6984559059143066,
+      "mean_token_accuracy": 0.8019823372364044,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7615563079212488,
+      "eval_loss": 0.7098539471626282,
+      "eval_mean_token_accuracy": 0.8047233508544022,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 90.3752,
+      "eval_samples_per_second": 15.712,
+      "eval_steps_per_second": 1.97,
+      "step": 240
+    },
+    {
+      "entropy": 0.787507726252079,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.9279311299324036,
+      "learning_rate": 0.00022762184231271568,
+      "loss": 0.708138370513916,
+      "mean_token_accuracy": 0.8013308703899383,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7856640896100676,
+      "eval_loss": 0.7191570401191711,
+      "eval_mean_token_accuracy": 0.8024677249153008,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 93.7336,
+      "eval_samples_per_second": 15.149,
+      "eval_steps_per_second": 1.899,
+      "step": 260
+    },
+    {
+      "entropy": 0.768489234894514,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.7992776036262512,
+      "learning_rate": 0.0002451988185530798,
+      "loss": 0.7035849571228028,
+      "mean_token_accuracy": 0.8031105332076549,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7682498740346244,
+      "eval_loss": 0.7149022221565247,
+      "eval_mean_token_accuracy": 0.7973497995499814,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 89.7309,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 280
+    },
+    {
+      "entropy": 0.7720955617725849,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.1931514739990234,
+      "learning_rate": 0.0002627757947934439,
+      "loss": 0.6965402603149414,
+      "mean_token_accuracy": 0.8040348663926125,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7214708304807042,
+      "eval_loss": 0.7350823879241943,
+      "eval_mean_token_accuracy": 0.7942911925610532,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 89.6451,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 300
+    },
+    {
+      "entropy": 0.7930095426738262,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8994467258453369,
+      "learning_rate": 0.0002803527710338081,
+      "loss": 0.7331492900848389,
+      "mean_token_accuracy": 0.796185302734375,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.8142535488926963,
+      "eval_loss": 0.7237834334373474,
+      "eval_mean_token_accuracy": 0.799495273761535,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 90.6759,
+      "eval_samples_per_second": 15.66,
+      "eval_steps_per_second": 1.963,
+      "step": 320
+    },
+    {
+      "entropy": 0.7825102441012859,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 1.8224239349365234,
+      "learning_rate": 0.00029792974727417223,
+      "loss": 0.7331175804138184,
+      "mean_token_accuracy": 0.7977667152881622,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7327135443017724,
+      "eval_loss": 0.7388784289360046,
+      "eval_mean_token_accuracy": 0.8007041252730938,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 89.7692,
+      "eval_samples_per_second": 15.818,
+      "eval_steps_per_second": 1.983,
+      "step": 340
+    },
+    {
+      "entropy": 0.819331557303667,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 1.0407779216766357,
+      "learning_rate": 0.0003155067235145364,
+      "loss": 0.7456695556640625,
+      "mean_token_accuracy": 0.7936980701982975,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.8065585420372781,
+      "eval_loss": 0.735519528388977,
+      "eval_mean_token_accuracy": 0.7974206296245704,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 89.4336,
+      "eval_samples_per_second": 15.878,
+      "eval_steps_per_second": 1.99,
+      "step": 360
+    },
+    {
+      "entropy": 0.8114653021097183,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.1209490299224854,
+      "learning_rate": 0.0003330836997549005,
+      "loss": 0.7340899467468261,
+      "mean_token_accuracy": 0.7914272703230381,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.8025527285056168,
+      "eval_loss": 0.7295576930046082,
+      "eval_mean_token_accuracy": 0.7990576628218876,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 89.9579,
+      "eval_samples_per_second": 15.785,
+      "eval_steps_per_second": 1.979,
+      "step": 380
+    },
+    {
+      "entropy": 0.8042497783899307,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 1.2437798976898193,
+      "learning_rate": 0.0003506606759952647,
+      "loss": 0.7381744384765625,
+      "mean_token_accuracy": 0.7967362694442273,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7717011336530193,
+      "eval_loss": 0.7436173558235168,
+      "eval_mean_token_accuracy": 0.7975849482450592,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 89.8411,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 400
+    },
+    {
+      "entropy": 0.7987750010056929,
+      "epoch": 1.012070006035003,
+      "grad_norm": 1.2390918731689453,
+      "learning_rate": 0.000364721224843344,
+      "loss": 0.7317790031433106,
+      "mean_token_accuracy": 0.7986087551364651,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7333170604170038,
+      "eval_loss": 0.7456948757171631,
+      "eval_mean_token_accuracy": 0.7964897577682238,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 89.7288,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 420
+    },
+    {
+      "entropy": 0.7372976846992969,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.1552740335464478,
+      "learning_rate": 0.00036468510102269575,
+      "loss": 0.6734249114990234,
+      "mean_token_accuracy": 0.805834824591875,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.7130355526891987,
+      "eval_loss": 0.7555333375930786,
+      "eval_mean_token_accuracy": 0.7961652781186479,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 89.8713,
+      "eval_samples_per_second": 15.8,
+      "eval_steps_per_second": 1.981,
+      "step": 440
+    },
+    {
+      "entropy": 0.7465668715536594,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 1.1991469860076904,
+      "learning_rate": 0.0003645973816745066,
+      "loss": 0.6871879577636719,
+      "mean_token_accuracy": 0.8042450234293937,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7772979542110743,
+      "eval_loss": 0.7561826705932617,
+      "eval_mean_token_accuracy": 0.7953876427720102,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 89.8395,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 460
+    },
+    {
+      "entropy": 0.7477855801582336,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 1.325486660003662,
+      "learning_rate": 0.00036445809162231435,
+      "loss": 0.6836830139160156,
+      "mean_token_accuracy": 0.8021619468927383,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.7098110676481483,
+      "eval_loss": 0.758007287979126,
+      "eval_mean_token_accuracy": 0.7909953189030122,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 89.6101,
+      "eval_samples_per_second": 15.846,
+      "eval_steps_per_second": 1.986,
+      "step": 480
+    },
+    {
+      "entropy": 0.7632691070437432,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 1.0122262239456177,
+      "learning_rate": 0.0003642672702835562,
+      "loss": 0.7071213245391845,
+      "mean_token_accuracy": 0.8000680930912495,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7645382720432924,
+      "eval_loss": 0.7492606043815613,
+      "eval_mean_token_accuracy": 0.7976858676149604,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 89.6092,
+      "eval_samples_per_second": 15.847,
+      "eval_steps_per_second": 1.986,
+      "step": 500
+    },
+    {
+      "entropy": 0.7623726457357407,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.367774486541748,
+      "learning_rate": 0.00036402497165841384,
+      "loss": 0.6965175151824952,
+      "mean_token_accuracy": 0.8016501650214195,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7868083715438843,
+      "eval_loss": 0.7481449842453003,
+      "eval_mean_token_accuracy": 0.799330582779445,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 89.8241,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 520
+    },
+    {
+      "entropy": 0.7535207405686378,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 1.213614821434021,
+      "learning_rate": 0.00036373126431453207,
+      "loss": 0.7078310489654541,
+      "mean_token_accuracy": 0.7998907208442688,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7829091950748743,
+      "eval_loss": 0.7416072487831116,
+      "eval_mean_token_accuracy": 0.7969146250339036,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 89.4607,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 540
+    },
+    {
+      "entropy": 0.7527363449335098,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 1.0929982662200928,
+      "learning_rate": 0.00036338623136761495,
+      "loss": 0.7058369636535644,
+      "mean_token_accuracy": 0.7997087433934211,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.728419938114252,
+      "eval_loss": 0.744873046875,
+      "eval_mean_token_accuracy": 0.7966634296299366,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 89.6597,
+      "eval_samples_per_second": 15.838,
+      "eval_steps_per_second": 1.985,
+      "step": 560
+    },
+    {
+      "entropy": 0.7406099259853363,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.409364104270935,
+      "learning_rate": 0.00036298997045790515,
+      "loss": 0.6902852058410645,
+      "mean_token_accuracy": 0.805386807769537,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.6991632758231645,
+      "eval_loss": 0.7480319738388062,
+      "eval_mean_token_accuracy": 0.7970068006033308,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 89.8329,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 580
+    },
+    {
+      "entropy": 0.7459449704736472,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.7053111791610718,
+      "learning_rate": 0.00036254259372255277,
+      "loss": 0.7044640064239502,
+      "mean_token_accuracy": 0.8005018755793571,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7454566660891758,
+      "eval_loss": 0.7551484704017639,
+      "eval_mean_token_accuracy": 0.7897093731365846,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 89.8313,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 600
+    },
+    {
+      "entropy": 0.7621391348540782,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 1.174443244934082,
+      "learning_rate": 0.000362044227763882,
+      "loss": 0.7111573696136475,
+      "mean_token_accuracy": 0.798752411454916,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7664387949397055,
+      "eval_loss": 0.7365804314613342,
+      "eval_mean_token_accuracy": 0.7963101200843126,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 89.9623,
+      "eval_samples_per_second": 15.784,
+      "eval_steps_per_second": 1.979,
+      "step": 620
+    },
+    {
+      "entropy": 0.7424944408237935,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.2898184061050415,
+      "learning_rate": 0.000361495013613564,
+      "loss": 0.6913110733032226,
+      "mean_token_accuracy": 0.8009015507996082,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7255479529332579,
+      "eval_loss": 0.7345991730690002,
+      "eval_mean_token_accuracy": 0.8013592945056015,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 89.817,
+      "eval_samples_per_second": 15.81,
+      "eval_steps_per_second": 1.982,
+      "step": 640
+    },
+    {
+      "entropy": 0.7303367637097835,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.1310476064682007,
+      "learning_rate": 0.00036089510669270666,
+      "loss": 0.6875617027282714,
+      "mean_token_accuracy": 0.8035397171974182,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7101308248016271,
+      "eval_loss": 0.7394917011260986,
+      "eval_mean_token_accuracy": 0.7990792942850777,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 89.7431,
+      "eval_samples_per_second": 15.823,
+      "eval_steps_per_second": 1.983,
+      "step": 660
+    },
+    {
+      "entropy": 0.7387678384780884,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.9645389914512634,
+      "learning_rate": 0.0003602446767678725,
+      "loss": 0.6911204338073731,
+      "mean_token_accuracy": 0.8056333385407924,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7281073737010527,
+      "eval_loss": 0.7376708984375,
+      "eval_mean_token_accuracy": 0.7997356415464637,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 90.0338,
+      "eval_samples_per_second": 15.772,
+      "eval_steps_per_second": 1.977,
+      "step": 680
+    },
+    {
+      "entropy": 0.7449554048478604,
+      "epoch": 1.687990343995172,
+      "grad_norm": 1.9069629907608032,
+      "learning_rate": 0.0003595439079030364,
+      "loss": 0.7042286396026611,
+      "mean_token_accuracy": 0.8018639378249646,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7022633817088738,
+      "eval_loss": 0.7320939898490906,
+      "eval_mean_token_accuracy": 0.8006051541044471,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 89.8948,
+      "eval_samples_per_second": 15.796,
+      "eval_steps_per_second": 1.98,
+      "step": 700
+    },
+    {
+      "entropy": 0.723249051719904,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.8333877325057983,
+      "learning_rate": 0.00035879299840749777,
+      "loss": 0.7043714046478271,
+      "mean_token_accuracy": 0.8028345607221127,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.672436946898364,
+      "eval_loss": 0.713233470916748,
+      "eval_mean_token_accuracy": 0.8047760577684038,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 89.6394,
+      "eval_samples_per_second": 15.841,
+      "eval_steps_per_second": 1.986,
+      "step": 720
+    },
+    {
+      "entropy": 0.730014206469059,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.4580037593841553,
+      "learning_rate": 0.00035799216077976137,
+      "loss": 0.6985176563262939,
+      "mean_token_accuracy": 0.8015547141432762,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.6908873795123582,
+      "eval_loss": 0.7198202013969421,
+      "eval_mean_token_accuracy": 0.8024131682481659,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 89.3771,
+      "eval_samples_per_second": 15.888,
+      "eval_steps_per_second": 1.992,
+      "step": 740
+    },
+    {
+      "entropy": 0.7219605796039105,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.51682710647583,
+      "learning_rate": 0.000357141621647403,
+      "loss": 0.6844424724578857,
+      "mean_token_accuracy": 0.8032813094556331,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7113401109582922,
+      "eval_loss": 0.7195309400558472,
+      "eval_mean_token_accuracy": 0.8019187035185568,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 89.4658,
+      "eval_samples_per_second": 15.872,
+      "eval_steps_per_second": 1.99,
+      "step": 760
+    },
+    {
+      "entropy": 0.729826743900776,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 1.0838000774383545,
+      "learning_rate": 0.0003562416217029361,
+      "loss": 0.685858678817749,
+      "mean_token_accuracy": 0.8050567395985126,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.649370267484965,
+      "eval_loss": 0.7177029252052307,
+      "eval_mean_token_accuracy": 0.8050174043419656,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 89.2287,
+      "eval_samples_per_second": 15.914,
+      "eval_steps_per_second": 1.995,
+      "step": 780
+    },
+    {
+      "entropy": 0.6799364153295755,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 1.1364924907684326,
+      "learning_rate": 0.00035529241563569903,
+      "loss": 0.6714428901672364,
+      "mean_token_accuracy": 0.8087341658771038,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6603799047095052,
+      "eval_loss": 0.7201809287071228,
+      "eval_mean_token_accuracy": 0.802863868769635,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 89.2718,
+      "eval_samples_per_second": 15.906,
+      "eval_steps_per_second": 1.994,
+      "step": 800
+    },
+    {
+      "entropy": 0.7078610375523567,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 1.2191942930221558,
+      "learning_rate": 0.0003542942720597808,
+      "loss": 0.6996590614318847,
+      "mean_token_accuracy": 0.8049512051045895,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.7240545264120852,
+      "eval_loss": 0.7301309108734131,
+      "eval_mean_token_accuracy": 0.8015973166133581,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 89.3438,
+      "eval_samples_per_second": 15.894,
+      "eval_steps_per_second": 1.992,
+      "step": 820
+    },
+    {
+      "entropy": 0.6557391556826505,
+      "epoch": 2.024140012070006,
+      "grad_norm": 1.36203134059906,
+      "learning_rate": 0.0003532474734380064,
+      "loss": 0.6327113628387451,
+      "mean_token_accuracy": 0.8142829847026181,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6573808648613062,
+      "eval_loss": 0.7334365844726562,
+      "eval_mean_token_accuracy": 0.8043633893634496,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 89.3359,
+      "eval_samples_per_second": 15.895,
+      "eval_steps_per_second": 1.992,
+      "step": 840
+    },
+    {
+      "entropy": 0.6409128502011299,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 1.5804955959320068,
+      "learning_rate": 0.0003521523160020035,
+      "loss": 0.6023559093475341,
+      "mean_token_accuracy": 0.8222251623868942,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6983234788594621,
+      "eval_loss": 0.7232135534286499,
+      "eval_mean_token_accuracy": 0.8010654268639811,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 89.2615,
+      "eval_samples_per_second": 15.908,
+      "eval_steps_per_second": 1.994,
+      "step": 860
+    },
+    {
+      "entropy": 0.6151149723678827,
+      "epoch": 2.12070006035003,
+      "grad_norm": 1.1975436210632324,
+      "learning_rate": 0.00035100910966837193,
+      "loss": 0.5803564548492431,
+      "mean_token_accuracy": 0.8256325736641884,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6636555573243773,
+      "eval_loss": 0.7205662727355957,
+      "eval_mean_token_accuracy": 0.8036768369460374,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 89.2957,
+      "eval_samples_per_second": 15.902,
+      "eval_steps_per_second": 1.993,
+      "step": 880
+    },
+    {
+      "entropy": 0.6201727617532015,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9907131195068359,
+      "learning_rate": 0.0003498181779509813,
+      "loss": 0.6101872444152832,
+      "mean_token_accuracy": 0.8196512959897518,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.6473788086617931,
+      "eval_loss": 0.7062397003173828,
+      "eval_mean_token_accuracy": 0.8078362184963869,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 88.6198,
+      "eval_samples_per_second": 16.024,
+      "eval_steps_per_second": 2.009,
+      "step": 900
+    },
+    {
+      "entropy": 0.6097332935780286,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.9055633544921875,
+      "learning_rate": 0.00034857985786942026,
+      "loss": 0.5956094264984131,
+      "mean_token_accuracy": 0.8240575887262821,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.612629868341296,
+      "eval_loss": 0.7106770277023315,
+      "eval_mean_token_accuracy": 0.805324277181304,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 89.4383,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 920
+    },
+    {
+      "entropy": 0.6182701248675585,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.9513736367225647,
+      "learning_rate": 0.00034729449985362404,
+      "loss": 0.6019269466400147,
+      "mean_token_accuracy": 0.8187106013298034,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6437647456533453,
+      "eval_loss": 0.7173502445220947,
+      "eval_mean_token_accuracy": 0.8056439573175451,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.9568,
+      "eval_samples_per_second": 15.963,
+      "eval_steps_per_second": 2.001,
+      "step": 940
+    },
+    {
+      "entropy": 0.6508570514619351,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.9606207013130188,
+      "learning_rate": 0.0003459624676447067,
+      "loss": 0.6296409606933594,
+      "mean_token_accuracy": 0.8167732007801533,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6288247135248077,
+      "eval_loss": 0.7160675525665283,
+      "eval_mean_token_accuracy": 0.8066010689467527,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 89.0044,
+      "eval_samples_per_second": 15.954,
+      "eval_steps_per_second": 2.0,
+      "step": 960
+    },
+    {
+      "entropy": 0.6380081418901682,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.9677181839942932,
+      "learning_rate": 0.000344584138192027,
+      "loss": 0.6210546493530273,
+      "mean_token_accuracy": 0.8186135642230511,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6407805031604981,
+      "eval_loss": 0.7055043578147888,
+      "eval_mean_token_accuracy": 0.8070103157772107,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 89.5606,
+      "eval_samples_per_second": 15.855,
+      "eval_steps_per_second": 1.987,
+      "step": 980
+    },
+    {
+      "entropy": 0.6310251638293266,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 2.1046407222747803,
+      "learning_rate": 0.000343159901546516,
+      "loss": 0.6165830135345459,
+      "mean_token_accuracy": 0.8207855716347694,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6461507470420237,
+      "eval_loss": 0.7292136549949646,
+      "eval_mean_token_accuracy": 0.8030418261383356,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 89.315,
+      "eval_samples_per_second": 15.899,
+      "eval_steps_per_second": 1.993,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6256617344915867,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 1.1768349409103394,
+      "learning_rate": 0.0003416901607502972,
+      "loss": 0.6142709255218506,
+      "mean_token_accuracy": 0.8220669947564602,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6150523993406403,
+      "eval_loss": 0.7019563317298889,
+      "eval_mean_token_accuracy": 0.8094323099998946,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 89.419,
+      "eval_samples_per_second": 15.88,
+      "eval_steps_per_second": 1.991,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6126735735684633,
+      "epoch": 2.506940253470127,
+      "grad_norm": 1.8065496683120728,
+      "learning_rate": 0.00034017533172263055,
+      "loss": 0.6056458473205566,
+      "mean_token_accuracy": 0.8203762136399746,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6544345456562685,
+      "eval_loss": 0.7086517214775085,
+      "eval_mean_token_accuracy": 0.806709805901131,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 88.981,
+      "eval_samples_per_second": 15.958,
+      "eval_steps_per_second": 2.0,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6311754353344441,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 1.0576369762420654,
+      "learning_rate": 0.00033861584314221236,
+      "loss": 0.6176392555236816,
+      "mean_token_accuracy": 0.8177410490810871,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6273799200406235,
+      "eval_loss": 0.708163321018219,
+      "eval_mean_token_accuracy": 0.8032017099053672,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 89.7668,
+      "eval_samples_per_second": 15.819,
+      "eval_steps_per_second": 1.983,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6200249589979648,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.997065007686615,
+      "learning_rate": 0.0003370121363258637,
+      "loss": 0.6203888893127442,
+      "mean_token_accuracy": 0.8196275025606156,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6457576811983344,
+      "eval_loss": 0.6972290277481079,
+      "eval_mean_token_accuracy": 0.8095980849158898,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 89.2027,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 1080
+    },
+    {
+      "entropy": 0.626810473203659,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.865475058555603,
+      "learning_rate": 0.0003353646651036438,
+      "loss": 0.6102585792541504,
+      "mean_token_accuracy": 0.819708751142025,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.6412022431914726,
+      "eval_loss": 0.7036139965057373,
+      "eval_mean_token_accuracy": 0.8070317135098275,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 89.4733,
+      "eval_samples_per_second": 15.871,
+      "eval_steps_per_second": 1.989,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6581619590520859,
+      "epoch": 2.700060350030175,
+      "grad_norm": 1.00307035446167,
+      "learning_rate": 0.00033367389569042064,
+      "loss": 0.6397720336914062,
+      "mean_token_accuracy": 0.81413309648633,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6288736402318719,
+      "eval_loss": 0.6966825723648071,
+      "eval_mean_token_accuracy": 0.8094039669867312,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 88.7341,
+      "eval_samples_per_second": 16.003,
+      "eval_steps_per_second": 2.006,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6399412982165813,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.8589329719543457,
+      "learning_rate": 0.0003319403065539384,
+      "loss": 0.6347338199615479,
+      "mean_token_accuracy": 0.8139534957706929,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.683491913455256,
+      "eval_loss": 0.6984149217605591,
+      "eval_mean_token_accuracy": 0.8083989469522841,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 89.3929,
+      "eval_samples_per_second": 15.885,
+      "eval_steps_per_second": 1.991,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6264835461974144,
+      "epoch": 2.796620398310199,
+      "grad_norm": 0.7807853817939758,
+      "learning_rate": 0.0003301643882794163,
+      "loss": 0.6162629127502441,
+      "mean_token_accuracy": 0.8216292977333068,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6578792159476977,
+      "eval_loss": 0.6921441555023193,
+      "eval_mean_token_accuracy": 0.8094299177775223,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 89.2356,
+      "eval_samples_per_second": 15.913,
+      "eval_steps_per_second": 1.995,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6269129011780024,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 1.1675593852996826,
+      "learning_rate": 0.0003283466434307189,
+      "loss": 0.6235532760620117,
+      "mean_token_accuracy": 0.8188897788524627,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6533248447970058,
+      "eval_loss": 0.696849524974823,
+      "eval_mean_token_accuracy": 0.8102703713968898,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2602,
+      "eval_samples_per_second": 16.089,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.6259873781353236,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.9596486687660217,
+      "learning_rate": 0.00032648758640813655,
+      "loss": 0.6118728160858155,
+      "mean_token_accuracy": 0.8171859130263328,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6212078754821521,
+      "eval_loss": 0.6942870020866394,
+      "eval_mean_token_accuracy": 0.810484343365337,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 89.4375,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 1200
+    },
+    {
+      "entropy": 0.6142537571489811,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 0.9625058770179749,
+      "learning_rate": 0.00032458774330281615,
+      "loss": 0.6060261249542236,
+      "mean_token_accuracy": 0.8223113484680653,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6553665667437436,
+      "eval_loss": 0.6835269331932068,
+      "eval_mean_token_accuracy": 0.8134723968720168,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 89.5092,
+      "eval_samples_per_second": 15.864,
+      "eval_steps_per_second": 1.989,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6301972426474094,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.9657333493232727,
+      "learning_rate": 0.0003226476517478835,
+      "loss": 0.6277759075164795,
+      "mean_token_accuracy": 0.8164266042411328,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6594253118788258,
+      "eval_loss": 0.6820935606956482,
+      "eval_mean_token_accuracy": 0.8132368198941263,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 89.6211,
+      "eval_samples_per_second": 15.844,
+      "eval_steps_per_second": 1.986,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5346100655469027,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8541185259819031,
+      "learning_rate": 0.0003206678607662996,
+      "loss": 0.5261692047119141,
+      "mean_token_accuracy": 0.8382289425119177,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5626645662476507,
+      "eval_loss": 0.7028542160987854,
+      "eval_mean_token_accuracy": 0.8133395572056931,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 89.1267,
+      "eval_samples_per_second": 15.932,
+      "eval_steps_per_second": 1.997,
+      "step": 1260
+    },
+    {
+      "entropy": 0.5001101832836866,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.8208815455436707,
+      "learning_rate": 0.0003186489306154935,
+      "loss": 0.4850132942199707,
+      "mean_token_accuracy": 0.8468983843922615,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5889462046743779,
+      "eval_loss": 0.691015899181366,
+      "eval_mean_token_accuracy": 0.8135226466012805,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.9677,
+      "eval_samples_per_second": 15.961,
+      "eval_steps_per_second": 2.001,
+      "step": 1280
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.2835248707651174e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1300/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1300/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1300/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1300/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1300/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..7ce4b072fcbe5b636867e13937c56fedee3b68b4
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1300/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.02728394274790723,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1300/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1300/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1300/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1300/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1300/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1300/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1300/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1300/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..2f0affb86e303161e5006b41ff37058a58fcc41e
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1300/trainer_state.json
@@ -0,0 +1,1399 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 3.1327700663850333,
+  "eval_steps": 20,
+  "global_step": 1300,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.024795526266098,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.945375919342041,
+      "learning_rate": 1.6698127428345936e-05,
+      "loss": 1.7825420379638672,
+      "mean_token_accuracy": 0.6315708436071873,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.29921497588747,
+      "eval_loss": 1.169131875038147,
+      "eval_mean_token_accuracy": 0.7185706342204233,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 91.7898,
+      "eval_samples_per_second": 15.47,
+      "eval_steps_per_second": 1.939,
+      "step": 20
+    },
+    {
+      "entropy": 1.0346344470977784,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.7916979789733887,
+      "learning_rate": 3.427510366871008e-05,
+      "loss": 0.9480395317077637,
+      "mean_token_accuracy": 0.7543429024517536,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 0.9673695430326997,
+      "eval_loss": 0.8588430285453796,
+      "eval_mean_token_accuracy": 0.7737233002534073,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 89.8219,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 40
+    },
+    {
+      "entropy": 0.8977566614747048,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7972187995910645,
+      "learning_rate": 5.185207990907422e-05,
+      "loss": 0.8029604911804199,
+      "mean_token_accuracy": 0.7818016350269318,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8436836982041263,
+      "eval_loss": 0.7893033623695374,
+      "eval_mean_token_accuracy": 0.7858914984076211,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 89.9349,
+      "eval_samples_per_second": 15.789,
+      "eval_steps_per_second": 1.979,
+      "step": 60
+    },
+    {
+      "entropy": 0.8650321319699288,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.2128729820251465,
+      "learning_rate": 6.942905614943836e-05,
+      "loss": 0.77417893409729,
+      "mean_token_accuracy": 0.7875479467213153,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8209857180547179,
+      "eval_loss": 0.7572280168533325,
+      "eval_mean_token_accuracy": 0.7925970383574453,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 89.9459,
+      "eval_samples_per_second": 15.787,
+      "eval_steps_per_second": 1.979,
+      "step": 80
+    },
+    {
+      "entropy": 0.8096660938113928,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.1254287958145142,
+      "learning_rate": 8.700603238980251e-05,
+      "loss": 0.7170331001281738,
+      "mean_token_accuracy": 0.7994555257260799,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.792780208788561,
+      "eval_loss": 0.7442984580993652,
+      "eval_mean_token_accuracy": 0.7954704573984896,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 90.294,
+      "eval_samples_per_second": 15.726,
+      "eval_steps_per_second": 1.971,
+      "step": 100
+    },
+    {
+      "entropy": 0.7979829199612141,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2140507698059082,
+      "learning_rate": 0.00010458300863016666,
+      "loss": 0.7141861915588379,
+      "mean_token_accuracy": 0.8000014744699001,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.7938692486018277,
+      "eval_loss": 0.7279797196388245,
+      "eval_mean_token_accuracy": 0.7989798734027348,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 89.7189,
+      "eval_samples_per_second": 15.827,
+      "eval_steps_per_second": 1.984,
+      "step": 120
+    },
+    {
+      "entropy": 0.7916347607970238,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.0506458282470703,
+      "learning_rate": 0.0001221599848705308,
+      "loss": 0.7117055416107178,
+      "mean_token_accuracy": 0.7970046654343605,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.8052781063519167,
+      "eval_loss": 0.722576916217804,
+      "eval_mean_token_accuracy": 0.7982530523551984,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 90.0376,
+      "eval_samples_per_second": 15.771,
+      "eval_steps_per_second": 1.977,
+      "step": 140
+    },
+    {
+      "entropy": 0.7880666613578796,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8409207463264465,
+      "learning_rate": 0.00013973696111089492,
+      "loss": 0.7052523136138916,
+      "mean_token_accuracy": 0.8000123649835587,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7787443134891853,
+      "eval_loss": 0.7137772440910339,
+      "eval_mean_token_accuracy": 0.8014209287220173,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 90.2562,
+      "eval_samples_per_second": 15.733,
+      "eval_steps_per_second": 1.972,
+      "step": 160
+    },
+    {
+      "entropy": 0.7696007996797561,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.0182477235794067,
+      "learning_rate": 0.00015731393735125907,
+      "loss": 0.7008797645568847,
+      "mean_token_accuracy": 0.8007259473204613,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.752259682068664,
+      "eval_loss": 0.7171286344528198,
+      "eval_mean_token_accuracy": 0.80052122574174,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 90.0797,
+      "eval_samples_per_second": 15.764,
+      "eval_steps_per_second": 1.976,
+      "step": 180
+    },
+    {
+      "entropy": 0.7828126326203346,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9553632736206055,
+      "learning_rate": 0.0001748909135916232,
+      "loss": 0.7046853542327881,
+      "mean_token_accuracy": 0.7988562889397144,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7389638169427936,
+      "eval_loss": 0.7187935709953308,
+      "eval_mean_token_accuracy": 0.7995899999409579,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 89.9223,
+      "eval_samples_per_second": 15.791,
+      "eval_steps_per_second": 1.979,
+      "step": 200
+    },
+    {
+      "entropy": 0.7909286297857762,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 1.134459376335144,
+      "learning_rate": 0.00019246788983198735,
+      "loss": 0.6989312171936035,
+      "mean_token_accuracy": 0.7991565138101577,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7496965199374082,
+      "eval_loss": 0.7106321454048157,
+      "eval_mean_token_accuracy": 0.8046183442131857,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 89.6899,
+      "eval_samples_per_second": 15.832,
+      "eval_steps_per_second": 1.985,
+      "step": 220
+    },
+    {
+      "entropy": 0.7651963323354721,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.9557585716247559,
+      "learning_rate": 0.0002100448660723515,
+      "loss": 0.6984559059143066,
+      "mean_token_accuracy": 0.8019823372364044,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7615563079212488,
+      "eval_loss": 0.7098539471626282,
+      "eval_mean_token_accuracy": 0.8047233508544022,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 90.3752,
+      "eval_samples_per_second": 15.712,
+      "eval_steps_per_second": 1.97,
+      "step": 240
+    },
+    {
+      "entropy": 0.787507726252079,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.9279311299324036,
+      "learning_rate": 0.00022762184231271568,
+      "loss": 0.708138370513916,
+      "mean_token_accuracy": 0.8013308703899383,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7856640896100676,
+      "eval_loss": 0.7191570401191711,
+      "eval_mean_token_accuracy": 0.8024677249153008,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 93.7336,
+      "eval_samples_per_second": 15.149,
+      "eval_steps_per_second": 1.899,
+      "step": 260
+    },
+    {
+      "entropy": 0.768489234894514,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.7992776036262512,
+      "learning_rate": 0.0002451988185530798,
+      "loss": 0.7035849571228028,
+      "mean_token_accuracy": 0.8031105332076549,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7682498740346244,
+      "eval_loss": 0.7149022221565247,
+      "eval_mean_token_accuracy": 0.7973497995499814,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 89.7309,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 280
+    },
+    {
+      "entropy": 0.7720955617725849,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.1931514739990234,
+      "learning_rate": 0.0002627757947934439,
+      "loss": 0.6965402603149414,
+      "mean_token_accuracy": 0.8040348663926125,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7214708304807042,
+      "eval_loss": 0.7350823879241943,
+      "eval_mean_token_accuracy": 0.7942911925610532,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 89.6451,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 300
+    },
+    {
+      "entropy": 0.7930095426738262,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8994467258453369,
+      "learning_rate": 0.0002803527710338081,
+      "loss": 0.7331492900848389,
+      "mean_token_accuracy": 0.796185302734375,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.8142535488926963,
+      "eval_loss": 0.7237834334373474,
+      "eval_mean_token_accuracy": 0.799495273761535,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 90.6759,
+      "eval_samples_per_second": 15.66,
+      "eval_steps_per_second": 1.963,
+      "step": 320
+    },
+    {
+      "entropy": 0.7825102441012859,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 1.8224239349365234,
+      "learning_rate": 0.00029792974727417223,
+      "loss": 0.7331175804138184,
+      "mean_token_accuracy": 0.7977667152881622,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7327135443017724,
+      "eval_loss": 0.7388784289360046,
+      "eval_mean_token_accuracy": 0.8007041252730938,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 89.7692,
+      "eval_samples_per_second": 15.818,
+      "eval_steps_per_second": 1.983,
+      "step": 340
+    },
+    {
+      "entropy": 0.819331557303667,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 1.0407779216766357,
+      "learning_rate": 0.0003155067235145364,
+      "loss": 0.7456695556640625,
+      "mean_token_accuracy": 0.7936980701982975,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.8065585420372781,
+      "eval_loss": 0.735519528388977,
+      "eval_mean_token_accuracy": 0.7974206296245704,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 89.4336,
+      "eval_samples_per_second": 15.878,
+      "eval_steps_per_second": 1.99,
+      "step": 360
+    },
+    {
+      "entropy": 0.8114653021097183,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.1209490299224854,
+      "learning_rate": 0.0003330836997549005,
+      "loss": 0.7340899467468261,
+      "mean_token_accuracy": 0.7914272703230381,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.8025527285056168,
+      "eval_loss": 0.7295576930046082,
+      "eval_mean_token_accuracy": 0.7990576628218876,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 89.9579,
+      "eval_samples_per_second": 15.785,
+      "eval_steps_per_second": 1.979,
+      "step": 380
+    },
+    {
+      "entropy": 0.8042497783899307,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 1.2437798976898193,
+      "learning_rate": 0.0003506606759952647,
+      "loss": 0.7381744384765625,
+      "mean_token_accuracy": 0.7967362694442273,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7717011336530193,
+      "eval_loss": 0.7436173558235168,
+      "eval_mean_token_accuracy": 0.7975849482450592,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 89.8411,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 400
+    },
+    {
+      "entropy": 0.7987750010056929,
+      "epoch": 1.012070006035003,
+      "grad_norm": 1.2390918731689453,
+      "learning_rate": 0.000364721224843344,
+      "loss": 0.7317790031433106,
+      "mean_token_accuracy": 0.7986087551364651,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7333170604170038,
+      "eval_loss": 0.7456948757171631,
+      "eval_mean_token_accuracy": 0.7964897577682238,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 89.7288,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 420
+    },
+    {
+      "entropy": 0.7372976846992969,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.1552740335464478,
+      "learning_rate": 0.00036468510102269575,
+      "loss": 0.6734249114990234,
+      "mean_token_accuracy": 0.805834824591875,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.7130355526891987,
+      "eval_loss": 0.7555333375930786,
+      "eval_mean_token_accuracy": 0.7961652781186479,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 89.8713,
+      "eval_samples_per_second": 15.8,
+      "eval_steps_per_second": 1.981,
+      "step": 440
+    },
+    {
+      "entropy": 0.7465668715536594,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 1.1991469860076904,
+      "learning_rate": 0.0003645973816745066,
+      "loss": 0.6871879577636719,
+      "mean_token_accuracy": 0.8042450234293937,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7772979542110743,
+      "eval_loss": 0.7561826705932617,
+      "eval_mean_token_accuracy": 0.7953876427720102,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 89.8395,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 460
+    },
+    {
+      "entropy": 0.7477855801582336,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 1.325486660003662,
+      "learning_rate": 0.00036445809162231435,
+      "loss": 0.6836830139160156,
+      "mean_token_accuracy": 0.8021619468927383,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.7098110676481483,
+      "eval_loss": 0.758007287979126,
+      "eval_mean_token_accuracy": 0.7909953189030122,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 89.6101,
+      "eval_samples_per_second": 15.846,
+      "eval_steps_per_second": 1.986,
+      "step": 480
+    },
+    {
+      "entropy": 0.7632691070437432,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 1.0122262239456177,
+      "learning_rate": 0.0003642672702835562,
+      "loss": 0.7071213245391845,
+      "mean_token_accuracy": 0.8000680930912495,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7645382720432924,
+      "eval_loss": 0.7492606043815613,
+      "eval_mean_token_accuracy": 0.7976858676149604,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 89.6092,
+      "eval_samples_per_second": 15.847,
+      "eval_steps_per_second": 1.986,
+      "step": 500
+    },
+    {
+      "entropy": 0.7623726457357407,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.367774486541748,
+      "learning_rate": 0.00036402497165841384,
+      "loss": 0.6965175151824952,
+      "mean_token_accuracy": 0.8016501650214195,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7868083715438843,
+      "eval_loss": 0.7481449842453003,
+      "eval_mean_token_accuracy": 0.799330582779445,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 89.8241,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 520
+    },
+    {
+      "entropy": 0.7535207405686378,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 1.213614821434021,
+      "learning_rate": 0.00036373126431453207,
+      "loss": 0.7078310489654541,
+      "mean_token_accuracy": 0.7998907208442688,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7829091950748743,
+      "eval_loss": 0.7416072487831116,
+      "eval_mean_token_accuracy": 0.7969146250339036,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 89.4607,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 540
+    },
+    {
+      "entropy": 0.7527363449335098,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 1.0929982662200928,
+      "learning_rate": 0.00036338623136761495,
+      "loss": 0.7058369636535644,
+      "mean_token_accuracy": 0.7997087433934211,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.728419938114252,
+      "eval_loss": 0.744873046875,
+      "eval_mean_token_accuracy": 0.7966634296299366,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 89.6597,
+      "eval_samples_per_second": 15.838,
+      "eval_steps_per_second": 1.985,
+      "step": 560
+    },
+    {
+      "entropy": 0.7406099259853363,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.409364104270935,
+      "learning_rate": 0.00036298997045790515,
+      "loss": 0.6902852058410645,
+      "mean_token_accuracy": 0.805386807769537,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.6991632758231645,
+      "eval_loss": 0.7480319738388062,
+      "eval_mean_token_accuracy": 0.7970068006033308,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 89.8329,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 580
+    },
+    {
+      "entropy": 0.7459449704736472,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.7053111791610718,
+      "learning_rate": 0.00036254259372255277,
+      "loss": 0.7044640064239502,
+      "mean_token_accuracy": 0.8005018755793571,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7454566660891758,
+      "eval_loss": 0.7551484704017639,
+      "eval_mean_token_accuracy": 0.7897093731365846,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 89.8313,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 600
+    },
+    {
+      "entropy": 0.7621391348540782,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 1.174443244934082,
+      "learning_rate": 0.000362044227763882,
+      "loss": 0.7111573696136475,
+      "mean_token_accuracy": 0.798752411454916,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7664387949397055,
+      "eval_loss": 0.7365804314613342,
+      "eval_mean_token_accuracy": 0.7963101200843126,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 89.9623,
+      "eval_samples_per_second": 15.784,
+      "eval_steps_per_second": 1.979,
+      "step": 620
+    },
+    {
+      "entropy": 0.7424944408237935,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.2898184061050415,
+      "learning_rate": 0.000361495013613564,
+      "loss": 0.6913110733032226,
+      "mean_token_accuracy": 0.8009015507996082,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7255479529332579,
+      "eval_loss": 0.7345991730690002,
+      "eval_mean_token_accuracy": 0.8013592945056015,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 89.817,
+      "eval_samples_per_second": 15.81,
+      "eval_steps_per_second": 1.982,
+      "step": 640
+    },
+    {
+      "entropy": 0.7303367637097835,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.1310476064682007,
+      "learning_rate": 0.00036089510669270666,
+      "loss": 0.6875617027282714,
+      "mean_token_accuracy": 0.8035397171974182,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7101308248016271,
+      "eval_loss": 0.7394917011260986,
+      "eval_mean_token_accuracy": 0.7990792942850777,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 89.7431,
+      "eval_samples_per_second": 15.823,
+      "eval_steps_per_second": 1.983,
+      "step": 660
+    },
+    {
+      "entropy": 0.7387678384780884,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.9645389914512634,
+      "learning_rate": 0.0003602446767678725,
+      "loss": 0.6911204338073731,
+      "mean_token_accuracy": 0.8056333385407924,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7281073737010527,
+      "eval_loss": 0.7376708984375,
+      "eval_mean_token_accuracy": 0.7997356415464637,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 90.0338,
+      "eval_samples_per_second": 15.772,
+      "eval_steps_per_second": 1.977,
+      "step": 680
+    },
+    {
+      "entropy": 0.7449554048478604,
+      "epoch": 1.687990343995172,
+      "grad_norm": 1.9069629907608032,
+      "learning_rate": 0.0003595439079030364,
+      "loss": 0.7042286396026611,
+      "mean_token_accuracy": 0.8018639378249646,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7022633817088738,
+      "eval_loss": 0.7320939898490906,
+      "eval_mean_token_accuracy": 0.8006051541044471,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 89.8948,
+      "eval_samples_per_second": 15.796,
+      "eval_steps_per_second": 1.98,
+      "step": 700
+    },
+    {
+      "entropy": 0.723249051719904,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.8333877325057983,
+      "learning_rate": 0.00035879299840749777,
+      "loss": 0.7043714046478271,
+      "mean_token_accuracy": 0.8028345607221127,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.672436946898364,
+      "eval_loss": 0.713233470916748,
+      "eval_mean_token_accuracy": 0.8047760577684038,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 89.6394,
+      "eval_samples_per_second": 15.841,
+      "eval_steps_per_second": 1.986,
+      "step": 720
+    },
+    {
+      "entropy": 0.730014206469059,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.4580037593841553,
+      "learning_rate": 0.00035799216077976137,
+      "loss": 0.6985176563262939,
+      "mean_token_accuracy": 0.8015547141432762,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.6908873795123582,
+      "eval_loss": 0.7198202013969421,
+      "eval_mean_token_accuracy": 0.8024131682481659,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 89.3771,
+      "eval_samples_per_second": 15.888,
+      "eval_steps_per_second": 1.992,
+      "step": 740
+    },
+    {
+      "entropy": 0.7219605796039105,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.51682710647583,
+      "learning_rate": 0.000357141621647403,
+      "loss": 0.6844424724578857,
+      "mean_token_accuracy": 0.8032813094556331,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7113401109582922,
+      "eval_loss": 0.7195309400558472,
+      "eval_mean_token_accuracy": 0.8019187035185568,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 89.4658,
+      "eval_samples_per_second": 15.872,
+      "eval_steps_per_second": 1.99,
+      "step": 760
+    },
+    {
+      "entropy": 0.729826743900776,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 1.0838000774383545,
+      "learning_rate": 0.0003562416217029361,
+      "loss": 0.685858678817749,
+      "mean_token_accuracy": 0.8050567395985126,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.649370267484965,
+      "eval_loss": 0.7177029252052307,
+      "eval_mean_token_accuracy": 0.8050174043419656,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 89.2287,
+      "eval_samples_per_second": 15.914,
+      "eval_steps_per_second": 1.995,
+      "step": 780
+    },
+    {
+      "entropy": 0.6799364153295755,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 1.1364924907684326,
+      "learning_rate": 0.00035529241563569903,
+      "loss": 0.6714428901672364,
+      "mean_token_accuracy": 0.8087341658771038,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6603799047095052,
+      "eval_loss": 0.7201809287071228,
+      "eval_mean_token_accuracy": 0.802863868769635,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 89.2718,
+      "eval_samples_per_second": 15.906,
+      "eval_steps_per_second": 1.994,
+      "step": 800
+    },
+    {
+      "entropy": 0.7078610375523567,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 1.2191942930221558,
+      "learning_rate": 0.0003542942720597808,
+      "loss": 0.6996590614318847,
+      "mean_token_accuracy": 0.8049512051045895,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.7240545264120852,
+      "eval_loss": 0.7301309108734131,
+      "eval_mean_token_accuracy": 0.8015973166133581,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 89.3438,
+      "eval_samples_per_second": 15.894,
+      "eval_steps_per_second": 1.992,
+      "step": 820
+    },
+    {
+      "entropy": 0.6557391556826505,
+      "epoch": 2.024140012070006,
+      "grad_norm": 1.36203134059906,
+      "learning_rate": 0.0003532474734380064,
+      "loss": 0.6327113628387451,
+      "mean_token_accuracy": 0.8142829847026181,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6573808648613062,
+      "eval_loss": 0.7334365844726562,
+      "eval_mean_token_accuracy": 0.8043633893634496,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 89.3359,
+      "eval_samples_per_second": 15.895,
+      "eval_steps_per_second": 1.992,
+      "step": 840
+    },
+    {
+      "entropy": 0.6409128502011299,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 1.5804955959320068,
+      "learning_rate": 0.0003521523160020035,
+      "loss": 0.6023559093475341,
+      "mean_token_accuracy": 0.8222251623868942,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6983234788594621,
+      "eval_loss": 0.7232135534286499,
+      "eval_mean_token_accuracy": 0.8010654268639811,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 89.2615,
+      "eval_samples_per_second": 15.908,
+      "eval_steps_per_second": 1.994,
+      "step": 860
+    },
+    {
+      "entropy": 0.6151149723678827,
+      "epoch": 2.12070006035003,
+      "grad_norm": 1.1975436210632324,
+      "learning_rate": 0.00035100910966837193,
+      "loss": 0.5803564548492431,
+      "mean_token_accuracy": 0.8256325736641884,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6636555573243773,
+      "eval_loss": 0.7205662727355957,
+      "eval_mean_token_accuracy": 0.8036768369460374,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 89.2957,
+      "eval_samples_per_second": 15.902,
+      "eval_steps_per_second": 1.993,
+      "step": 880
+    },
+    {
+      "entropy": 0.6201727617532015,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9907131195068359,
+      "learning_rate": 0.0003498181779509813,
+      "loss": 0.6101872444152832,
+      "mean_token_accuracy": 0.8196512959897518,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.6473788086617931,
+      "eval_loss": 0.7062397003173828,
+      "eval_mean_token_accuracy": 0.8078362184963869,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 88.6198,
+      "eval_samples_per_second": 16.024,
+      "eval_steps_per_second": 2.009,
+      "step": 900
+    },
+    {
+      "entropy": 0.6097332935780286,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.9055633544921875,
+      "learning_rate": 0.00034857985786942026,
+      "loss": 0.5956094264984131,
+      "mean_token_accuracy": 0.8240575887262821,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.612629868341296,
+      "eval_loss": 0.7106770277023315,
+      "eval_mean_token_accuracy": 0.805324277181304,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 89.4383,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 920
+    },
+    {
+      "entropy": 0.6182701248675585,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.9513736367225647,
+      "learning_rate": 0.00034729449985362404,
+      "loss": 0.6019269466400147,
+      "mean_token_accuracy": 0.8187106013298034,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6437647456533453,
+      "eval_loss": 0.7173502445220947,
+      "eval_mean_token_accuracy": 0.8056439573175451,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.9568,
+      "eval_samples_per_second": 15.963,
+      "eval_steps_per_second": 2.001,
+      "step": 940
+    },
+    {
+      "entropy": 0.6508570514619351,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.9606207013130188,
+      "learning_rate": 0.0003459624676447067,
+      "loss": 0.6296409606933594,
+      "mean_token_accuracy": 0.8167732007801533,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6288247135248077,
+      "eval_loss": 0.7160675525665283,
+      "eval_mean_token_accuracy": 0.8066010689467527,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 89.0044,
+      "eval_samples_per_second": 15.954,
+      "eval_steps_per_second": 2.0,
+      "step": 960
+    },
+    {
+      "entropy": 0.6380081418901682,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.9677181839942932,
+      "learning_rate": 0.000344584138192027,
+      "loss": 0.6210546493530273,
+      "mean_token_accuracy": 0.8186135642230511,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6407805031604981,
+      "eval_loss": 0.7055043578147888,
+      "eval_mean_token_accuracy": 0.8070103157772107,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 89.5606,
+      "eval_samples_per_second": 15.855,
+      "eval_steps_per_second": 1.987,
+      "step": 980
+    },
+    {
+      "entropy": 0.6310251638293266,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 2.1046407222747803,
+      "learning_rate": 0.000343159901546516,
+      "loss": 0.6165830135345459,
+      "mean_token_accuracy": 0.8207855716347694,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6461507470420237,
+      "eval_loss": 0.7292136549949646,
+      "eval_mean_token_accuracy": 0.8030418261383356,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 89.315,
+      "eval_samples_per_second": 15.899,
+      "eval_steps_per_second": 1.993,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6256617344915867,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 1.1768349409103394,
+      "learning_rate": 0.0003416901607502972,
+      "loss": 0.6142709255218506,
+      "mean_token_accuracy": 0.8220669947564602,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6150523993406403,
+      "eval_loss": 0.7019563317298889,
+      "eval_mean_token_accuracy": 0.8094323099998946,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 89.419,
+      "eval_samples_per_second": 15.88,
+      "eval_steps_per_second": 1.991,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6126735735684633,
+      "epoch": 2.506940253470127,
+      "grad_norm": 1.8065496683120728,
+      "learning_rate": 0.00034017533172263055,
+      "loss": 0.6056458473205566,
+      "mean_token_accuracy": 0.8203762136399746,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6544345456562685,
+      "eval_loss": 0.7086517214775085,
+      "eval_mean_token_accuracy": 0.806709805901131,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 88.981,
+      "eval_samples_per_second": 15.958,
+      "eval_steps_per_second": 2.0,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6311754353344441,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 1.0576369762420654,
+      "learning_rate": 0.00033861584314221236,
+      "loss": 0.6176392555236816,
+      "mean_token_accuracy": 0.8177410490810871,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6273799200406235,
+      "eval_loss": 0.708163321018219,
+      "eval_mean_token_accuracy": 0.8032017099053672,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 89.7668,
+      "eval_samples_per_second": 15.819,
+      "eval_steps_per_second": 1.983,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6200249589979648,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.997065007686615,
+      "learning_rate": 0.0003370121363258637,
+      "loss": 0.6203888893127442,
+      "mean_token_accuracy": 0.8196275025606156,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6457576811983344,
+      "eval_loss": 0.6972290277481079,
+      "eval_mean_token_accuracy": 0.8095980849158898,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 89.2027,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 1080
+    },
+    {
+      "entropy": 0.626810473203659,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.865475058555603,
+      "learning_rate": 0.0003353646651036438,
+      "loss": 0.6102585792541504,
+      "mean_token_accuracy": 0.819708751142025,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.6412022431914726,
+      "eval_loss": 0.7036139965057373,
+      "eval_mean_token_accuracy": 0.8070317135098275,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 89.4733,
+      "eval_samples_per_second": 15.871,
+      "eval_steps_per_second": 1.989,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6581619590520859,
+      "epoch": 2.700060350030175,
+      "grad_norm": 1.00307035446167,
+      "learning_rate": 0.00033367389569042064,
+      "loss": 0.6397720336914062,
+      "mean_token_accuracy": 0.81413309648633,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6288736402318719,
+      "eval_loss": 0.6966825723648071,
+      "eval_mean_token_accuracy": 0.8094039669867312,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 88.7341,
+      "eval_samples_per_second": 16.003,
+      "eval_steps_per_second": 2.006,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6399412982165813,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.8589329719543457,
+      "learning_rate": 0.0003319403065539384,
+      "loss": 0.6347338199615479,
+      "mean_token_accuracy": 0.8139534957706929,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.683491913455256,
+      "eval_loss": 0.6984149217605591,
+      "eval_mean_token_accuracy": 0.8083989469522841,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 89.3929,
+      "eval_samples_per_second": 15.885,
+      "eval_steps_per_second": 1.991,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6264835461974144,
+      "epoch": 2.796620398310199,
+      "grad_norm": 0.7807853817939758,
+      "learning_rate": 0.0003301643882794163,
+      "loss": 0.6162629127502441,
+      "mean_token_accuracy": 0.8216292977333068,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6578792159476977,
+      "eval_loss": 0.6921441555023193,
+      "eval_mean_token_accuracy": 0.8094299177775223,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 89.2356,
+      "eval_samples_per_second": 15.913,
+      "eval_steps_per_second": 1.995,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6269129011780024,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 1.1675593852996826,
+      "learning_rate": 0.0003283466434307189,
+      "loss": 0.6235532760620117,
+      "mean_token_accuracy": 0.8188897788524627,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6533248447970058,
+      "eval_loss": 0.696849524974823,
+      "eval_mean_token_accuracy": 0.8102703713968898,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2602,
+      "eval_samples_per_second": 16.089,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.6259873781353236,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.9596486687660217,
+      "learning_rate": 0.00032648758640813655,
+      "loss": 0.6118728160858155,
+      "mean_token_accuracy": 0.8171859130263328,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6212078754821521,
+      "eval_loss": 0.6942870020866394,
+      "eval_mean_token_accuracy": 0.810484343365337,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 89.4375,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 1200
+    },
+    {
+      "entropy": 0.6142537571489811,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 0.9625058770179749,
+      "learning_rate": 0.00032458774330281615,
+      "loss": 0.6060261249542236,
+      "mean_token_accuracy": 0.8223113484680653,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6553665667437436,
+      "eval_loss": 0.6835269331932068,
+      "eval_mean_token_accuracy": 0.8134723968720168,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 89.5092,
+      "eval_samples_per_second": 15.864,
+      "eval_steps_per_second": 1.989,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6301972426474094,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.9657333493232727,
+      "learning_rate": 0.0003226476517478835,
+      "loss": 0.6277759075164795,
+      "mean_token_accuracy": 0.8164266042411328,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6594253118788258,
+      "eval_loss": 0.6820935606956482,
+      "eval_mean_token_accuracy": 0.8132368198941263,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 89.6211,
+      "eval_samples_per_second": 15.844,
+      "eval_steps_per_second": 1.986,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5346100655469027,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8541185259819031,
+      "learning_rate": 0.0003206678607662996,
+      "loss": 0.5261692047119141,
+      "mean_token_accuracy": 0.8382289425119177,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5626645662476507,
+      "eval_loss": 0.7028542160987854,
+      "eval_mean_token_accuracy": 0.8133395572056931,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 89.1267,
+      "eval_samples_per_second": 15.932,
+      "eval_steps_per_second": 1.997,
+      "step": 1260
+    },
+    {
+      "entropy": 0.5001101832836866,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.8208815455436707,
+      "learning_rate": 0.0003186489306154935,
+      "loss": 0.4850132942199707,
+      "mean_token_accuracy": 0.8468983843922615,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5889462046743779,
+      "eval_loss": 0.691015899181366,
+      "eval_mean_token_accuracy": 0.8135226466012805,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.9677,
+      "eval_samples_per_second": 15.961,
+      "eval_steps_per_second": 2.001,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5242766339331866,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 0.7893072962760925,
+      "learning_rate": 0.0003165914326288163,
+      "loss": 0.5061577320098877,
+      "mean_token_accuracy": 0.8425608821213245,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5543691962957382,
+      "eval_loss": 0.6913734078407288,
+      "eval_mean_token_accuracy": 0.8154889680026622,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 89.6963,
+      "eval_samples_per_second": 15.831,
+      "eval_steps_per_second": 1.984,
+      "step": 1300
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.3048837115230618e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1320/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1320/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1320/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1320/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1320/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..7ce4b072fcbe5b636867e13937c56fedee3b68b4
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1320/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.02728394274790723,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1320/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1320/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1320/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1320/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1320/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1320/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1320/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1320/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..dd41560cc02589cf90577156d8b7c800a2e5af28
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1320/trainer_state.json
@@ -0,0 +1,1420 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 3.181050090525045,
+  "eval_steps": 20,
+  "global_step": 1320,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.024795526266098,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.945375919342041,
+      "learning_rate": 1.6698127428345936e-05,
+      "loss": 1.7825420379638672,
+      "mean_token_accuracy": 0.6315708436071873,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.29921497588747,
+      "eval_loss": 1.169131875038147,
+      "eval_mean_token_accuracy": 0.7185706342204233,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 91.7898,
+      "eval_samples_per_second": 15.47,
+      "eval_steps_per_second": 1.939,
+      "step": 20
+    },
+    {
+      "entropy": 1.0346344470977784,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.7916979789733887,
+      "learning_rate": 3.427510366871008e-05,
+      "loss": 0.9480395317077637,
+      "mean_token_accuracy": 0.7543429024517536,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 0.9673695430326997,
+      "eval_loss": 0.8588430285453796,
+      "eval_mean_token_accuracy": 0.7737233002534073,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 89.8219,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 40
+    },
+    {
+      "entropy": 0.8977566614747048,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7972187995910645,
+      "learning_rate": 5.185207990907422e-05,
+      "loss": 0.8029604911804199,
+      "mean_token_accuracy": 0.7818016350269318,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8436836982041263,
+      "eval_loss": 0.7893033623695374,
+      "eval_mean_token_accuracy": 0.7858914984076211,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 89.9349,
+      "eval_samples_per_second": 15.789,
+      "eval_steps_per_second": 1.979,
+      "step": 60
+    },
+    {
+      "entropy": 0.8650321319699288,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.2128729820251465,
+      "learning_rate": 6.942905614943836e-05,
+      "loss": 0.77417893409729,
+      "mean_token_accuracy": 0.7875479467213153,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8209857180547179,
+      "eval_loss": 0.7572280168533325,
+      "eval_mean_token_accuracy": 0.7925970383574453,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 89.9459,
+      "eval_samples_per_second": 15.787,
+      "eval_steps_per_second": 1.979,
+      "step": 80
+    },
+    {
+      "entropy": 0.8096660938113928,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.1254287958145142,
+      "learning_rate": 8.700603238980251e-05,
+      "loss": 0.7170331001281738,
+      "mean_token_accuracy": 0.7994555257260799,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.792780208788561,
+      "eval_loss": 0.7442984580993652,
+      "eval_mean_token_accuracy": 0.7954704573984896,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 90.294,
+      "eval_samples_per_second": 15.726,
+      "eval_steps_per_second": 1.971,
+      "step": 100
+    },
+    {
+      "entropy": 0.7979829199612141,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2140507698059082,
+      "learning_rate": 0.00010458300863016666,
+      "loss": 0.7141861915588379,
+      "mean_token_accuracy": 0.8000014744699001,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.7938692486018277,
+      "eval_loss": 0.7279797196388245,
+      "eval_mean_token_accuracy": 0.7989798734027348,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 89.7189,
+      "eval_samples_per_second": 15.827,
+      "eval_steps_per_second": 1.984,
+      "step": 120
+    },
+    {
+      "entropy": 0.7916347607970238,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.0506458282470703,
+      "learning_rate": 0.0001221599848705308,
+      "loss": 0.7117055416107178,
+      "mean_token_accuracy": 0.7970046654343605,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.8052781063519167,
+      "eval_loss": 0.722576916217804,
+      "eval_mean_token_accuracy": 0.7982530523551984,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 90.0376,
+      "eval_samples_per_second": 15.771,
+      "eval_steps_per_second": 1.977,
+      "step": 140
+    },
+    {
+      "entropy": 0.7880666613578796,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8409207463264465,
+      "learning_rate": 0.00013973696111089492,
+      "loss": 0.7052523136138916,
+      "mean_token_accuracy": 0.8000123649835587,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7787443134891853,
+      "eval_loss": 0.7137772440910339,
+      "eval_mean_token_accuracy": 0.8014209287220173,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 90.2562,
+      "eval_samples_per_second": 15.733,
+      "eval_steps_per_second": 1.972,
+      "step": 160
+    },
+    {
+      "entropy": 0.7696007996797561,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.0182477235794067,
+      "learning_rate": 0.00015731393735125907,
+      "loss": 0.7008797645568847,
+      "mean_token_accuracy": 0.8007259473204613,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.752259682068664,
+      "eval_loss": 0.7171286344528198,
+      "eval_mean_token_accuracy": 0.80052122574174,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 90.0797,
+      "eval_samples_per_second": 15.764,
+      "eval_steps_per_second": 1.976,
+      "step": 180
+    },
+    {
+      "entropy": 0.7828126326203346,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9553632736206055,
+      "learning_rate": 0.0001748909135916232,
+      "loss": 0.7046853542327881,
+      "mean_token_accuracy": 0.7988562889397144,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7389638169427936,
+      "eval_loss": 0.7187935709953308,
+      "eval_mean_token_accuracy": 0.7995899999409579,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 89.9223,
+      "eval_samples_per_second": 15.791,
+      "eval_steps_per_second": 1.979,
+      "step": 200
+    },
+    {
+      "entropy": 0.7909286297857762,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 1.134459376335144,
+      "learning_rate": 0.00019246788983198735,
+      "loss": 0.6989312171936035,
+      "mean_token_accuracy": 0.7991565138101577,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7496965199374082,
+      "eval_loss": 0.7106321454048157,
+      "eval_mean_token_accuracy": 0.8046183442131857,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 89.6899,
+      "eval_samples_per_second": 15.832,
+      "eval_steps_per_second": 1.985,
+      "step": 220
+    },
+    {
+      "entropy": 0.7651963323354721,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.9557585716247559,
+      "learning_rate": 0.0002100448660723515,
+      "loss": 0.6984559059143066,
+      "mean_token_accuracy": 0.8019823372364044,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7615563079212488,
+      "eval_loss": 0.7098539471626282,
+      "eval_mean_token_accuracy": 0.8047233508544022,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 90.3752,
+      "eval_samples_per_second": 15.712,
+      "eval_steps_per_second": 1.97,
+      "step": 240
+    },
+    {
+      "entropy": 0.787507726252079,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.9279311299324036,
+      "learning_rate": 0.00022762184231271568,
+      "loss": 0.708138370513916,
+      "mean_token_accuracy": 0.8013308703899383,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7856640896100676,
+      "eval_loss": 0.7191570401191711,
+      "eval_mean_token_accuracy": 0.8024677249153008,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 93.7336,
+      "eval_samples_per_second": 15.149,
+      "eval_steps_per_second": 1.899,
+      "step": 260
+    },
+    {
+      "entropy": 0.768489234894514,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.7992776036262512,
+      "learning_rate": 0.0002451988185530798,
+      "loss": 0.7035849571228028,
+      "mean_token_accuracy": 0.8031105332076549,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7682498740346244,
+      "eval_loss": 0.7149022221565247,
+      "eval_mean_token_accuracy": 0.7973497995499814,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 89.7309,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 280
+    },
+    {
+      "entropy": 0.7720955617725849,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.1931514739990234,
+      "learning_rate": 0.0002627757947934439,
+      "loss": 0.6965402603149414,
+      "mean_token_accuracy": 0.8040348663926125,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7214708304807042,
+      "eval_loss": 0.7350823879241943,
+      "eval_mean_token_accuracy": 0.7942911925610532,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 89.6451,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 300
+    },
+    {
+      "entropy": 0.7930095426738262,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8994467258453369,
+      "learning_rate": 0.0002803527710338081,
+      "loss": 0.7331492900848389,
+      "mean_token_accuracy": 0.796185302734375,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.8142535488926963,
+      "eval_loss": 0.7237834334373474,
+      "eval_mean_token_accuracy": 0.799495273761535,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 90.6759,
+      "eval_samples_per_second": 15.66,
+      "eval_steps_per_second": 1.963,
+      "step": 320
+    },
+    {
+      "entropy": 0.7825102441012859,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 1.8224239349365234,
+      "learning_rate": 0.00029792974727417223,
+      "loss": 0.7331175804138184,
+      "mean_token_accuracy": 0.7977667152881622,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7327135443017724,
+      "eval_loss": 0.7388784289360046,
+      "eval_mean_token_accuracy": 0.8007041252730938,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 89.7692,
+      "eval_samples_per_second": 15.818,
+      "eval_steps_per_second": 1.983,
+      "step": 340
+    },
+    {
+      "entropy": 0.819331557303667,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 1.0407779216766357,
+      "learning_rate": 0.0003155067235145364,
+      "loss": 0.7456695556640625,
+      "mean_token_accuracy": 0.7936980701982975,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.8065585420372781,
+      "eval_loss": 0.735519528388977,
+      "eval_mean_token_accuracy": 0.7974206296245704,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 89.4336,
+      "eval_samples_per_second": 15.878,
+      "eval_steps_per_second": 1.99,
+      "step": 360
+    },
+    {
+      "entropy": 0.8114653021097183,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.1209490299224854,
+      "learning_rate": 0.0003330836997549005,
+      "loss": 0.7340899467468261,
+      "mean_token_accuracy": 0.7914272703230381,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.8025527285056168,
+      "eval_loss": 0.7295576930046082,
+      "eval_mean_token_accuracy": 0.7990576628218876,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 89.9579,
+      "eval_samples_per_second": 15.785,
+      "eval_steps_per_second": 1.979,
+      "step": 380
+    },
+    {
+      "entropy": 0.8042497783899307,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 1.2437798976898193,
+      "learning_rate": 0.0003506606759952647,
+      "loss": 0.7381744384765625,
+      "mean_token_accuracy": 0.7967362694442273,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7717011336530193,
+      "eval_loss": 0.7436173558235168,
+      "eval_mean_token_accuracy": 0.7975849482450592,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 89.8411,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 400
+    },
+    {
+      "entropy": 0.7987750010056929,
+      "epoch": 1.012070006035003,
+      "grad_norm": 1.2390918731689453,
+      "learning_rate": 0.000364721224843344,
+      "loss": 0.7317790031433106,
+      "mean_token_accuracy": 0.7986087551364651,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7333170604170038,
+      "eval_loss": 0.7456948757171631,
+      "eval_mean_token_accuracy": 0.7964897577682238,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 89.7288,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 420
+    },
+    {
+      "entropy": 0.7372976846992969,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.1552740335464478,
+      "learning_rate": 0.00036468510102269575,
+      "loss": 0.6734249114990234,
+      "mean_token_accuracy": 0.805834824591875,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.7130355526891987,
+      "eval_loss": 0.7555333375930786,
+      "eval_mean_token_accuracy": 0.7961652781186479,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 89.8713,
+      "eval_samples_per_second": 15.8,
+      "eval_steps_per_second": 1.981,
+      "step": 440
+    },
+    {
+      "entropy": 0.7465668715536594,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 1.1991469860076904,
+      "learning_rate": 0.0003645973816745066,
+      "loss": 0.6871879577636719,
+      "mean_token_accuracy": 0.8042450234293937,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7772979542110743,
+      "eval_loss": 0.7561826705932617,
+      "eval_mean_token_accuracy": 0.7953876427720102,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 89.8395,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 460
+    },
+    {
+      "entropy": 0.7477855801582336,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 1.325486660003662,
+      "learning_rate": 0.00036445809162231435,
+      "loss": 0.6836830139160156,
+      "mean_token_accuracy": 0.8021619468927383,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.7098110676481483,
+      "eval_loss": 0.758007287979126,
+      "eval_mean_token_accuracy": 0.7909953189030122,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 89.6101,
+      "eval_samples_per_second": 15.846,
+      "eval_steps_per_second": 1.986,
+      "step": 480
+    },
+    {
+      "entropy": 0.7632691070437432,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 1.0122262239456177,
+      "learning_rate": 0.0003642672702835562,
+      "loss": 0.7071213245391845,
+      "mean_token_accuracy": 0.8000680930912495,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7645382720432924,
+      "eval_loss": 0.7492606043815613,
+      "eval_mean_token_accuracy": 0.7976858676149604,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 89.6092,
+      "eval_samples_per_second": 15.847,
+      "eval_steps_per_second": 1.986,
+      "step": 500
+    },
+    {
+      "entropy": 0.7623726457357407,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.367774486541748,
+      "learning_rate": 0.00036402497165841384,
+      "loss": 0.6965175151824952,
+      "mean_token_accuracy": 0.8016501650214195,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7868083715438843,
+      "eval_loss": 0.7481449842453003,
+      "eval_mean_token_accuracy": 0.799330582779445,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 89.8241,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 520
+    },
+    {
+      "entropy": 0.7535207405686378,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 1.213614821434021,
+      "learning_rate": 0.00036373126431453207,
+      "loss": 0.7078310489654541,
+      "mean_token_accuracy": 0.7998907208442688,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7829091950748743,
+      "eval_loss": 0.7416072487831116,
+      "eval_mean_token_accuracy": 0.7969146250339036,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 89.4607,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 540
+    },
+    {
+      "entropy": 0.7527363449335098,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 1.0929982662200928,
+      "learning_rate": 0.00036338623136761495,
+      "loss": 0.7058369636535644,
+      "mean_token_accuracy": 0.7997087433934211,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.728419938114252,
+      "eval_loss": 0.744873046875,
+      "eval_mean_token_accuracy": 0.7966634296299366,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 89.6597,
+      "eval_samples_per_second": 15.838,
+      "eval_steps_per_second": 1.985,
+      "step": 560
+    },
+    {
+      "entropy": 0.7406099259853363,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.409364104270935,
+      "learning_rate": 0.00036298997045790515,
+      "loss": 0.6902852058410645,
+      "mean_token_accuracy": 0.805386807769537,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.6991632758231645,
+      "eval_loss": 0.7480319738388062,
+      "eval_mean_token_accuracy": 0.7970068006033308,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 89.8329,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 580
+    },
+    {
+      "entropy": 0.7459449704736472,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.7053111791610718,
+      "learning_rate": 0.00036254259372255277,
+      "loss": 0.7044640064239502,
+      "mean_token_accuracy": 0.8005018755793571,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7454566660891758,
+      "eval_loss": 0.7551484704017639,
+      "eval_mean_token_accuracy": 0.7897093731365846,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 89.8313,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 600
+    },
+    {
+      "entropy": 0.7621391348540782,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 1.174443244934082,
+      "learning_rate": 0.000362044227763882,
+      "loss": 0.7111573696136475,
+      "mean_token_accuracy": 0.798752411454916,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7664387949397055,
+      "eval_loss": 0.7365804314613342,
+      "eval_mean_token_accuracy": 0.7963101200843126,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 89.9623,
+      "eval_samples_per_second": 15.784,
+      "eval_steps_per_second": 1.979,
+      "step": 620
+    },
+    {
+      "entropy": 0.7424944408237935,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.2898184061050415,
+      "learning_rate": 0.000361495013613564,
+      "loss": 0.6913110733032226,
+      "mean_token_accuracy": 0.8009015507996082,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7255479529332579,
+      "eval_loss": 0.7345991730690002,
+      "eval_mean_token_accuracy": 0.8013592945056015,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 89.817,
+      "eval_samples_per_second": 15.81,
+      "eval_steps_per_second": 1.982,
+      "step": 640
+    },
+    {
+      "entropy": 0.7303367637097835,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.1310476064682007,
+      "learning_rate": 0.00036089510669270666,
+      "loss": 0.6875617027282714,
+      "mean_token_accuracy": 0.8035397171974182,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7101308248016271,
+      "eval_loss": 0.7394917011260986,
+      "eval_mean_token_accuracy": 0.7990792942850777,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 89.7431,
+      "eval_samples_per_second": 15.823,
+      "eval_steps_per_second": 1.983,
+      "step": 660
+    },
+    {
+      "entropy": 0.7387678384780884,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.9645389914512634,
+      "learning_rate": 0.0003602446767678725,
+      "loss": 0.6911204338073731,
+      "mean_token_accuracy": 0.8056333385407924,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7281073737010527,
+      "eval_loss": 0.7376708984375,
+      "eval_mean_token_accuracy": 0.7997356415464637,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 90.0338,
+      "eval_samples_per_second": 15.772,
+      "eval_steps_per_second": 1.977,
+      "step": 680
+    },
+    {
+      "entropy": 0.7449554048478604,
+      "epoch": 1.687990343995172,
+      "grad_norm": 1.9069629907608032,
+      "learning_rate": 0.0003595439079030364,
+      "loss": 0.7042286396026611,
+      "mean_token_accuracy": 0.8018639378249646,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7022633817088738,
+      "eval_loss": 0.7320939898490906,
+      "eval_mean_token_accuracy": 0.8006051541044471,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 89.8948,
+      "eval_samples_per_second": 15.796,
+      "eval_steps_per_second": 1.98,
+      "step": 700
+    },
+    {
+      "entropy": 0.723249051719904,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.8333877325057983,
+      "learning_rate": 0.00035879299840749777,
+      "loss": 0.7043714046478271,
+      "mean_token_accuracy": 0.8028345607221127,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.672436946898364,
+      "eval_loss": 0.713233470916748,
+      "eval_mean_token_accuracy": 0.8047760577684038,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 89.6394,
+      "eval_samples_per_second": 15.841,
+      "eval_steps_per_second": 1.986,
+      "step": 720
+    },
+    {
+      "entropy": 0.730014206469059,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.4580037593841553,
+      "learning_rate": 0.00035799216077976137,
+      "loss": 0.6985176563262939,
+      "mean_token_accuracy": 0.8015547141432762,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.6908873795123582,
+      "eval_loss": 0.7198202013969421,
+      "eval_mean_token_accuracy": 0.8024131682481659,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 89.3771,
+      "eval_samples_per_second": 15.888,
+      "eval_steps_per_second": 1.992,
+      "step": 740
+    },
+    {
+      "entropy": 0.7219605796039105,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.51682710647583,
+      "learning_rate": 0.000357141621647403,
+      "loss": 0.6844424724578857,
+      "mean_token_accuracy": 0.8032813094556331,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7113401109582922,
+      "eval_loss": 0.7195309400558472,
+      "eval_mean_token_accuracy": 0.8019187035185568,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 89.4658,
+      "eval_samples_per_second": 15.872,
+      "eval_steps_per_second": 1.99,
+      "step": 760
+    },
+    {
+      "entropy": 0.729826743900776,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 1.0838000774383545,
+      "learning_rate": 0.0003562416217029361,
+      "loss": 0.685858678817749,
+      "mean_token_accuracy": 0.8050567395985126,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.649370267484965,
+      "eval_loss": 0.7177029252052307,
+      "eval_mean_token_accuracy": 0.8050174043419656,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 89.2287,
+      "eval_samples_per_second": 15.914,
+      "eval_steps_per_second": 1.995,
+      "step": 780
+    },
+    {
+      "entropy": 0.6799364153295755,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 1.1364924907684326,
+      "learning_rate": 0.00035529241563569903,
+      "loss": 0.6714428901672364,
+      "mean_token_accuracy": 0.8087341658771038,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6603799047095052,
+      "eval_loss": 0.7201809287071228,
+      "eval_mean_token_accuracy": 0.802863868769635,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 89.2718,
+      "eval_samples_per_second": 15.906,
+      "eval_steps_per_second": 1.994,
+      "step": 800
+    },
+    {
+      "entropy": 0.7078610375523567,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 1.2191942930221558,
+      "learning_rate": 0.0003542942720597808,
+      "loss": 0.6996590614318847,
+      "mean_token_accuracy": 0.8049512051045895,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.7240545264120852,
+      "eval_loss": 0.7301309108734131,
+      "eval_mean_token_accuracy": 0.8015973166133581,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 89.3438,
+      "eval_samples_per_second": 15.894,
+      "eval_steps_per_second": 1.992,
+      "step": 820
+    },
+    {
+      "entropy": 0.6557391556826505,
+      "epoch": 2.024140012070006,
+      "grad_norm": 1.36203134059906,
+      "learning_rate": 0.0003532474734380064,
+      "loss": 0.6327113628387451,
+      "mean_token_accuracy": 0.8142829847026181,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6573808648613062,
+      "eval_loss": 0.7334365844726562,
+      "eval_mean_token_accuracy": 0.8043633893634496,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 89.3359,
+      "eval_samples_per_second": 15.895,
+      "eval_steps_per_second": 1.992,
+      "step": 840
+    },
+    {
+      "entropy": 0.6409128502011299,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 1.5804955959320068,
+      "learning_rate": 0.0003521523160020035,
+      "loss": 0.6023559093475341,
+      "mean_token_accuracy": 0.8222251623868942,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6983234788594621,
+      "eval_loss": 0.7232135534286499,
+      "eval_mean_token_accuracy": 0.8010654268639811,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 89.2615,
+      "eval_samples_per_second": 15.908,
+      "eval_steps_per_second": 1.994,
+      "step": 860
+    },
+    {
+      "entropy": 0.6151149723678827,
+      "epoch": 2.12070006035003,
+      "grad_norm": 1.1975436210632324,
+      "learning_rate": 0.00035100910966837193,
+      "loss": 0.5803564548492431,
+      "mean_token_accuracy": 0.8256325736641884,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6636555573243773,
+      "eval_loss": 0.7205662727355957,
+      "eval_mean_token_accuracy": 0.8036768369460374,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 89.2957,
+      "eval_samples_per_second": 15.902,
+      "eval_steps_per_second": 1.993,
+      "step": 880
+    },
+    {
+      "entropy": 0.6201727617532015,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9907131195068359,
+      "learning_rate": 0.0003498181779509813,
+      "loss": 0.6101872444152832,
+      "mean_token_accuracy": 0.8196512959897518,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.6473788086617931,
+      "eval_loss": 0.7062397003173828,
+      "eval_mean_token_accuracy": 0.8078362184963869,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 88.6198,
+      "eval_samples_per_second": 16.024,
+      "eval_steps_per_second": 2.009,
+      "step": 900
+    },
+    {
+      "entropy": 0.6097332935780286,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.9055633544921875,
+      "learning_rate": 0.00034857985786942026,
+      "loss": 0.5956094264984131,
+      "mean_token_accuracy": 0.8240575887262821,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.612629868341296,
+      "eval_loss": 0.7106770277023315,
+      "eval_mean_token_accuracy": 0.805324277181304,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 89.4383,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 920
+    },
+    {
+      "entropy": 0.6182701248675585,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.9513736367225647,
+      "learning_rate": 0.00034729449985362404,
+      "loss": 0.6019269466400147,
+      "mean_token_accuracy": 0.8187106013298034,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6437647456533453,
+      "eval_loss": 0.7173502445220947,
+      "eval_mean_token_accuracy": 0.8056439573175451,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.9568,
+      "eval_samples_per_second": 15.963,
+      "eval_steps_per_second": 2.001,
+      "step": 940
+    },
+    {
+      "entropy": 0.6508570514619351,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.9606207013130188,
+      "learning_rate": 0.0003459624676447067,
+      "loss": 0.6296409606933594,
+      "mean_token_accuracy": 0.8167732007801533,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6288247135248077,
+      "eval_loss": 0.7160675525665283,
+      "eval_mean_token_accuracy": 0.8066010689467527,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 89.0044,
+      "eval_samples_per_second": 15.954,
+      "eval_steps_per_second": 2.0,
+      "step": 960
+    },
+    {
+      "entropy": 0.6380081418901682,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.9677181839942932,
+      "learning_rate": 0.000344584138192027,
+      "loss": 0.6210546493530273,
+      "mean_token_accuracy": 0.8186135642230511,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6407805031604981,
+      "eval_loss": 0.7055043578147888,
+      "eval_mean_token_accuracy": 0.8070103157772107,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 89.5606,
+      "eval_samples_per_second": 15.855,
+      "eval_steps_per_second": 1.987,
+      "step": 980
+    },
+    {
+      "entropy": 0.6310251638293266,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 2.1046407222747803,
+      "learning_rate": 0.000343159901546516,
+      "loss": 0.6165830135345459,
+      "mean_token_accuracy": 0.8207855716347694,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6461507470420237,
+      "eval_loss": 0.7292136549949646,
+      "eval_mean_token_accuracy": 0.8030418261383356,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 89.315,
+      "eval_samples_per_second": 15.899,
+      "eval_steps_per_second": 1.993,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6256617344915867,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 1.1768349409103394,
+      "learning_rate": 0.0003416901607502972,
+      "loss": 0.6142709255218506,
+      "mean_token_accuracy": 0.8220669947564602,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6150523993406403,
+      "eval_loss": 0.7019563317298889,
+      "eval_mean_token_accuracy": 0.8094323099998946,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 89.419,
+      "eval_samples_per_second": 15.88,
+      "eval_steps_per_second": 1.991,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6126735735684633,
+      "epoch": 2.506940253470127,
+      "grad_norm": 1.8065496683120728,
+      "learning_rate": 0.00034017533172263055,
+      "loss": 0.6056458473205566,
+      "mean_token_accuracy": 0.8203762136399746,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6544345456562685,
+      "eval_loss": 0.7086517214775085,
+      "eval_mean_token_accuracy": 0.806709805901131,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 88.981,
+      "eval_samples_per_second": 15.958,
+      "eval_steps_per_second": 2.0,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6311754353344441,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 1.0576369762420654,
+      "learning_rate": 0.00033861584314221236,
+      "loss": 0.6176392555236816,
+      "mean_token_accuracy": 0.8177410490810871,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6273799200406235,
+      "eval_loss": 0.708163321018219,
+      "eval_mean_token_accuracy": 0.8032017099053672,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 89.7668,
+      "eval_samples_per_second": 15.819,
+      "eval_steps_per_second": 1.983,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6200249589979648,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.997065007686615,
+      "learning_rate": 0.0003370121363258637,
+      "loss": 0.6203888893127442,
+      "mean_token_accuracy": 0.8196275025606156,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6457576811983344,
+      "eval_loss": 0.6972290277481079,
+      "eval_mean_token_accuracy": 0.8095980849158898,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 89.2027,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 1080
+    },
+    {
+      "entropy": 0.626810473203659,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.865475058555603,
+      "learning_rate": 0.0003353646651036438,
+      "loss": 0.6102585792541504,
+      "mean_token_accuracy": 0.819708751142025,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.6412022431914726,
+      "eval_loss": 0.7036139965057373,
+      "eval_mean_token_accuracy": 0.8070317135098275,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 89.4733,
+      "eval_samples_per_second": 15.871,
+      "eval_steps_per_second": 1.989,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6581619590520859,
+      "epoch": 2.700060350030175,
+      "grad_norm": 1.00307035446167,
+      "learning_rate": 0.00033367389569042064,
+      "loss": 0.6397720336914062,
+      "mean_token_accuracy": 0.81413309648633,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6288736402318719,
+      "eval_loss": 0.6966825723648071,
+      "eval_mean_token_accuracy": 0.8094039669867312,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 88.7341,
+      "eval_samples_per_second": 16.003,
+      "eval_steps_per_second": 2.006,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6399412982165813,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.8589329719543457,
+      "learning_rate": 0.0003319403065539384,
+      "loss": 0.6347338199615479,
+      "mean_token_accuracy": 0.8139534957706929,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.683491913455256,
+      "eval_loss": 0.6984149217605591,
+      "eval_mean_token_accuracy": 0.8083989469522841,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 89.3929,
+      "eval_samples_per_second": 15.885,
+      "eval_steps_per_second": 1.991,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6264835461974144,
+      "epoch": 2.796620398310199,
+      "grad_norm": 0.7807853817939758,
+      "learning_rate": 0.0003301643882794163,
+      "loss": 0.6162629127502441,
+      "mean_token_accuracy": 0.8216292977333068,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6578792159476977,
+      "eval_loss": 0.6921441555023193,
+      "eval_mean_token_accuracy": 0.8094299177775223,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 89.2356,
+      "eval_samples_per_second": 15.913,
+      "eval_steps_per_second": 1.995,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6269129011780024,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 1.1675593852996826,
+      "learning_rate": 0.0003283466434307189,
+      "loss": 0.6235532760620117,
+      "mean_token_accuracy": 0.8188897788524627,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6533248447970058,
+      "eval_loss": 0.696849524974823,
+      "eval_mean_token_accuracy": 0.8102703713968898,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2602,
+      "eval_samples_per_second": 16.089,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.6259873781353236,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.9596486687660217,
+      "learning_rate": 0.00032648758640813655,
+      "loss": 0.6118728160858155,
+      "mean_token_accuracy": 0.8171859130263328,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6212078754821521,
+      "eval_loss": 0.6942870020866394,
+      "eval_mean_token_accuracy": 0.810484343365337,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 89.4375,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 1200
+    },
+    {
+      "entropy": 0.6142537571489811,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 0.9625058770179749,
+      "learning_rate": 0.00032458774330281615,
+      "loss": 0.6060261249542236,
+      "mean_token_accuracy": 0.8223113484680653,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6553665667437436,
+      "eval_loss": 0.6835269331932068,
+      "eval_mean_token_accuracy": 0.8134723968720168,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 89.5092,
+      "eval_samples_per_second": 15.864,
+      "eval_steps_per_second": 1.989,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6301972426474094,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.9657333493232727,
+      "learning_rate": 0.0003226476517478835,
+      "loss": 0.6277759075164795,
+      "mean_token_accuracy": 0.8164266042411328,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6594253118788258,
+      "eval_loss": 0.6820935606956482,
+      "eval_mean_token_accuracy": 0.8132368198941263,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 89.6211,
+      "eval_samples_per_second": 15.844,
+      "eval_steps_per_second": 1.986,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5346100655469027,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8541185259819031,
+      "learning_rate": 0.0003206678607662996,
+      "loss": 0.5261692047119141,
+      "mean_token_accuracy": 0.8382289425119177,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5626645662476507,
+      "eval_loss": 0.7028542160987854,
+      "eval_mean_token_accuracy": 0.8133395572056931,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 89.1267,
+      "eval_samples_per_second": 15.932,
+      "eval_steps_per_second": 1.997,
+      "step": 1260
+    },
+    {
+      "entropy": 0.5001101832836866,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.8208815455436707,
+      "learning_rate": 0.0003186489306154935,
+      "loss": 0.4850132942199707,
+      "mean_token_accuracy": 0.8468983843922615,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5889462046743779,
+      "eval_loss": 0.691015899181366,
+      "eval_mean_token_accuracy": 0.8135226466012805,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.9677,
+      "eval_samples_per_second": 15.961,
+      "eval_steps_per_second": 2.001,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5242766339331866,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 0.7893072962760925,
+      "learning_rate": 0.0003165914326288163,
+      "loss": 0.5061577320098877,
+      "mean_token_accuracy": 0.8425608821213245,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5543691962957382,
+      "eval_loss": 0.6913734078407288,
+      "eval_mean_token_accuracy": 0.8154889680026622,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 89.6963,
+      "eval_samples_per_second": 15.831,
+      "eval_steps_per_second": 1.984,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5327721010893584,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.8992080092430115,
+      "learning_rate": 0.0003144959490538604,
+      "loss": 0.5068036556243897,
+      "mean_token_accuracy": 0.8420542575418949,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5734540803378887,
+      "eval_loss": 0.6948716044425964,
+      "eval_mean_token_accuracy": 0.8137217944257715,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 89.4597,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 1320
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.326625819426898e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1340/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1340/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1340/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1340/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1340/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..7ce4b072fcbe5b636867e13937c56fedee3b68b4
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1340/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.02728394274790723,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1340/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1340/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1340/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1340/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1340/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1340/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1340/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1340/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..f4acabad58753fbfbc178d0b9274e94dbc689bdd
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1340/trainer_state.json
@@ -0,0 +1,1441 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 3.2293301146650575,
+  "eval_steps": 20,
+  "global_step": 1340,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.024795526266098,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.945375919342041,
+      "learning_rate": 1.6698127428345936e-05,
+      "loss": 1.7825420379638672,
+      "mean_token_accuracy": 0.6315708436071873,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.29921497588747,
+      "eval_loss": 1.169131875038147,
+      "eval_mean_token_accuracy": 0.7185706342204233,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 91.7898,
+      "eval_samples_per_second": 15.47,
+      "eval_steps_per_second": 1.939,
+      "step": 20
+    },
+    {
+      "entropy": 1.0346344470977784,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.7916979789733887,
+      "learning_rate": 3.427510366871008e-05,
+      "loss": 0.9480395317077637,
+      "mean_token_accuracy": 0.7543429024517536,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 0.9673695430326997,
+      "eval_loss": 0.8588430285453796,
+      "eval_mean_token_accuracy": 0.7737233002534073,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 89.8219,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 40
+    },
+    {
+      "entropy": 0.8977566614747048,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7972187995910645,
+      "learning_rate": 5.185207990907422e-05,
+      "loss": 0.8029604911804199,
+      "mean_token_accuracy": 0.7818016350269318,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8436836982041263,
+      "eval_loss": 0.7893033623695374,
+      "eval_mean_token_accuracy": 0.7858914984076211,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 89.9349,
+      "eval_samples_per_second": 15.789,
+      "eval_steps_per_second": 1.979,
+      "step": 60
+    },
+    {
+      "entropy": 0.8650321319699288,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.2128729820251465,
+      "learning_rate": 6.942905614943836e-05,
+      "loss": 0.77417893409729,
+      "mean_token_accuracy": 0.7875479467213153,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8209857180547179,
+      "eval_loss": 0.7572280168533325,
+      "eval_mean_token_accuracy": 0.7925970383574453,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 89.9459,
+      "eval_samples_per_second": 15.787,
+      "eval_steps_per_second": 1.979,
+      "step": 80
+    },
+    {
+      "entropy": 0.8096660938113928,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.1254287958145142,
+      "learning_rate": 8.700603238980251e-05,
+      "loss": 0.7170331001281738,
+      "mean_token_accuracy": 0.7994555257260799,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.792780208788561,
+      "eval_loss": 0.7442984580993652,
+      "eval_mean_token_accuracy": 0.7954704573984896,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 90.294,
+      "eval_samples_per_second": 15.726,
+      "eval_steps_per_second": 1.971,
+      "step": 100
+    },
+    {
+      "entropy": 0.7979829199612141,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2140507698059082,
+      "learning_rate": 0.00010458300863016666,
+      "loss": 0.7141861915588379,
+      "mean_token_accuracy": 0.8000014744699001,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.7938692486018277,
+      "eval_loss": 0.7279797196388245,
+      "eval_mean_token_accuracy": 0.7989798734027348,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 89.7189,
+      "eval_samples_per_second": 15.827,
+      "eval_steps_per_second": 1.984,
+      "step": 120
+    },
+    {
+      "entropy": 0.7916347607970238,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.0506458282470703,
+      "learning_rate": 0.0001221599848705308,
+      "loss": 0.7117055416107178,
+      "mean_token_accuracy": 0.7970046654343605,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.8052781063519167,
+      "eval_loss": 0.722576916217804,
+      "eval_mean_token_accuracy": 0.7982530523551984,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 90.0376,
+      "eval_samples_per_second": 15.771,
+      "eval_steps_per_second": 1.977,
+      "step": 140
+    },
+    {
+      "entropy": 0.7880666613578796,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8409207463264465,
+      "learning_rate": 0.00013973696111089492,
+      "loss": 0.7052523136138916,
+      "mean_token_accuracy": 0.8000123649835587,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7787443134891853,
+      "eval_loss": 0.7137772440910339,
+      "eval_mean_token_accuracy": 0.8014209287220173,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 90.2562,
+      "eval_samples_per_second": 15.733,
+      "eval_steps_per_second": 1.972,
+      "step": 160
+    },
+    {
+      "entropy": 0.7696007996797561,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.0182477235794067,
+      "learning_rate": 0.00015731393735125907,
+      "loss": 0.7008797645568847,
+      "mean_token_accuracy": 0.8007259473204613,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.752259682068664,
+      "eval_loss": 0.7171286344528198,
+      "eval_mean_token_accuracy": 0.80052122574174,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 90.0797,
+      "eval_samples_per_second": 15.764,
+      "eval_steps_per_second": 1.976,
+      "step": 180
+    },
+    {
+      "entropy": 0.7828126326203346,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9553632736206055,
+      "learning_rate": 0.0001748909135916232,
+      "loss": 0.7046853542327881,
+      "mean_token_accuracy": 0.7988562889397144,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7389638169427936,
+      "eval_loss": 0.7187935709953308,
+      "eval_mean_token_accuracy": 0.7995899999409579,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 89.9223,
+      "eval_samples_per_second": 15.791,
+      "eval_steps_per_second": 1.979,
+      "step": 200
+    },
+    {
+      "entropy": 0.7909286297857762,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 1.134459376335144,
+      "learning_rate": 0.00019246788983198735,
+      "loss": 0.6989312171936035,
+      "mean_token_accuracy": 0.7991565138101577,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7496965199374082,
+      "eval_loss": 0.7106321454048157,
+      "eval_mean_token_accuracy": 0.8046183442131857,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 89.6899,
+      "eval_samples_per_second": 15.832,
+      "eval_steps_per_second": 1.985,
+      "step": 220
+    },
+    {
+      "entropy": 0.7651963323354721,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.9557585716247559,
+      "learning_rate": 0.0002100448660723515,
+      "loss": 0.6984559059143066,
+      "mean_token_accuracy": 0.8019823372364044,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7615563079212488,
+      "eval_loss": 0.7098539471626282,
+      "eval_mean_token_accuracy": 0.8047233508544022,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 90.3752,
+      "eval_samples_per_second": 15.712,
+      "eval_steps_per_second": 1.97,
+      "step": 240
+    },
+    {
+      "entropy": 0.787507726252079,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.9279311299324036,
+      "learning_rate": 0.00022762184231271568,
+      "loss": 0.708138370513916,
+      "mean_token_accuracy": 0.8013308703899383,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7856640896100676,
+      "eval_loss": 0.7191570401191711,
+      "eval_mean_token_accuracy": 0.8024677249153008,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 93.7336,
+      "eval_samples_per_second": 15.149,
+      "eval_steps_per_second": 1.899,
+      "step": 260
+    },
+    {
+      "entropy": 0.768489234894514,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.7992776036262512,
+      "learning_rate": 0.0002451988185530798,
+      "loss": 0.7035849571228028,
+      "mean_token_accuracy": 0.8031105332076549,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7682498740346244,
+      "eval_loss": 0.7149022221565247,
+      "eval_mean_token_accuracy": 0.7973497995499814,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 89.7309,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 280
+    },
+    {
+      "entropy": 0.7720955617725849,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.1931514739990234,
+      "learning_rate": 0.0002627757947934439,
+      "loss": 0.6965402603149414,
+      "mean_token_accuracy": 0.8040348663926125,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7214708304807042,
+      "eval_loss": 0.7350823879241943,
+      "eval_mean_token_accuracy": 0.7942911925610532,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 89.6451,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 300
+    },
+    {
+      "entropy": 0.7930095426738262,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8994467258453369,
+      "learning_rate": 0.0002803527710338081,
+      "loss": 0.7331492900848389,
+      "mean_token_accuracy": 0.796185302734375,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.8142535488926963,
+      "eval_loss": 0.7237834334373474,
+      "eval_mean_token_accuracy": 0.799495273761535,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 90.6759,
+      "eval_samples_per_second": 15.66,
+      "eval_steps_per_second": 1.963,
+      "step": 320
+    },
+    {
+      "entropy": 0.7825102441012859,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 1.8224239349365234,
+      "learning_rate": 0.00029792974727417223,
+      "loss": 0.7331175804138184,
+      "mean_token_accuracy": 0.7977667152881622,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7327135443017724,
+      "eval_loss": 0.7388784289360046,
+      "eval_mean_token_accuracy": 0.8007041252730938,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 89.7692,
+      "eval_samples_per_second": 15.818,
+      "eval_steps_per_second": 1.983,
+      "step": 340
+    },
+    {
+      "entropy": 0.819331557303667,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 1.0407779216766357,
+      "learning_rate": 0.0003155067235145364,
+      "loss": 0.7456695556640625,
+      "mean_token_accuracy": 0.7936980701982975,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.8065585420372781,
+      "eval_loss": 0.735519528388977,
+      "eval_mean_token_accuracy": 0.7974206296245704,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 89.4336,
+      "eval_samples_per_second": 15.878,
+      "eval_steps_per_second": 1.99,
+      "step": 360
+    },
+    {
+      "entropy": 0.8114653021097183,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.1209490299224854,
+      "learning_rate": 0.0003330836997549005,
+      "loss": 0.7340899467468261,
+      "mean_token_accuracy": 0.7914272703230381,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.8025527285056168,
+      "eval_loss": 0.7295576930046082,
+      "eval_mean_token_accuracy": 0.7990576628218876,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 89.9579,
+      "eval_samples_per_second": 15.785,
+      "eval_steps_per_second": 1.979,
+      "step": 380
+    },
+    {
+      "entropy": 0.8042497783899307,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 1.2437798976898193,
+      "learning_rate": 0.0003506606759952647,
+      "loss": 0.7381744384765625,
+      "mean_token_accuracy": 0.7967362694442273,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7717011336530193,
+      "eval_loss": 0.7436173558235168,
+      "eval_mean_token_accuracy": 0.7975849482450592,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 89.8411,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 400
+    },
+    {
+      "entropy": 0.7987750010056929,
+      "epoch": 1.012070006035003,
+      "grad_norm": 1.2390918731689453,
+      "learning_rate": 0.000364721224843344,
+      "loss": 0.7317790031433106,
+      "mean_token_accuracy": 0.7986087551364651,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7333170604170038,
+      "eval_loss": 0.7456948757171631,
+      "eval_mean_token_accuracy": 0.7964897577682238,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 89.7288,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 420
+    },
+    {
+      "entropy": 0.7372976846992969,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.1552740335464478,
+      "learning_rate": 0.00036468510102269575,
+      "loss": 0.6734249114990234,
+      "mean_token_accuracy": 0.805834824591875,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.7130355526891987,
+      "eval_loss": 0.7555333375930786,
+      "eval_mean_token_accuracy": 0.7961652781186479,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 89.8713,
+      "eval_samples_per_second": 15.8,
+      "eval_steps_per_second": 1.981,
+      "step": 440
+    },
+    {
+      "entropy": 0.7465668715536594,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 1.1991469860076904,
+      "learning_rate": 0.0003645973816745066,
+      "loss": 0.6871879577636719,
+      "mean_token_accuracy": 0.8042450234293937,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7772979542110743,
+      "eval_loss": 0.7561826705932617,
+      "eval_mean_token_accuracy": 0.7953876427720102,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 89.8395,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 460
+    },
+    {
+      "entropy": 0.7477855801582336,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 1.325486660003662,
+      "learning_rate": 0.00036445809162231435,
+      "loss": 0.6836830139160156,
+      "mean_token_accuracy": 0.8021619468927383,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.7098110676481483,
+      "eval_loss": 0.758007287979126,
+      "eval_mean_token_accuracy": 0.7909953189030122,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 89.6101,
+      "eval_samples_per_second": 15.846,
+      "eval_steps_per_second": 1.986,
+      "step": 480
+    },
+    {
+      "entropy": 0.7632691070437432,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 1.0122262239456177,
+      "learning_rate": 0.0003642672702835562,
+      "loss": 0.7071213245391845,
+      "mean_token_accuracy": 0.8000680930912495,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7645382720432924,
+      "eval_loss": 0.7492606043815613,
+      "eval_mean_token_accuracy": 0.7976858676149604,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 89.6092,
+      "eval_samples_per_second": 15.847,
+      "eval_steps_per_second": 1.986,
+      "step": 500
+    },
+    {
+      "entropy": 0.7623726457357407,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.367774486541748,
+      "learning_rate": 0.00036402497165841384,
+      "loss": 0.6965175151824952,
+      "mean_token_accuracy": 0.8016501650214195,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7868083715438843,
+      "eval_loss": 0.7481449842453003,
+      "eval_mean_token_accuracy": 0.799330582779445,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 89.8241,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 520
+    },
+    {
+      "entropy": 0.7535207405686378,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 1.213614821434021,
+      "learning_rate": 0.00036373126431453207,
+      "loss": 0.7078310489654541,
+      "mean_token_accuracy": 0.7998907208442688,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7829091950748743,
+      "eval_loss": 0.7416072487831116,
+      "eval_mean_token_accuracy": 0.7969146250339036,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 89.4607,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 540
+    },
+    {
+      "entropy": 0.7527363449335098,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 1.0929982662200928,
+      "learning_rate": 0.00036338623136761495,
+      "loss": 0.7058369636535644,
+      "mean_token_accuracy": 0.7997087433934211,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.728419938114252,
+      "eval_loss": 0.744873046875,
+      "eval_mean_token_accuracy": 0.7966634296299366,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 89.6597,
+      "eval_samples_per_second": 15.838,
+      "eval_steps_per_second": 1.985,
+      "step": 560
+    },
+    {
+      "entropy": 0.7406099259853363,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.409364104270935,
+      "learning_rate": 0.00036298997045790515,
+      "loss": 0.6902852058410645,
+      "mean_token_accuracy": 0.805386807769537,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.6991632758231645,
+      "eval_loss": 0.7480319738388062,
+      "eval_mean_token_accuracy": 0.7970068006033308,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 89.8329,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 580
+    },
+    {
+      "entropy": 0.7459449704736472,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.7053111791610718,
+      "learning_rate": 0.00036254259372255277,
+      "loss": 0.7044640064239502,
+      "mean_token_accuracy": 0.8005018755793571,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7454566660891758,
+      "eval_loss": 0.7551484704017639,
+      "eval_mean_token_accuracy": 0.7897093731365846,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 89.8313,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 600
+    },
+    {
+      "entropy": 0.7621391348540782,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 1.174443244934082,
+      "learning_rate": 0.000362044227763882,
+      "loss": 0.7111573696136475,
+      "mean_token_accuracy": 0.798752411454916,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7664387949397055,
+      "eval_loss": 0.7365804314613342,
+      "eval_mean_token_accuracy": 0.7963101200843126,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 89.9623,
+      "eval_samples_per_second": 15.784,
+      "eval_steps_per_second": 1.979,
+      "step": 620
+    },
+    {
+      "entropy": 0.7424944408237935,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.2898184061050415,
+      "learning_rate": 0.000361495013613564,
+      "loss": 0.6913110733032226,
+      "mean_token_accuracy": 0.8009015507996082,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7255479529332579,
+      "eval_loss": 0.7345991730690002,
+      "eval_mean_token_accuracy": 0.8013592945056015,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 89.817,
+      "eval_samples_per_second": 15.81,
+      "eval_steps_per_second": 1.982,
+      "step": 640
+    },
+    {
+      "entropy": 0.7303367637097835,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.1310476064682007,
+      "learning_rate": 0.00036089510669270666,
+      "loss": 0.6875617027282714,
+      "mean_token_accuracy": 0.8035397171974182,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7101308248016271,
+      "eval_loss": 0.7394917011260986,
+      "eval_mean_token_accuracy": 0.7990792942850777,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 89.7431,
+      "eval_samples_per_second": 15.823,
+      "eval_steps_per_second": 1.983,
+      "step": 660
+    },
+    {
+      "entropy": 0.7387678384780884,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.9645389914512634,
+      "learning_rate": 0.0003602446767678725,
+      "loss": 0.6911204338073731,
+      "mean_token_accuracy": 0.8056333385407924,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7281073737010527,
+      "eval_loss": 0.7376708984375,
+      "eval_mean_token_accuracy": 0.7997356415464637,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 90.0338,
+      "eval_samples_per_second": 15.772,
+      "eval_steps_per_second": 1.977,
+      "step": 680
+    },
+    {
+      "entropy": 0.7449554048478604,
+      "epoch": 1.687990343995172,
+      "grad_norm": 1.9069629907608032,
+      "learning_rate": 0.0003595439079030364,
+      "loss": 0.7042286396026611,
+      "mean_token_accuracy": 0.8018639378249646,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7022633817088738,
+      "eval_loss": 0.7320939898490906,
+      "eval_mean_token_accuracy": 0.8006051541044471,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 89.8948,
+      "eval_samples_per_second": 15.796,
+      "eval_steps_per_second": 1.98,
+      "step": 700
+    },
+    {
+      "entropy": 0.723249051719904,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.8333877325057983,
+      "learning_rate": 0.00035879299840749777,
+      "loss": 0.7043714046478271,
+      "mean_token_accuracy": 0.8028345607221127,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.672436946898364,
+      "eval_loss": 0.713233470916748,
+      "eval_mean_token_accuracy": 0.8047760577684038,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 89.6394,
+      "eval_samples_per_second": 15.841,
+      "eval_steps_per_second": 1.986,
+      "step": 720
+    },
+    {
+      "entropy": 0.730014206469059,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.4580037593841553,
+      "learning_rate": 0.00035799216077976137,
+      "loss": 0.6985176563262939,
+      "mean_token_accuracy": 0.8015547141432762,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.6908873795123582,
+      "eval_loss": 0.7198202013969421,
+      "eval_mean_token_accuracy": 0.8024131682481659,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 89.3771,
+      "eval_samples_per_second": 15.888,
+      "eval_steps_per_second": 1.992,
+      "step": 740
+    },
+    {
+      "entropy": 0.7219605796039105,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.51682710647583,
+      "learning_rate": 0.000357141621647403,
+      "loss": 0.6844424724578857,
+      "mean_token_accuracy": 0.8032813094556331,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7113401109582922,
+      "eval_loss": 0.7195309400558472,
+      "eval_mean_token_accuracy": 0.8019187035185568,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 89.4658,
+      "eval_samples_per_second": 15.872,
+      "eval_steps_per_second": 1.99,
+      "step": 760
+    },
+    {
+      "entropy": 0.729826743900776,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 1.0838000774383545,
+      "learning_rate": 0.0003562416217029361,
+      "loss": 0.685858678817749,
+      "mean_token_accuracy": 0.8050567395985126,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.649370267484965,
+      "eval_loss": 0.7177029252052307,
+      "eval_mean_token_accuracy": 0.8050174043419656,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 89.2287,
+      "eval_samples_per_second": 15.914,
+      "eval_steps_per_second": 1.995,
+      "step": 780
+    },
+    {
+      "entropy": 0.6799364153295755,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 1.1364924907684326,
+      "learning_rate": 0.00035529241563569903,
+      "loss": 0.6714428901672364,
+      "mean_token_accuracy": 0.8087341658771038,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6603799047095052,
+      "eval_loss": 0.7201809287071228,
+      "eval_mean_token_accuracy": 0.802863868769635,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 89.2718,
+      "eval_samples_per_second": 15.906,
+      "eval_steps_per_second": 1.994,
+      "step": 800
+    },
+    {
+      "entropy": 0.7078610375523567,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 1.2191942930221558,
+      "learning_rate": 0.0003542942720597808,
+      "loss": 0.6996590614318847,
+      "mean_token_accuracy": 0.8049512051045895,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.7240545264120852,
+      "eval_loss": 0.7301309108734131,
+      "eval_mean_token_accuracy": 0.8015973166133581,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 89.3438,
+      "eval_samples_per_second": 15.894,
+      "eval_steps_per_second": 1.992,
+      "step": 820
+    },
+    {
+      "entropy": 0.6557391556826505,
+      "epoch": 2.024140012070006,
+      "grad_norm": 1.36203134059906,
+      "learning_rate": 0.0003532474734380064,
+      "loss": 0.6327113628387451,
+      "mean_token_accuracy": 0.8142829847026181,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6573808648613062,
+      "eval_loss": 0.7334365844726562,
+      "eval_mean_token_accuracy": 0.8043633893634496,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 89.3359,
+      "eval_samples_per_second": 15.895,
+      "eval_steps_per_second": 1.992,
+      "step": 840
+    },
+    {
+      "entropy": 0.6409128502011299,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 1.5804955959320068,
+      "learning_rate": 0.0003521523160020035,
+      "loss": 0.6023559093475341,
+      "mean_token_accuracy": 0.8222251623868942,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6983234788594621,
+      "eval_loss": 0.7232135534286499,
+      "eval_mean_token_accuracy": 0.8010654268639811,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 89.2615,
+      "eval_samples_per_second": 15.908,
+      "eval_steps_per_second": 1.994,
+      "step": 860
+    },
+    {
+      "entropy": 0.6151149723678827,
+      "epoch": 2.12070006035003,
+      "grad_norm": 1.1975436210632324,
+      "learning_rate": 0.00035100910966837193,
+      "loss": 0.5803564548492431,
+      "mean_token_accuracy": 0.8256325736641884,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6636555573243773,
+      "eval_loss": 0.7205662727355957,
+      "eval_mean_token_accuracy": 0.8036768369460374,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 89.2957,
+      "eval_samples_per_second": 15.902,
+      "eval_steps_per_second": 1.993,
+      "step": 880
+    },
+    {
+      "entropy": 0.6201727617532015,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9907131195068359,
+      "learning_rate": 0.0003498181779509813,
+      "loss": 0.6101872444152832,
+      "mean_token_accuracy": 0.8196512959897518,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.6473788086617931,
+      "eval_loss": 0.7062397003173828,
+      "eval_mean_token_accuracy": 0.8078362184963869,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 88.6198,
+      "eval_samples_per_second": 16.024,
+      "eval_steps_per_second": 2.009,
+      "step": 900
+    },
+    {
+      "entropy": 0.6097332935780286,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.9055633544921875,
+      "learning_rate": 0.00034857985786942026,
+      "loss": 0.5956094264984131,
+      "mean_token_accuracy": 0.8240575887262821,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.612629868341296,
+      "eval_loss": 0.7106770277023315,
+      "eval_mean_token_accuracy": 0.805324277181304,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 89.4383,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 920
+    },
+    {
+      "entropy": 0.6182701248675585,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.9513736367225647,
+      "learning_rate": 0.00034729449985362404,
+      "loss": 0.6019269466400147,
+      "mean_token_accuracy": 0.8187106013298034,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6437647456533453,
+      "eval_loss": 0.7173502445220947,
+      "eval_mean_token_accuracy": 0.8056439573175451,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.9568,
+      "eval_samples_per_second": 15.963,
+      "eval_steps_per_second": 2.001,
+      "step": 940
+    },
+    {
+      "entropy": 0.6508570514619351,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.9606207013130188,
+      "learning_rate": 0.0003459624676447067,
+      "loss": 0.6296409606933594,
+      "mean_token_accuracy": 0.8167732007801533,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6288247135248077,
+      "eval_loss": 0.7160675525665283,
+      "eval_mean_token_accuracy": 0.8066010689467527,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 89.0044,
+      "eval_samples_per_second": 15.954,
+      "eval_steps_per_second": 2.0,
+      "step": 960
+    },
+    {
+      "entropy": 0.6380081418901682,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.9677181839942932,
+      "learning_rate": 0.000344584138192027,
+      "loss": 0.6210546493530273,
+      "mean_token_accuracy": 0.8186135642230511,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6407805031604981,
+      "eval_loss": 0.7055043578147888,
+      "eval_mean_token_accuracy": 0.8070103157772107,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 89.5606,
+      "eval_samples_per_second": 15.855,
+      "eval_steps_per_second": 1.987,
+      "step": 980
+    },
+    {
+      "entropy": 0.6310251638293266,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 2.1046407222747803,
+      "learning_rate": 0.000343159901546516,
+      "loss": 0.6165830135345459,
+      "mean_token_accuracy": 0.8207855716347694,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6461507470420237,
+      "eval_loss": 0.7292136549949646,
+      "eval_mean_token_accuracy": 0.8030418261383356,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 89.315,
+      "eval_samples_per_second": 15.899,
+      "eval_steps_per_second": 1.993,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6256617344915867,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 1.1768349409103394,
+      "learning_rate": 0.0003416901607502972,
+      "loss": 0.6142709255218506,
+      "mean_token_accuracy": 0.8220669947564602,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6150523993406403,
+      "eval_loss": 0.7019563317298889,
+      "eval_mean_token_accuracy": 0.8094323099998946,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 89.419,
+      "eval_samples_per_second": 15.88,
+      "eval_steps_per_second": 1.991,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6126735735684633,
+      "epoch": 2.506940253470127,
+      "grad_norm": 1.8065496683120728,
+      "learning_rate": 0.00034017533172263055,
+      "loss": 0.6056458473205566,
+      "mean_token_accuracy": 0.8203762136399746,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6544345456562685,
+      "eval_loss": 0.7086517214775085,
+      "eval_mean_token_accuracy": 0.806709805901131,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 88.981,
+      "eval_samples_per_second": 15.958,
+      "eval_steps_per_second": 2.0,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6311754353344441,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 1.0576369762420654,
+      "learning_rate": 0.00033861584314221236,
+      "loss": 0.6176392555236816,
+      "mean_token_accuracy": 0.8177410490810871,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6273799200406235,
+      "eval_loss": 0.708163321018219,
+      "eval_mean_token_accuracy": 0.8032017099053672,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 89.7668,
+      "eval_samples_per_second": 15.819,
+      "eval_steps_per_second": 1.983,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6200249589979648,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.997065007686615,
+      "learning_rate": 0.0003370121363258637,
+      "loss": 0.6203888893127442,
+      "mean_token_accuracy": 0.8196275025606156,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6457576811983344,
+      "eval_loss": 0.6972290277481079,
+      "eval_mean_token_accuracy": 0.8095980849158898,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 89.2027,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 1080
+    },
+    {
+      "entropy": 0.626810473203659,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.865475058555603,
+      "learning_rate": 0.0003353646651036438,
+      "loss": 0.6102585792541504,
+      "mean_token_accuracy": 0.819708751142025,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.6412022431914726,
+      "eval_loss": 0.7036139965057373,
+      "eval_mean_token_accuracy": 0.8070317135098275,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 89.4733,
+      "eval_samples_per_second": 15.871,
+      "eval_steps_per_second": 1.989,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6581619590520859,
+      "epoch": 2.700060350030175,
+      "grad_norm": 1.00307035446167,
+      "learning_rate": 0.00033367389569042064,
+      "loss": 0.6397720336914062,
+      "mean_token_accuracy": 0.81413309648633,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6288736402318719,
+      "eval_loss": 0.6966825723648071,
+      "eval_mean_token_accuracy": 0.8094039669867312,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 88.7341,
+      "eval_samples_per_second": 16.003,
+      "eval_steps_per_second": 2.006,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6399412982165813,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.8589329719543457,
+      "learning_rate": 0.0003319403065539384,
+      "loss": 0.6347338199615479,
+      "mean_token_accuracy": 0.8139534957706929,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.683491913455256,
+      "eval_loss": 0.6984149217605591,
+      "eval_mean_token_accuracy": 0.8083989469522841,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 89.3929,
+      "eval_samples_per_second": 15.885,
+      "eval_steps_per_second": 1.991,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6264835461974144,
+      "epoch": 2.796620398310199,
+      "grad_norm": 0.7807853817939758,
+      "learning_rate": 0.0003301643882794163,
+      "loss": 0.6162629127502441,
+      "mean_token_accuracy": 0.8216292977333068,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6578792159476977,
+      "eval_loss": 0.6921441555023193,
+      "eval_mean_token_accuracy": 0.8094299177775223,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 89.2356,
+      "eval_samples_per_second": 15.913,
+      "eval_steps_per_second": 1.995,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6269129011780024,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 1.1675593852996826,
+      "learning_rate": 0.0003283466434307189,
+      "loss": 0.6235532760620117,
+      "mean_token_accuracy": 0.8188897788524627,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6533248447970058,
+      "eval_loss": 0.696849524974823,
+      "eval_mean_token_accuracy": 0.8102703713968898,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2602,
+      "eval_samples_per_second": 16.089,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.6259873781353236,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.9596486687660217,
+      "learning_rate": 0.00032648758640813655,
+      "loss": 0.6118728160858155,
+      "mean_token_accuracy": 0.8171859130263328,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6212078754821521,
+      "eval_loss": 0.6942870020866394,
+      "eval_mean_token_accuracy": 0.810484343365337,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 89.4375,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 1200
+    },
+    {
+      "entropy": 0.6142537571489811,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 0.9625058770179749,
+      "learning_rate": 0.00032458774330281615,
+      "loss": 0.6060261249542236,
+      "mean_token_accuracy": 0.8223113484680653,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6553665667437436,
+      "eval_loss": 0.6835269331932068,
+      "eval_mean_token_accuracy": 0.8134723968720168,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 89.5092,
+      "eval_samples_per_second": 15.864,
+      "eval_steps_per_second": 1.989,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6301972426474094,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.9657333493232727,
+      "learning_rate": 0.0003226476517478835,
+      "loss": 0.6277759075164795,
+      "mean_token_accuracy": 0.8164266042411328,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6594253118788258,
+      "eval_loss": 0.6820935606956482,
+      "eval_mean_token_accuracy": 0.8132368198941263,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 89.6211,
+      "eval_samples_per_second": 15.844,
+      "eval_steps_per_second": 1.986,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5346100655469027,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8541185259819031,
+      "learning_rate": 0.0003206678607662996,
+      "loss": 0.5261692047119141,
+      "mean_token_accuracy": 0.8382289425119177,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5626645662476507,
+      "eval_loss": 0.7028542160987854,
+      "eval_mean_token_accuracy": 0.8133395572056931,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 89.1267,
+      "eval_samples_per_second": 15.932,
+      "eval_steps_per_second": 1.997,
+      "step": 1260
+    },
+    {
+      "entropy": 0.5001101832836866,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.8208815455436707,
+      "learning_rate": 0.0003186489306154935,
+      "loss": 0.4850132942199707,
+      "mean_token_accuracy": 0.8468983843922615,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5889462046743779,
+      "eval_loss": 0.691015899181366,
+      "eval_mean_token_accuracy": 0.8135226466012805,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.9677,
+      "eval_samples_per_second": 15.961,
+      "eval_steps_per_second": 2.001,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5242766339331866,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 0.7893072962760925,
+      "learning_rate": 0.0003165914326288163,
+      "loss": 0.5061577320098877,
+      "mean_token_accuracy": 0.8425608821213245,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5543691962957382,
+      "eval_loss": 0.6913734078407288,
+      "eval_mean_token_accuracy": 0.8154889680026622,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 89.6963,
+      "eval_samples_per_second": 15.831,
+      "eval_steps_per_second": 1.984,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5327721010893584,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.8992080092430115,
+      "learning_rate": 0.0003144959490538604,
+      "loss": 0.5068036556243897,
+      "mean_token_accuracy": 0.8420542575418949,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5734540803378887,
+      "eval_loss": 0.6948716044425964,
+      "eval_mean_token_accuracy": 0.8137217944257715,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 89.4597,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5247377116233111,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.9291845560073853,
+      "learning_rate": 0.0003123630728876902,
+      "loss": 0.5023272037506104,
+      "mean_token_accuracy": 0.8435182586312294,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5492362935891312,
+      "eval_loss": 0.6980633735656738,
+      "eval_mean_token_accuracy": 0.8155082919624415,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 89.2124,
+      "eval_samples_per_second": 15.917,
+      "eval_steps_per_second": 1.995,
+      "step": 1340
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.345957024394281e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1360/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1360/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1360/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1360/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1360/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..7ce4b072fcbe5b636867e13937c56fedee3b68b4
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1360/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.02728394274790723,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1360/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1360/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1360/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1360/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1360/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1360/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1360/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1360/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..c043b7a028490c807463aab4b8fc8966566842fd
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1360/trainer_state.json
@@ -0,0 +1,1462 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 3.2776101388050694,
+  "eval_steps": 20,
+  "global_step": 1360,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.024795526266098,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.945375919342041,
+      "learning_rate": 1.6698127428345936e-05,
+      "loss": 1.7825420379638672,
+      "mean_token_accuracy": 0.6315708436071873,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.29921497588747,
+      "eval_loss": 1.169131875038147,
+      "eval_mean_token_accuracy": 0.7185706342204233,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 91.7898,
+      "eval_samples_per_second": 15.47,
+      "eval_steps_per_second": 1.939,
+      "step": 20
+    },
+    {
+      "entropy": 1.0346344470977784,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.7916979789733887,
+      "learning_rate": 3.427510366871008e-05,
+      "loss": 0.9480395317077637,
+      "mean_token_accuracy": 0.7543429024517536,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 0.9673695430326997,
+      "eval_loss": 0.8588430285453796,
+      "eval_mean_token_accuracy": 0.7737233002534073,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 89.8219,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 40
+    },
+    {
+      "entropy": 0.8977566614747048,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7972187995910645,
+      "learning_rate": 5.185207990907422e-05,
+      "loss": 0.8029604911804199,
+      "mean_token_accuracy": 0.7818016350269318,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8436836982041263,
+      "eval_loss": 0.7893033623695374,
+      "eval_mean_token_accuracy": 0.7858914984076211,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 89.9349,
+      "eval_samples_per_second": 15.789,
+      "eval_steps_per_second": 1.979,
+      "step": 60
+    },
+    {
+      "entropy": 0.8650321319699288,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.2128729820251465,
+      "learning_rate": 6.942905614943836e-05,
+      "loss": 0.77417893409729,
+      "mean_token_accuracy": 0.7875479467213153,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8209857180547179,
+      "eval_loss": 0.7572280168533325,
+      "eval_mean_token_accuracy": 0.7925970383574453,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 89.9459,
+      "eval_samples_per_second": 15.787,
+      "eval_steps_per_second": 1.979,
+      "step": 80
+    },
+    {
+      "entropy": 0.8096660938113928,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.1254287958145142,
+      "learning_rate": 8.700603238980251e-05,
+      "loss": 0.7170331001281738,
+      "mean_token_accuracy": 0.7994555257260799,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.792780208788561,
+      "eval_loss": 0.7442984580993652,
+      "eval_mean_token_accuracy": 0.7954704573984896,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 90.294,
+      "eval_samples_per_second": 15.726,
+      "eval_steps_per_second": 1.971,
+      "step": 100
+    },
+    {
+      "entropy": 0.7979829199612141,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2140507698059082,
+      "learning_rate": 0.00010458300863016666,
+      "loss": 0.7141861915588379,
+      "mean_token_accuracy": 0.8000014744699001,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.7938692486018277,
+      "eval_loss": 0.7279797196388245,
+      "eval_mean_token_accuracy": 0.7989798734027348,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 89.7189,
+      "eval_samples_per_second": 15.827,
+      "eval_steps_per_second": 1.984,
+      "step": 120
+    },
+    {
+      "entropy": 0.7916347607970238,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.0506458282470703,
+      "learning_rate": 0.0001221599848705308,
+      "loss": 0.7117055416107178,
+      "mean_token_accuracy": 0.7970046654343605,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.8052781063519167,
+      "eval_loss": 0.722576916217804,
+      "eval_mean_token_accuracy": 0.7982530523551984,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 90.0376,
+      "eval_samples_per_second": 15.771,
+      "eval_steps_per_second": 1.977,
+      "step": 140
+    },
+    {
+      "entropy": 0.7880666613578796,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8409207463264465,
+      "learning_rate": 0.00013973696111089492,
+      "loss": 0.7052523136138916,
+      "mean_token_accuracy": 0.8000123649835587,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7787443134891853,
+      "eval_loss": 0.7137772440910339,
+      "eval_mean_token_accuracy": 0.8014209287220173,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 90.2562,
+      "eval_samples_per_second": 15.733,
+      "eval_steps_per_second": 1.972,
+      "step": 160
+    },
+    {
+      "entropy": 0.7696007996797561,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.0182477235794067,
+      "learning_rate": 0.00015731393735125907,
+      "loss": 0.7008797645568847,
+      "mean_token_accuracy": 0.8007259473204613,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.752259682068664,
+      "eval_loss": 0.7171286344528198,
+      "eval_mean_token_accuracy": 0.80052122574174,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 90.0797,
+      "eval_samples_per_second": 15.764,
+      "eval_steps_per_second": 1.976,
+      "step": 180
+    },
+    {
+      "entropy": 0.7828126326203346,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9553632736206055,
+      "learning_rate": 0.0001748909135916232,
+      "loss": 0.7046853542327881,
+      "mean_token_accuracy": 0.7988562889397144,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7389638169427936,
+      "eval_loss": 0.7187935709953308,
+      "eval_mean_token_accuracy": 0.7995899999409579,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 89.9223,
+      "eval_samples_per_second": 15.791,
+      "eval_steps_per_second": 1.979,
+      "step": 200
+    },
+    {
+      "entropy": 0.7909286297857762,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 1.134459376335144,
+      "learning_rate": 0.00019246788983198735,
+      "loss": 0.6989312171936035,
+      "mean_token_accuracy": 0.7991565138101577,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7496965199374082,
+      "eval_loss": 0.7106321454048157,
+      "eval_mean_token_accuracy": 0.8046183442131857,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 89.6899,
+      "eval_samples_per_second": 15.832,
+      "eval_steps_per_second": 1.985,
+      "step": 220
+    },
+    {
+      "entropy": 0.7651963323354721,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.9557585716247559,
+      "learning_rate": 0.0002100448660723515,
+      "loss": 0.6984559059143066,
+      "mean_token_accuracy": 0.8019823372364044,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7615563079212488,
+      "eval_loss": 0.7098539471626282,
+      "eval_mean_token_accuracy": 0.8047233508544022,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 90.3752,
+      "eval_samples_per_second": 15.712,
+      "eval_steps_per_second": 1.97,
+      "step": 240
+    },
+    {
+      "entropy": 0.787507726252079,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.9279311299324036,
+      "learning_rate": 0.00022762184231271568,
+      "loss": 0.708138370513916,
+      "mean_token_accuracy": 0.8013308703899383,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7856640896100676,
+      "eval_loss": 0.7191570401191711,
+      "eval_mean_token_accuracy": 0.8024677249153008,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 93.7336,
+      "eval_samples_per_second": 15.149,
+      "eval_steps_per_second": 1.899,
+      "step": 260
+    },
+    {
+      "entropy": 0.768489234894514,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.7992776036262512,
+      "learning_rate": 0.0002451988185530798,
+      "loss": 0.7035849571228028,
+      "mean_token_accuracy": 0.8031105332076549,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7682498740346244,
+      "eval_loss": 0.7149022221565247,
+      "eval_mean_token_accuracy": 0.7973497995499814,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 89.7309,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 280
+    },
+    {
+      "entropy": 0.7720955617725849,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.1931514739990234,
+      "learning_rate": 0.0002627757947934439,
+      "loss": 0.6965402603149414,
+      "mean_token_accuracy": 0.8040348663926125,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7214708304807042,
+      "eval_loss": 0.7350823879241943,
+      "eval_mean_token_accuracy": 0.7942911925610532,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 89.6451,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 300
+    },
+    {
+      "entropy": 0.7930095426738262,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8994467258453369,
+      "learning_rate": 0.0002803527710338081,
+      "loss": 0.7331492900848389,
+      "mean_token_accuracy": 0.796185302734375,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.8142535488926963,
+      "eval_loss": 0.7237834334373474,
+      "eval_mean_token_accuracy": 0.799495273761535,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 90.6759,
+      "eval_samples_per_second": 15.66,
+      "eval_steps_per_second": 1.963,
+      "step": 320
+    },
+    {
+      "entropy": 0.7825102441012859,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 1.8224239349365234,
+      "learning_rate": 0.00029792974727417223,
+      "loss": 0.7331175804138184,
+      "mean_token_accuracy": 0.7977667152881622,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7327135443017724,
+      "eval_loss": 0.7388784289360046,
+      "eval_mean_token_accuracy": 0.8007041252730938,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 89.7692,
+      "eval_samples_per_second": 15.818,
+      "eval_steps_per_second": 1.983,
+      "step": 340
+    },
+    {
+      "entropy": 0.819331557303667,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 1.0407779216766357,
+      "learning_rate": 0.0003155067235145364,
+      "loss": 0.7456695556640625,
+      "mean_token_accuracy": 0.7936980701982975,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.8065585420372781,
+      "eval_loss": 0.735519528388977,
+      "eval_mean_token_accuracy": 0.7974206296245704,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 89.4336,
+      "eval_samples_per_second": 15.878,
+      "eval_steps_per_second": 1.99,
+      "step": 360
+    },
+    {
+      "entropy": 0.8114653021097183,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.1209490299224854,
+      "learning_rate": 0.0003330836997549005,
+      "loss": 0.7340899467468261,
+      "mean_token_accuracy": 0.7914272703230381,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.8025527285056168,
+      "eval_loss": 0.7295576930046082,
+      "eval_mean_token_accuracy": 0.7990576628218876,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 89.9579,
+      "eval_samples_per_second": 15.785,
+      "eval_steps_per_second": 1.979,
+      "step": 380
+    },
+    {
+      "entropy": 0.8042497783899307,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 1.2437798976898193,
+      "learning_rate": 0.0003506606759952647,
+      "loss": 0.7381744384765625,
+      "mean_token_accuracy": 0.7967362694442273,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7717011336530193,
+      "eval_loss": 0.7436173558235168,
+      "eval_mean_token_accuracy": 0.7975849482450592,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 89.8411,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 400
+    },
+    {
+      "entropy": 0.7987750010056929,
+      "epoch": 1.012070006035003,
+      "grad_norm": 1.2390918731689453,
+      "learning_rate": 0.000364721224843344,
+      "loss": 0.7317790031433106,
+      "mean_token_accuracy": 0.7986087551364651,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7333170604170038,
+      "eval_loss": 0.7456948757171631,
+      "eval_mean_token_accuracy": 0.7964897577682238,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 89.7288,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 420
+    },
+    {
+      "entropy": 0.7372976846992969,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.1552740335464478,
+      "learning_rate": 0.00036468510102269575,
+      "loss": 0.6734249114990234,
+      "mean_token_accuracy": 0.805834824591875,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.7130355526891987,
+      "eval_loss": 0.7555333375930786,
+      "eval_mean_token_accuracy": 0.7961652781186479,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 89.8713,
+      "eval_samples_per_second": 15.8,
+      "eval_steps_per_second": 1.981,
+      "step": 440
+    },
+    {
+      "entropy": 0.7465668715536594,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 1.1991469860076904,
+      "learning_rate": 0.0003645973816745066,
+      "loss": 0.6871879577636719,
+      "mean_token_accuracy": 0.8042450234293937,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7772979542110743,
+      "eval_loss": 0.7561826705932617,
+      "eval_mean_token_accuracy": 0.7953876427720102,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 89.8395,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 460
+    },
+    {
+      "entropy": 0.7477855801582336,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 1.325486660003662,
+      "learning_rate": 0.00036445809162231435,
+      "loss": 0.6836830139160156,
+      "mean_token_accuracy": 0.8021619468927383,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.7098110676481483,
+      "eval_loss": 0.758007287979126,
+      "eval_mean_token_accuracy": 0.7909953189030122,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 89.6101,
+      "eval_samples_per_second": 15.846,
+      "eval_steps_per_second": 1.986,
+      "step": 480
+    },
+    {
+      "entropy": 0.7632691070437432,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 1.0122262239456177,
+      "learning_rate": 0.0003642672702835562,
+      "loss": 0.7071213245391845,
+      "mean_token_accuracy": 0.8000680930912495,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7645382720432924,
+      "eval_loss": 0.7492606043815613,
+      "eval_mean_token_accuracy": 0.7976858676149604,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 89.6092,
+      "eval_samples_per_second": 15.847,
+      "eval_steps_per_second": 1.986,
+      "step": 500
+    },
+    {
+      "entropy": 0.7623726457357407,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.367774486541748,
+      "learning_rate": 0.00036402497165841384,
+      "loss": 0.6965175151824952,
+      "mean_token_accuracy": 0.8016501650214195,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7868083715438843,
+      "eval_loss": 0.7481449842453003,
+      "eval_mean_token_accuracy": 0.799330582779445,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 89.8241,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 520
+    },
+    {
+      "entropy": 0.7535207405686378,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 1.213614821434021,
+      "learning_rate": 0.00036373126431453207,
+      "loss": 0.7078310489654541,
+      "mean_token_accuracy": 0.7998907208442688,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7829091950748743,
+      "eval_loss": 0.7416072487831116,
+      "eval_mean_token_accuracy": 0.7969146250339036,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 89.4607,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 540
+    },
+    {
+      "entropy": 0.7527363449335098,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 1.0929982662200928,
+      "learning_rate": 0.00036338623136761495,
+      "loss": 0.7058369636535644,
+      "mean_token_accuracy": 0.7997087433934211,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.728419938114252,
+      "eval_loss": 0.744873046875,
+      "eval_mean_token_accuracy": 0.7966634296299366,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 89.6597,
+      "eval_samples_per_second": 15.838,
+      "eval_steps_per_second": 1.985,
+      "step": 560
+    },
+    {
+      "entropy": 0.7406099259853363,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.409364104270935,
+      "learning_rate": 0.00036298997045790515,
+      "loss": 0.6902852058410645,
+      "mean_token_accuracy": 0.805386807769537,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.6991632758231645,
+      "eval_loss": 0.7480319738388062,
+      "eval_mean_token_accuracy": 0.7970068006033308,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 89.8329,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 580
+    },
+    {
+      "entropy": 0.7459449704736472,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.7053111791610718,
+      "learning_rate": 0.00036254259372255277,
+      "loss": 0.7044640064239502,
+      "mean_token_accuracy": 0.8005018755793571,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7454566660891758,
+      "eval_loss": 0.7551484704017639,
+      "eval_mean_token_accuracy": 0.7897093731365846,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 89.8313,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 600
+    },
+    {
+      "entropy": 0.7621391348540782,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 1.174443244934082,
+      "learning_rate": 0.000362044227763882,
+      "loss": 0.7111573696136475,
+      "mean_token_accuracy": 0.798752411454916,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7664387949397055,
+      "eval_loss": 0.7365804314613342,
+      "eval_mean_token_accuracy": 0.7963101200843126,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 89.9623,
+      "eval_samples_per_second": 15.784,
+      "eval_steps_per_second": 1.979,
+      "step": 620
+    },
+    {
+      "entropy": 0.7424944408237935,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.2898184061050415,
+      "learning_rate": 0.000361495013613564,
+      "loss": 0.6913110733032226,
+      "mean_token_accuracy": 0.8009015507996082,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7255479529332579,
+      "eval_loss": 0.7345991730690002,
+      "eval_mean_token_accuracy": 0.8013592945056015,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 89.817,
+      "eval_samples_per_second": 15.81,
+      "eval_steps_per_second": 1.982,
+      "step": 640
+    },
+    {
+      "entropy": 0.7303367637097835,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.1310476064682007,
+      "learning_rate": 0.00036089510669270666,
+      "loss": 0.6875617027282714,
+      "mean_token_accuracy": 0.8035397171974182,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7101308248016271,
+      "eval_loss": 0.7394917011260986,
+      "eval_mean_token_accuracy": 0.7990792942850777,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 89.7431,
+      "eval_samples_per_second": 15.823,
+      "eval_steps_per_second": 1.983,
+      "step": 660
+    },
+    {
+      "entropy": 0.7387678384780884,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.9645389914512634,
+      "learning_rate": 0.0003602446767678725,
+      "loss": 0.6911204338073731,
+      "mean_token_accuracy": 0.8056333385407924,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7281073737010527,
+      "eval_loss": 0.7376708984375,
+      "eval_mean_token_accuracy": 0.7997356415464637,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 90.0338,
+      "eval_samples_per_second": 15.772,
+      "eval_steps_per_second": 1.977,
+      "step": 680
+    },
+    {
+      "entropy": 0.7449554048478604,
+      "epoch": 1.687990343995172,
+      "grad_norm": 1.9069629907608032,
+      "learning_rate": 0.0003595439079030364,
+      "loss": 0.7042286396026611,
+      "mean_token_accuracy": 0.8018639378249646,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7022633817088738,
+      "eval_loss": 0.7320939898490906,
+      "eval_mean_token_accuracy": 0.8006051541044471,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 89.8948,
+      "eval_samples_per_second": 15.796,
+      "eval_steps_per_second": 1.98,
+      "step": 700
+    },
+    {
+      "entropy": 0.723249051719904,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.8333877325057983,
+      "learning_rate": 0.00035879299840749777,
+      "loss": 0.7043714046478271,
+      "mean_token_accuracy": 0.8028345607221127,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.672436946898364,
+      "eval_loss": 0.713233470916748,
+      "eval_mean_token_accuracy": 0.8047760577684038,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 89.6394,
+      "eval_samples_per_second": 15.841,
+      "eval_steps_per_second": 1.986,
+      "step": 720
+    },
+    {
+      "entropy": 0.730014206469059,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.4580037593841553,
+      "learning_rate": 0.00035799216077976137,
+      "loss": 0.6985176563262939,
+      "mean_token_accuracy": 0.8015547141432762,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.6908873795123582,
+      "eval_loss": 0.7198202013969421,
+      "eval_mean_token_accuracy": 0.8024131682481659,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 89.3771,
+      "eval_samples_per_second": 15.888,
+      "eval_steps_per_second": 1.992,
+      "step": 740
+    },
+    {
+      "entropy": 0.7219605796039105,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.51682710647583,
+      "learning_rate": 0.000357141621647403,
+      "loss": 0.6844424724578857,
+      "mean_token_accuracy": 0.8032813094556331,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7113401109582922,
+      "eval_loss": 0.7195309400558472,
+      "eval_mean_token_accuracy": 0.8019187035185568,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 89.4658,
+      "eval_samples_per_second": 15.872,
+      "eval_steps_per_second": 1.99,
+      "step": 760
+    },
+    {
+      "entropy": 0.729826743900776,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 1.0838000774383545,
+      "learning_rate": 0.0003562416217029361,
+      "loss": 0.685858678817749,
+      "mean_token_accuracy": 0.8050567395985126,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.649370267484965,
+      "eval_loss": 0.7177029252052307,
+      "eval_mean_token_accuracy": 0.8050174043419656,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 89.2287,
+      "eval_samples_per_second": 15.914,
+      "eval_steps_per_second": 1.995,
+      "step": 780
+    },
+    {
+      "entropy": 0.6799364153295755,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 1.1364924907684326,
+      "learning_rate": 0.00035529241563569903,
+      "loss": 0.6714428901672364,
+      "mean_token_accuracy": 0.8087341658771038,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6603799047095052,
+      "eval_loss": 0.7201809287071228,
+      "eval_mean_token_accuracy": 0.802863868769635,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 89.2718,
+      "eval_samples_per_second": 15.906,
+      "eval_steps_per_second": 1.994,
+      "step": 800
+    },
+    {
+      "entropy": 0.7078610375523567,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 1.2191942930221558,
+      "learning_rate": 0.0003542942720597808,
+      "loss": 0.6996590614318847,
+      "mean_token_accuracy": 0.8049512051045895,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.7240545264120852,
+      "eval_loss": 0.7301309108734131,
+      "eval_mean_token_accuracy": 0.8015973166133581,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 89.3438,
+      "eval_samples_per_second": 15.894,
+      "eval_steps_per_second": 1.992,
+      "step": 820
+    },
+    {
+      "entropy": 0.6557391556826505,
+      "epoch": 2.024140012070006,
+      "grad_norm": 1.36203134059906,
+      "learning_rate": 0.0003532474734380064,
+      "loss": 0.6327113628387451,
+      "mean_token_accuracy": 0.8142829847026181,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6573808648613062,
+      "eval_loss": 0.7334365844726562,
+      "eval_mean_token_accuracy": 0.8043633893634496,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 89.3359,
+      "eval_samples_per_second": 15.895,
+      "eval_steps_per_second": 1.992,
+      "step": 840
+    },
+    {
+      "entropy": 0.6409128502011299,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 1.5804955959320068,
+      "learning_rate": 0.0003521523160020035,
+      "loss": 0.6023559093475341,
+      "mean_token_accuracy": 0.8222251623868942,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6983234788594621,
+      "eval_loss": 0.7232135534286499,
+      "eval_mean_token_accuracy": 0.8010654268639811,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 89.2615,
+      "eval_samples_per_second": 15.908,
+      "eval_steps_per_second": 1.994,
+      "step": 860
+    },
+    {
+      "entropy": 0.6151149723678827,
+      "epoch": 2.12070006035003,
+      "grad_norm": 1.1975436210632324,
+      "learning_rate": 0.00035100910966837193,
+      "loss": 0.5803564548492431,
+      "mean_token_accuracy": 0.8256325736641884,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6636555573243773,
+      "eval_loss": 0.7205662727355957,
+      "eval_mean_token_accuracy": 0.8036768369460374,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 89.2957,
+      "eval_samples_per_second": 15.902,
+      "eval_steps_per_second": 1.993,
+      "step": 880
+    },
+    {
+      "entropy": 0.6201727617532015,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9907131195068359,
+      "learning_rate": 0.0003498181779509813,
+      "loss": 0.6101872444152832,
+      "mean_token_accuracy": 0.8196512959897518,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.6473788086617931,
+      "eval_loss": 0.7062397003173828,
+      "eval_mean_token_accuracy": 0.8078362184963869,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 88.6198,
+      "eval_samples_per_second": 16.024,
+      "eval_steps_per_second": 2.009,
+      "step": 900
+    },
+    {
+      "entropy": 0.6097332935780286,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.9055633544921875,
+      "learning_rate": 0.00034857985786942026,
+      "loss": 0.5956094264984131,
+      "mean_token_accuracy": 0.8240575887262821,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.612629868341296,
+      "eval_loss": 0.7106770277023315,
+      "eval_mean_token_accuracy": 0.805324277181304,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 89.4383,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 920
+    },
+    {
+      "entropy": 0.6182701248675585,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.9513736367225647,
+      "learning_rate": 0.00034729449985362404,
+      "loss": 0.6019269466400147,
+      "mean_token_accuracy": 0.8187106013298034,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6437647456533453,
+      "eval_loss": 0.7173502445220947,
+      "eval_mean_token_accuracy": 0.8056439573175451,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.9568,
+      "eval_samples_per_second": 15.963,
+      "eval_steps_per_second": 2.001,
+      "step": 940
+    },
+    {
+      "entropy": 0.6508570514619351,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.9606207013130188,
+      "learning_rate": 0.0003459624676447067,
+      "loss": 0.6296409606933594,
+      "mean_token_accuracy": 0.8167732007801533,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6288247135248077,
+      "eval_loss": 0.7160675525665283,
+      "eval_mean_token_accuracy": 0.8066010689467527,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 89.0044,
+      "eval_samples_per_second": 15.954,
+      "eval_steps_per_second": 2.0,
+      "step": 960
+    },
+    {
+      "entropy": 0.6380081418901682,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.9677181839942932,
+      "learning_rate": 0.000344584138192027,
+      "loss": 0.6210546493530273,
+      "mean_token_accuracy": 0.8186135642230511,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6407805031604981,
+      "eval_loss": 0.7055043578147888,
+      "eval_mean_token_accuracy": 0.8070103157772107,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 89.5606,
+      "eval_samples_per_second": 15.855,
+      "eval_steps_per_second": 1.987,
+      "step": 980
+    },
+    {
+      "entropy": 0.6310251638293266,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 2.1046407222747803,
+      "learning_rate": 0.000343159901546516,
+      "loss": 0.6165830135345459,
+      "mean_token_accuracy": 0.8207855716347694,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6461507470420237,
+      "eval_loss": 0.7292136549949646,
+      "eval_mean_token_accuracy": 0.8030418261383356,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 89.315,
+      "eval_samples_per_second": 15.899,
+      "eval_steps_per_second": 1.993,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6256617344915867,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 1.1768349409103394,
+      "learning_rate": 0.0003416901607502972,
+      "loss": 0.6142709255218506,
+      "mean_token_accuracy": 0.8220669947564602,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6150523993406403,
+      "eval_loss": 0.7019563317298889,
+      "eval_mean_token_accuracy": 0.8094323099998946,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 89.419,
+      "eval_samples_per_second": 15.88,
+      "eval_steps_per_second": 1.991,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6126735735684633,
+      "epoch": 2.506940253470127,
+      "grad_norm": 1.8065496683120728,
+      "learning_rate": 0.00034017533172263055,
+      "loss": 0.6056458473205566,
+      "mean_token_accuracy": 0.8203762136399746,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6544345456562685,
+      "eval_loss": 0.7086517214775085,
+      "eval_mean_token_accuracy": 0.806709805901131,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 88.981,
+      "eval_samples_per_second": 15.958,
+      "eval_steps_per_second": 2.0,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6311754353344441,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 1.0576369762420654,
+      "learning_rate": 0.00033861584314221236,
+      "loss": 0.6176392555236816,
+      "mean_token_accuracy": 0.8177410490810871,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6273799200406235,
+      "eval_loss": 0.708163321018219,
+      "eval_mean_token_accuracy": 0.8032017099053672,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 89.7668,
+      "eval_samples_per_second": 15.819,
+      "eval_steps_per_second": 1.983,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6200249589979648,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.997065007686615,
+      "learning_rate": 0.0003370121363258637,
+      "loss": 0.6203888893127442,
+      "mean_token_accuracy": 0.8196275025606156,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6457576811983344,
+      "eval_loss": 0.6972290277481079,
+      "eval_mean_token_accuracy": 0.8095980849158898,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 89.2027,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 1080
+    },
+    {
+      "entropy": 0.626810473203659,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.865475058555603,
+      "learning_rate": 0.0003353646651036438,
+      "loss": 0.6102585792541504,
+      "mean_token_accuracy": 0.819708751142025,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.6412022431914726,
+      "eval_loss": 0.7036139965057373,
+      "eval_mean_token_accuracy": 0.8070317135098275,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 89.4733,
+      "eval_samples_per_second": 15.871,
+      "eval_steps_per_second": 1.989,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6581619590520859,
+      "epoch": 2.700060350030175,
+      "grad_norm": 1.00307035446167,
+      "learning_rate": 0.00033367389569042064,
+      "loss": 0.6397720336914062,
+      "mean_token_accuracy": 0.81413309648633,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6288736402318719,
+      "eval_loss": 0.6966825723648071,
+      "eval_mean_token_accuracy": 0.8094039669867312,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 88.7341,
+      "eval_samples_per_second": 16.003,
+      "eval_steps_per_second": 2.006,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6399412982165813,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.8589329719543457,
+      "learning_rate": 0.0003319403065539384,
+      "loss": 0.6347338199615479,
+      "mean_token_accuracy": 0.8139534957706929,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.683491913455256,
+      "eval_loss": 0.6984149217605591,
+      "eval_mean_token_accuracy": 0.8083989469522841,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 89.3929,
+      "eval_samples_per_second": 15.885,
+      "eval_steps_per_second": 1.991,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6264835461974144,
+      "epoch": 2.796620398310199,
+      "grad_norm": 0.7807853817939758,
+      "learning_rate": 0.0003301643882794163,
+      "loss": 0.6162629127502441,
+      "mean_token_accuracy": 0.8216292977333068,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6578792159476977,
+      "eval_loss": 0.6921441555023193,
+      "eval_mean_token_accuracy": 0.8094299177775223,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 89.2356,
+      "eval_samples_per_second": 15.913,
+      "eval_steps_per_second": 1.995,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6269129011780024,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 1.1675593852996826,
+      "learning_rate": 0.0003283466434307189,
+      "loss": 0.6235532760620117,
+      "mean_token_accuracy": 0.8188897788524627,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6533248447970058,
+      "eval_loss": 0.696849524974823,
+      "eval_mean_token_accuracy": 0.8102703713968898,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2602,
+      "eval_samples_per_second": 16.089,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.6259873781353236,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.9596486687660217,
+      "learning_rate": 0.00032648758640813655,
+      "loss": 0.6118728160858155,
+      "mean_token_accuracy": 0.8171859130263328,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6212078754821521,
+      "eval_loss": 0.6942870020866394,
+      "eval_mean_token_accuracy": 0.810484343365337,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 89.4375,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 1200
+    },
+    {
+      "entropy": 0.6142537571489811,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 0.9625058770179749,
+      "learning_rate": 0.00032458774330281615,
+      "loss": 0.6060261249542236,
+      "mean_token_accuracy": 0.8223113484680653,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6553665667437436,
+      "eval_loss": 0.6835269331932068,
+      "eval_mean_token_accuracy": 0.8134723968720168,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 89.5092,
+      "eval_samples_per_second": 15.864,
+      "eval_steps_per_second": 1.989,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6301972426474094,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.9657333493232727,
+      "learning_rate": 0.0003226476517478835,
+      "loss": 0.6277759075164795,
+      "mean_token_accuracy": 0.8164266042411328,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6594253118788258,
+      "eval_loss": 0.6820935606956482,
+      "eval_mean_token_accuracy": 0.8132368198941263,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 89.6211,
+      "eval_samples_per_second": 15.844,
+      "eval_steps_per_second": 1.986,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5346100655469027,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8541185259819031,
+      "learning_rate": 0.0003206678607662996,
+      "loss": 0.5261692047119141,
+      "mean_token_accuracy": 0.8382289425119177,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5626645662476507,
+      "eval_loss": 0.7028542160987854,
+      "eval_mean_token_accuracy": 0.8133395572056931,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 89.1267,
+      "eval_samples_per_second": 15.932,
+      "eval_steps_per_second": 1.997,
+      "step": 1260
+    },
+    {
+      "entropy": 0.5001101832836866,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.8208815455436707,
+      "learning_rate": 0.0003186489306154935,
+      "loss": 0.4850132942199707,
+      "mean_token_accuracy": 0.8468983843922615,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5889462046743779,
+      "eval_loss": 0.691015899181366,
+      "eval_mean_token_accuracy": 0.8135226466012805,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.9677,
+      "eval_samples_per_second": 15.961,
+      "eval_steps_per_second": 2.001,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5242766339331866,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 0.7893072962760925,
+      "learning_rate": 0.0003165914326288163,
+      "loss": 0.5061577320098877,
+      "mean_token_accuracy": 0.8425608821213245,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5543691962957382,
+      "eval_loss": 0.6913734078407288,
+      "eval_mean_token_accuracy": 0.8154889680026622,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 89.6963,
+      "eval_samples_per_second": 15.831,
+      "eval_steps_per_second": 1.984,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5327721010893584,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.8992080092430115,
+      "learning_rate": 0.0003144959490538604,
+      "loss": 0.5068036556243897,
+      "mean_token_accuracy": 0.8420542575418949,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5734540803378887,
+      "eval_loss": 0.6948716044425964,
+      "eval_mean_token_accuracy": 0.8137217944257715,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 89.4597,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5247377116233111,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.9291845560073853,
+      "learning_rate": 0.0003123630728876902,
+      "loss": 0.5023272037506104,
+      "mean_token_accuracy": 0.8435182586312294,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5492362935891312,
+      "eval_loss": 0.6980633735656738,
+      "eval_mean_token_accuracy": 0.8155082919624415,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 89.2124,
+      "eval_samples_per_second": 15.917,
+      "eval_steps_per_second": 1.995,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5380499072372913,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 1.0477524995803833,
+      "learning_rate": 0.00031019340770903136,
+      "loss": 0.5264591693878173,
+      "mean_token_accuracy": 0.8391959741711617,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5896503307511297,
+      "eval_loss": 0.6997144222259521,
+      "eval_mean_token_accuracy": 0.8124373062942805,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 89.1147,
+      "eval_samples_per_second": 15.935,
+      "eval_steps_per_second": 1.997,
+      "step": 1360
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.3663042912097894e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1380/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1380/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1380/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1380/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1380/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..7ce4b072fcbe5b636867e13937c56fedee3b68b4
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1380/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.02728394274790723,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1380/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1380/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1380/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1380/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1380/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1380/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1380/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1380/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..957854d293c6f9d48ee1d82af92b96066ed35c91
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1380/trainer_state.json
@@ -0,0 +1,1483 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 3.3258901629450817,
+  "eval_steps": 20,
+  "global_step": 1380,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.024795526266098,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.945375919342041,
+      "learning_rate": 1.6698127428345936e-05,
+      "loss": 1.7825420379638672,
+      "mean_token_accuracy": 0.6315708436071873,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.29921497588747,
+      "eval_loss": 1.169131875038147,
+      "eval_mean_token_accuracy": 0.7185706342204233,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 91.7898,
+      "eval_samples_per_second": 15.47,
+      "eval_steps_per_second": 1.939,
+      "step": 20
+    },
+    {
+      "entropy": 1.0346344470977784,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.7916979789733887,
+      "learning_rate": 3.427510366871008e-05,
+      "loss": 0.9480395317077637,
+      "mean_token_accuracy": 0.7543429024517536,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 0.9673695430326997,
+      "eval_loss": 0.8588430285453796,
+      "eval_mean_token_accuracy": 0.7737233002534073,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 89.8219,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 40
+    },
+    {
+      "entropy": 0.8977566614747048,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7972187995910645,
+      "learning_rate": 5.185207990907422e-05,
+      "loss": 0.8029604911804199,
+      "mean_token_accuracy": 0.7818016350269318,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8436836982041263,
+      "eval_loss": 0.7893033623695374,
+      "eval_mean_token_accuracy": 0.7858914984076211,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 89.9349,
+      "eval_samples_per_second": 15.789,
+      "eval_steps_per_second": 1.979,
+      "step": 60
+    },
+    {
+      "entropy": 0.8650321319699288,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.2128729820251465,
+      "learning_rate": 6.942905614943836e-05,
+      "loss": 0.77417893409729,
+      "mean_token_accuracy": 0.7875479467213153,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8209857180547179,
+      "eval_loss": 0.7572280168533325,
+      "eval_mean_token_accuracy": 0.7925970383574453,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 89.9459,
+      "eval_samples_per_second": 15.787,
+      "eval_steps_per_second": 1.979,
+      "step": 80
+    },
+    {
+      "entropy": 0.8096660938113928,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.1254287958145142,
+      "learning_rate": 8.700603238980251e-05,
+      "loss": 0.7170331001281738,
+      "mean_token_accuracy": 0.7994555257260799,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.792780208788561,
+      "eval_loss": 0.7442984580993652,
+      "eval_mean_token_accuracy": 0.7954704573984896,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 90.294,
+      "eval_samples_per_second": 15.726,
+      "eval_steps_per_second": 1.971,
+      "step": 100
+    },
+    {
+      "entropy": 0.7979829199612141,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2140507698059082,
+      "learning_rate": 0.00010458300863016666,
+      "loss": 0.7141861915588379,
+      "mean_token_accuracy": 0.8000014744699001,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.7938692486018277,
+      "eval_loss": 0.7279797196388245,
+      "eval_mean_token_accuracy": 0.7989798734027348,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 89.7189,
+      "eval_samples_per_second": 15.827,
+      "eval_steps_per_second": 1.984,
+      "step": 120
+    },
+    {
+      "entropy": 0.7916347607970238,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.0506458282470703,
+      "learning_rate": 0.0001221599848705308,
+      "loss": 0.7117055416107178,
+      "mean_token_accuracy": 0.7970046654343605,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.8052781063519167,
+      "eval_loss": 0.722576916217804,
+      "eval_mean_token_accuracy": 0.7982530523551984,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 90.0376,
+      "eval_samples_per_second": 15.771,
+      "eval_steps_per_second": 1.977,
+      "step": 140
+    },
+    {
+      "entropy": 0.7880666613578796,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8409207463264465,
+      "learning_rate": 0.00013973696111089492,
+      "loss": 0.7052523136138916,
+      "mean_token_accuracy": 0.8000123649835587,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7787443134891853,
+      "eval_loss": 0.7137772440910339,
+      "eval_mean_token_accuracy": 0.8014209287220173,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 90.2562,
+      "eval_samples_per_second": 15.733,
+      "eval_steps_per_second": 1.972,
+      "step": 160
+    },
+    {
+      "entropy": 0.7696007996797561,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.0182477235794067,
+      "learning_rate": 0.00015731393735125907,
+      "loss": 0.7008797645568847,
+      "mean_token_accuracy": 0.8007259473204613,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.752259682068664,
+      "eval_loss": 0.7171286344528198,
+      "eval_mean_token_accuracy": 0.80052122574174,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 90.0797,
+      "eval_samples_per_second": 15.764,
+      "eval_steps_per_second": 1.976,
+      "step": 180
+    },
+    {
+      "entropy": 0.7828126326203346,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9553632736206055,
+      "learning_rate": 0.0001748909135916232,
+      "loss": 0.7046853542327881,
+      "mean_token_accuracy": 0.7988562889397144,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7389638169427936,
+      "eval_loss": 0.7187935709953308,
+      "eval_mean_token_accuracy": 0.7995899999409579,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 89.9223,
+      "eval_samples_per_second": 15.791,
+      "eval_steps_per_second": 1.979,
+      "step": 200
+    },
+    {
+      "entropy": 0.7909286297857762,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 1.134459376335144,
+      "learning_rate": 0.00019246788983198735,
+      "loss": 0.6989312171936035,
+      "mean_token_accuracy": 0.7991565138101577,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7496965199374082,
+      "eval_loss": 0.7106321454048157,
+      "eval_mean_token_accuracy": 0.8046183442131857,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 89.6899,
+      "eval_samples_per_second": 15.832,
+      "eval_steps_per_second": 1.985,
+      "step": 220
+    },
+    {
+      "entropy": 0.7651963323354721,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.9557585716247559,
+      "learning_rate": 0.0002100448660723515,
+      "loss": 0.6984559059143066,
+      "mean_token_accuracy": 0.8019823372364044,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7615563079212488,
+      "eval_loss": 0.7098539471626282,
+      "eval_mean_token_accuracy": 0.8047233508544022,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 90.3752,
+      "eval_samples_per_second": 15.712,
+      "eval_steps_per_second": 1.97,
+      "step": 240
+    },
+    {
+      "entropy": 0.787507726252079,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.9279311299324036,
+      "learning_rate": 0.00022762184231271568,
+      "loss": 0.708138370513916,
+      "mean_token_accuracy": 0.8013308703899383,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7856640896100676,
+      "eval_loss": 0.7191570401191711,
+      "eval_mean_token_accuracy": 0.8024677249153008,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 93.7336,
+      "eval_samples_per_second": 15.149,
+      "eval_steps_per_second": 1.899,
+      "step": 260
+    },
+    {
+      "entropy": 0.768489234894514,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.7992776036262512,
+      "learning_rate": 0.0002451988185530798,
+      "loss": 0.7035849571228028,
+      "mean_token_accuracy": 0.8031105332076549,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7682498740346244,
+      "eval_loss": 0.7149022221565247,
+      "eval_mean_token_accuracy": 0.7973497995499814,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 89.7309,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 280
+    },
+    {
+      "entropy": 0.7720955617725849,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.1931514739990234,
+      "learning_rate": 0.0002627757947934439,
+      "loss": 0.6965402603149414,
+      "mean_token_accuracy": 0.8040348663926125,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7214708304807042,
+      "eval_loss": 0.7350823879241943,
+      "eval_mean_token_accuracy": 0.7942911925610532,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 89.6451,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 300
+    },
+    {
+      "entropy": 0.7930095426738262,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8994467258453369,
+      "learning_rate": 0.0002803527710338081,
+      "loss": 0.7331492900848389,
+      "mean_token_accuracy": 0.796185302734375,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.8142535488926963,
+      "eval_loss": 0.7237834334373474,
+      "eval_mean_token_accuracy": 0.799495273761535,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 90.6759,
+      "eval_samples_per_second": 15.66,
+      "eval_steps_per_second": 1.963,
+      "step": 320
+    },
+    {
+      "entropy": 0.7825102441012859,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 1.8224239349365234,
+      "learning_rate": 0.00029792974727417223,
+      "loss": 0.7331175804138184,
+      "mean_token_accuracy": 0.7977667152881622,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7327135443017724,
+      "eval_loss": 0.7388784289360046,
+      "eval_mean_token_accuracy": 0.8007041252730938,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 89.7692,
+      "eval_samples_per_second": 15.818,
+      "eval_steps_per_second": 1.983,
+      "step": 340
+    },
+    {
+      "entropy": 0.819331557303667,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 1.0407779216766357,
+      "learning_rate": 0.0003155067235145364,
+      "loss": 0.7456695556640625,
+      "mean_token_accuracy": 0.7936980701982975,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.8065585420372781,
+      "eval_loss": 0.735519528388977,
+      "eval_mean_token_accuracy": 0.7974206296245704,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 89.4336,
+      "eval_samples_per_second": 15.878,
+      "eval_steps_per_second": 1.99,
+      "step": 360
+    },
+    {
+      "entropy": 0.8114653021097183,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.1209490299224854,
+      "learning_rate": 0.0003330836997549005,
+      "loss": 0.7340899467468261,
+      "mean_token_accuracy": 0.7914272703230381,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.8025527285056168,
+      "eval_loss": 0.7295576930046082,
+      "eval_mean_token_accuracy": 0.7990576628218876,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 89.9579,
+      "eval_samples_per_second": 15.785,
+      "eval_steps_per_second": 1.979,
+      "step": 380
+    },
+    {
+      "entropy": 0.8042497783899307,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 1.2437798976898193,
+      "learning_rate": 0.0003506606759952647,
+      "loss": 0.7381744384765625,
+      "mean_token_accuracy": 0.7967362694442273,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7717011336530193,
+      "eval_loss": 0.7436173558235168,
+      "eval_mean_token_accuracy": 0.7975849482450592,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 89.8411,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 400
+    },
+    {
+      "entropy": 0.7987750010056929,
+      "epoch": 1.012070006035003,
+      "grad_norm": 1.2390918731689453,
+      "learning_rate": 0.000364721224843344,
+      "loss": 0.7317790031433106,
+      "mean_token_accuracy": 0.7986087551364651,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7333170604170038,
+      "eval_loss": 0.7456948757171631,
+      "eval_mean_token_accuracy": 0.7964897577682238,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 89.7288,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 420
+    },
+    {
+      "entropy": 0.7372976846992969,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.1552740335464478,
+      "learning_rate": 0.00036468510102269575,
+      "loss": 0.6734249114990234,
+      "mean_token_accuracy": 0.805834824591875,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.7130355526891987,
+      "eval_loss": 0.7555333375930786,
+      "eval_mean_token_accuracy": 0.7961652781186479,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 89.8713,
+      "eval_samples_per_second": 15.8,
+      "eval_steps_per_second": 1.981,
+      "step": 440
+    },
+    {
+      "entropy": 0.7465668715536594,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 1.1991469860076904,
+      "learning_rate": 0.0003645973816745066,
+      "loss": 0.6871879577636719,
+      "mean_token_accuracy": 0.8042450234293937,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7772979542110743,
+      "eval_loss": 0.7561826705932617,
+      "eval_mean_token_accuracy": 0.7953876427720102,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 89.8395,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 460
+    },
+    {
+      "entropy": 0.7477855801582336,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 1.325486660003662,
+      "learning_rate": 0.00036445809162231435,
+      "loss": 0.6836830139160156,
+      "mean_token_accuracy": 0.8021619468927383,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.7098110676481483,
+      "eval_loss": 0.758007287979126,
+      "eval_mean_token_accuracy": 0.7909953189030122,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 89.6101,
+      "eval_samples_per_second": 15.846,
+      "eval_steps_per_second": 1.986,
+      "step": 480
+    },
+    {
+      "entropy": 0.7632691070437432,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 1.0122262239456177,
+      "learning_rate": 0.0003642672702835562,
+      "loss": 0.7071213245391845,
+      "mean_token_accuracy": 0.8000680930912495,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7645382720432924,
+      "eval_loss": 0.7492606043815613,
+      "eval_mean_token_accuracy": 0.7976858676149604,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 89.6092,
+      "eval_samples_per_second": 15.847,
+      "eval_steps_per_second": 1.986,
+      "step": 500
+    },
+    {
+      "entropy": 0.7623726457357407,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.367774486541748,
+      "learning_rate": 0.00036402497165841384,
+      "loss": 0.6965175151824952,
+      "mean_token_accuracy": 0.8016501650214195,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7868083715438843,
+      "eval_loss": 0.7481449842453003,
+      "eval_mean_token_accuracy": 0.799330582779445,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 89.8241,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 520
+    },
+    {
+      "entropy": 0.7535207405686378,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 1.213614821434021,
+      "learning_rate": 0.00036373126431453207,
+      "loss": 0.7078310489654541,
+      "mean_token_accuracy": 0.7998907208442688,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7829091950748743,
+      "eval_loss": 0.7416072487831116,
+      "eval_mean_token_accuracy": 0.7969146250339036,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 89.4607,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 540
+    },
+    {
+      "entropy": 0.7527363449335098,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 1.0929982662200928,
+      "learning_rate": 0.00036338623136761495,
+      "loss": 0.7058369636535644,
+      "mean_token_accuracy": 0.7997087433934211,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.728419938114252,
+      "eval_loss": 0.744873046875,
+      "eval_mean_token_accuracy": 0.7966634296299366,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 89.6597,
+      "eval_samples_per_second": 15.838,
+      "eval_steps_per_second": 1.985,
+      "step": 560
+    },
+    {
+      "entropy": 0.7406099259853363,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.409364104270935,
+      "learning_rate": 0.00036298997045790515,
+      "loss": 0.6902852058410645,
+      "mean_token_accuracy": 0.805386807769537,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.6991632758231645,
+      "eval_loss": 0.7480319738388062,
+      "eval_mean_token_accuracy": 0.7970068006033308,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 89.8329,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 580
+    },
+    {
+      "entropy": 0.7459449704736472,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.7053111791610718,
+      "learning_rate": 0.00036254259372255277,
+      "loss": 0.7044640064239502,
+      "mean_token_accuracy": 0.8005018755793571,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7454566660891758,
+      "eval_loss": 0.7551484704017639,
+      "eval_mean_token_accuracy": 0.7897093731365846,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 89.8313,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 600
+    },
+    {
+      "entropy": 0.7621391348540782,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 1.174443244934082,
+      "learning_rate": 0.000362044227763882,
+      "loss": 0.7111573696136475,
+      "mean_token_accuracy": 0.798752411454916,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7664387949397055,
+      "eval_loss": 0.7365804314613342,
+      "eval_mean_token_accuracy": 0.7963101200843126,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 89.9623,
+      "eval_samples_per_second": 15.784,
+      "eval_steps_per_second": 1.979,
+      "step": 620
+    },
+    {
+      "entropy": 0.7424944408237935,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.2898184061050415,
+      "learning_rate": 0.000361495013613564,
+      "loss": 0.6913110733032226,
+      "mean_token_accuracy": 0.8009015507996082,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7255479529332579,
+      "eval_loss": 0.7345991730690002,
+      "eval_mean_token_accuracy": 0.8013592945056015,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 89.817,
+      "eval_samples_per_second": 15.81,
+      "eval_steps_per_second": 1.982,
+      "step": 640
+    },
+    {
+      "entropy": 0.7303367637097835,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.1310476064682007,
+      "learning_rate": 0.00036089510669270666,
+      "loss": 0.6875617027282714,
+      "mean_token_accuracy": 0.8035397171974182,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7101308248016271,
+      "eval_loss": 0.7394917011260986,
+      "eval_mean_token_accuracy": 0.7990792942850777,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 89.7431,
+      "eval_samples_per_second": 15.823,
+      "eval_steps_per_second": 1.983,
+      "step": 660
+    },
+    {
+      "entropy": 0.7387678384780884,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.9645389914512634,
+      "learning_rate": 0.0003602446767678725,
+      "loss": 0.6911204338073731,
+      "mean_token_accuracy": 0.8056333385407924,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7281073737010527,
+      "eval_loss": 0.7376708984375,
+      "eval_mean_token_accuracy": 0.7997356415464637,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 90.0338,
+      "eval_samples_per_second": 15.772,
+      "eval_steps_per_second": 1.977,
+      "step": 680
+    },
+    {
+      "entropy": 0.7449554048478604,
+      "epoch": 1.687990343995172,
+      "grad_norm": 1.9069629907608032,
+      "learning_rate": 0.0003595439079030364,
+      "loss": 0.7042286396026611,
+      "mean_token_accuracy": 0.8018639378249646,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7022633817088738,
+      "eval_loss": 0.7320939898490906,
+      "eval_mean_token_accuracy": 0.8006051541044471,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 89.8948,
+      "eval_samples_per_second": 15.796,
+      "eval_steps_per_second": 1.98,
+      "step": 700
+    },
+    {
+      "entropy": 0.723249051719904,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.8333877325057983,
+      "learning_rate": 0.00035879299840749777,
+      "loss": 0.7043714046478271,
+      "mean_token_accuracy": 0.8028345607221127,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.672436946898364,
+      "eval_loss": 0.713233470916748,
+      "eval_mean_token_accuracy": 0.8047760577684038,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 89.6394,
+      "eval_samples_per_second": 15.841,
+      "eval_steps_per_second": 1.986,
+      "step": 720
+    },
+    {
+      "entropy": 0.730014206469059,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.4580037593841553,
+      "learning_rate": 0.00035799216077976137,
+      "loss": 0.6985176563262939,
+      "mean_token_accuracy": 0.8015547141432762,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.6908873795123582,
+      "eval_loss": 0.7198202013969421,
+      "eval_mean_token_accuracy": 0.8024131682481659,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 89.3771,
+      "eval_samples_per_second": 15.888,
+      "eval_steps_per_second": 1.992,
+      "step": 740
+    },
+    {
+      "entropy": 0.7219605796039105,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.51682710647583,
+      "learning_rate": 0.000357141621647403,
+      "loss": 0.6844424724578857,
+      "mean_token_accuracy": 0.8032813094556331,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7113401109582922,
+      "eval_loss": 0.7195309400558472,
+      "eval_mean_token_accuracy": 0.8019187035185568,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 89.4658,
+      "eval_samples_per_second": 15.872,
+      "eval_steps_per_second": 1.99,
+      "step": 760
+    },
+    {
+      "entropy": 0.729826743900776,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 1.0838000774383545,
+      "learning_rate": 0.0003562416217029361,
+      "loss": 0.685858678817749,
+      "mean_token_accuracy": 0.8050567395985126,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.649370267484965,
+      "eval_loss": 0.7177029252052307,
+      "eval_mean_token_accuracy": 0.8050174043419656,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 89.2287,
+      "eval_samples_per_second": 15.914,
+      "eval_steps_per_second": 1.995,
+      "step": 780
+    },
+    {
+      "entropy": 0.6799364153295755,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 1.1364924907684326,
+      "learning_rate": 0.00035529241563569903,
+      "loss": 0.6714428901672364,
+      "mean_token_accuracy": 0.8087341658771038,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6603799047095052,
+      "eval_loss": 0.7201809287071228,
+      "eval_mean_token_accuracy": 0.802863868769635,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 89.2718,
+      "eval_samples_per_second": 15.906,
+      "eval_steps_per_second": 1.994,
+      "step": 800
+    },
+    {
+      "entropy": 0.7078610375523567,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 1.2191942930221558,
+      "learning_rate": 0.0003542942720597808,
+      "loss": 0.6996590614318847,
+      "mean_token_accuracy": 0.8049512051045895,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.7240545264120852,
+      "eval_loss": 0.7301309108734131,
+      "eval_mean_token_accuracy": 0.8015973166133581,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 89.3438,
+      "eval_samples_per_second": 15.894,
+      "eval_steps_per_second": 1.992,
+      "step": 820
+    },
+    {
+      "entropy": 0.6557391556826505,
+      "epoch": 2.024140012070006,
+      "grad_norm": 1.36203134059906,
+      "learning_rate": 0.0003532474734380064,
+      "loss": 0.6327113628387451,
+      "mean_token_accuracy": 0.8142829847026181,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6573808648613062,
+      "eval_loss": 0.7334365844726562,
+      "eval_mean_token_accuracy": 0.8043633893634496,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 89.3359,
+      "eval_samples_per_second": 15.895,
+      "eval_steps_per_second": 1.992,
+      "step": 840
+    },
+    {
+      "entropy": 0.6409128502011299,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 1.5804955959320068,
+      "learning_rate": 0.0003521523160020035,
+      "loss": 0.6023559093475341,
+      "mean_token_accuracy": 0.8222251623868942,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6983234788594621,
+      "eval_loss": 0.7232135534286499,
+      "eval_mean_token_accuracy": 0.8010654268639811,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 89.2615,
+      "eval_samples_per_second": 15.908,
+      "eval_steps_per_second": 1.994,
+      "step": 860
+    },
+    {
+      "entropy": 0.6151149723678827,
+      "epoch": 2.12070006035003,
+      "grad_norm": 1.1975436210632324,
+      "learning_rate": 0.00035100910966837193,
+      "loss": 0.5803564548492431,
+      "mean_token_accuracy": 0.8256325736641884,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6636555573243773,
+      "eval_loss": 0.7205662727355957,
+      "eval_mean_token_accuracy": 0.8036768369460374,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 89.2957,
+      "eval_samples_per_second": 15.902,
+      "eval_steps_per_second": 1.993,
+      "step": 880
+    },
+    {
+      "entropy": 0.6201727617532015,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9907131195068359,
+      "learning_rate": 0.0003498181779509813,
+      "loss": 0.6101872444152832,
+      "mean_token_accuracy": 0.8196512959897518,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.6473788086617931,
+      "eval_loss": 0.7062397003173828,
+      "eval_mean_token_accuracy": 0.8078362184963869,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 88.6198,
+      "eval_samples_per_second": 16.024,
+      "eval_steps_per_second": 2.009,
+      "step": 900
+    },
+    {
+      "entropy": 0.6097332935780286,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.9055633544921875,
+      "learning_rate": 0.00034857985786942026,
+      "loss": 0.5956094264984131,
+      "mean_token_accuracy": 0.8240575887262821,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.612629868341296,
+      "eval_loss": 0.7106770277023315,
+      "eval_mean_token_accuracy": 0.805324277181304,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 89.4383,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 920
+    },
+    {
+      "entropy": 0.6182701248675585,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.9513736367225647,
+      "learning_rate": 0.00034729449985362404,
+      "loss": 0.6019269466400147,
+      "mean_token_accuracy": 0.8187106013298034,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6437647456533453,
+      "eval_loss": 0.7173502445220947,
+      "eval_mean_token_accuracy": 0.8056439573175451,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.9568,
+      "eval_samples_per_second": 15.963,
+      "eval_steps_per_second": 2.001,
+      "step": 940
+    },
+    {
+      "entropy": 0.6508570514619351,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.9606207013130188,
+      "learning_rate": 0.0003459624676447067,
+      "loss": 0.6296409606933594,
+      "mean_token_accuracy": 0.8167732007801533,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6288247135248077,
+      "eval_loss": 0.7160675525665283,
+      "eval_mean_token_accuracy": 0.8066010689467527,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 89.0044,
+      "eval_samples_per_second": 15.954,
+      "eval_steps_per_second": 2.0,
+      "step": 960
+    },
+    {
+      "entropy": 0.6380081418901682,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.9677181839942932,
+      "learning_rate": 0.000344584138192027,
+      "loss": 0.6210546493530273,
+      "mean_token_accuracy": 0.8186135642230511,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6407805031604981,
+      "eval_loss": 0.7055043578147888,
+      "eval_mean_token_accuracy": 0.8070103157772107,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 89.5606,
+      "eval_samples_per_second": 15.855,
+      "eval_steps_per_second": 1.987,
+      "step": 980
+    },
+    {
+      "entropy": 0.6310251638293266,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 2.1046407222747803,
+      "learning_rate": 0.000343159901546516,
+      "loss": 0.6165830135345459,
+      "mean_token_accuracy": 0.8207855716347694,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6461507470420237,
+      "eval_loss": 0.7292136549949646,
+      "eval_mean_token_accuracy": 0.8030418261383356,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 89.315,
+      "eval_samples_per_second": 15.899,
+      "eval_steps_per_second": 1.993,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6256617344915867,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 1.1768349409103394,
+      "learning_rate": 0.0003416901607502972,
+      "loss": 0.6142709255218506,
+      "mean_token_accuracy": 0.8220669947564602,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6150523993406403,
+      "eval_loss": 0.7019563317298889,
+      "eval_mean_token_accuracy": 0.8094323099998946,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 89.419,
+      "eval_samples_per_second": 15.88,
+      "eval_steps_per_second": 1.991,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6126735735684633,
+      "epoch": 2.506940253470127,
+      "grad_norm": 1.8065496683120728,
+      "learning_rate": 0.00034017533172263055,
+      "loss": 0.6056458473205566,
+      "mean_token_accuracy": 0.8203762136399746,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6544345456562685,
+      "eval_loss": 0.7086517214775085,
+      "eval_mean_token_accuracy": 0.806709805901131,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 88.981,
+      "eval_samples_per_second": 15.958,
+      "eval_steps_per_second": 2.0,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6311754353344441,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 1.0576369762420654,
+      "learning_rate": 0.00033861584314221236,
+      "loss": 0.6176392555236816,
+      "mean_token_accuracy": 0.8177410490810871,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6273799200406235,
+      "eval_loss": 0.708163321018219,
+      "eval_mean_token_accuracy": 0.8032017099053672,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 89.7668,
+      "eval_samples_per_second": 15.819,
+      "eval_steps_per_second": 1.983,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6200249589979648,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.997065007686615,
+      "learning_rate": 0.0003370121363258637,
+      "loss": 0.6203888893127442,
+      "mean_token_accuracy": 0.8196275025606156,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6457576811983344,
+      "eval_loss": 0.6972290277481079,
+      "eval_mean_token_accuracy": 0.8095980849158898,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 89.2027,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 1080
+    },
+    {
+      "entropy": 0.626810473203659,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.865475058555603,
+      "learning_rate": 0.0003353646651036438,
+      "loss": 0.6102585792541504,
+      "mean_token_accuracy": 0.819708751142025,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.6412022431914726,
+      "eval_loss": 0.7036139965057373,
+      "eval_mean_token_accuracy": 0.8070317135098275,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 89.4733,
+      "eval_samples_per_second": 15.871,
+      "eval_steps_per_second": 1.989,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6581619590520859,
+      "epoch": 2.700060350030175,
+      "grad_norm": 1.00307035446167,
+      "learning_rate": 0.00033367389569042064,
+      "loss": 0.6397720336914062,
+      "mean_token_accuracy": 0.81413309648633,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6288736402318719,
+      "eval_loss": 0.6966825723648071,
+      "eval_mean_token_accuracy": 0.8094039669867312,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 88.7341,
+      "eval_samples_per_second": 16.003,
+      "eval_steps_per_second": 2.006,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6399412982165813,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.8589329719543457,
+      "learning_rate": 0.0003319403065539384,
+      "loss": 0.6347338199615479,
+      "mean_token_accuracy": 0.8139534957706929,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.683491913455256,
+      "eval_loss": 0.6984149217605591,
+      "eval_mean_token_accuracy": 0.8083989469522841,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 89.3929,
+      "eval_samples_per_second": 15.885,
+      "eval_steps_per_second": 1.991,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6264835461974144,
+      "epoch": 2.796620398310199,
+      "grad_norm": 0.7807853817939758,
+      "learning_rate": 0.0003301643882794163,
+      "loss": 0.6162629127502441,
+      "mean_token_accuracy": 0.8216292977333068,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6578792159476977,
+      "eval_loss": 0.6921441555023193,
+      "eval_mean_token_accuracy": 0.8094299177775223,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 89.2356,
+      "eval_samples_per_second": 15.913,
+      "eval_steps_per_second": 1.995,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6269129011780024,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 1.1675593852996826,
+      "learning_rate": 0.0003283466434307189,
+      "loss": 0.6235532760620117,
+      "mean_token_accuracy": 0.8188897788524627,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6533248447970058,
+      "eval_loss": 0.696849524974823,
+      "eval_mean_token_accuracy": 0.8102703713968898,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2602,
+      "eval_samples_per_second": 16.089,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.6259873781353236,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.9596486687660217,
+      "learning_rate": 0.00032648758640813655,
+      "loss": 0.6118728160858155,
+      "mean_token_accuracy": 0.8171859130263328,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6212078754821521,
+      "eval_loss": 0.6942870020866394,
+      "eval_mean_token_accuracy": 0.810484343365337,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 89.4375,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 1200
+    },
+    {
+      "entropy": 0.6142537571489811,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 0.9625058770179749,
+      "learning_rate": 0.00032458774330281615,
+      "loss": 0.6060261249542236,
+      "mean_token_accuracy": 0.8223113484680653,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6553665667437436,
+      "eval_loss": 0.6835269331932068,
+      "eval_mean_token_accuracy": 0.8134723968720168,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 89.5092,
+      "eval_samples_per_second": 15.864,
+      "eval_steps_per_second": 1.989,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6301972426474094,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.9657333493232727,
+      "learning_rate": 0.0003226476517478835,
+      "loss": 0.6277759075164795,
+      "mean_token_accuracy": 0.8164266042411328,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6594253118788258,
+      "eval_loss": 0.6820935606956482,
+      "eval_mean_token_accuracy": 0.8132368198941263,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 89.6211,
+      "eval_samples_per_second": 15.844,
+      "eval_steps_per_second": 1.986,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5346100655469027,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8541185259819031,
+      "learning_rate": 0.0003206678607662996,
+      "loss": 0.5261692047119141,
+      "mean_token_accuracy": 0.8382289425119177,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5626645662476507,
+      "eval_loss": 0.7028542160987854,
+      "eval_mean_token_accuracy": 0.8133395572056931,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 89.1267,
+      "eval_samples_per_second": 15.932,
+      "eval_steps_per_second": 1.997,
+      "step": 1260
+    },
+    {
+      "entropy": 0.5001101832836866,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.8208815455436707,
+      "learning_rate": 0.0003186489306154935,
+      "loss": 0.4850132942199707,
+      "mean_token_accuracy": 0.8468983843922615,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5889462046743779,
+      "eval_loss": 0.691015899181366,
+      "eval_mean_token_accuracy": 0.8135226466012805,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.9677,
+      "eval_samples_per_second": 15.961,
+      "eval_steps_per_second": 2.001,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5242766339331866,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 0.7893072962760925,
+      "learning_rate": 0.0003165914326288163,
+      "loss": 0.5061577320098877,
+      "mean_token_accuracy": 0.8425608821213245,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5543691962957382,
+      "eval_loss": 0.6913734078407288,
+      "eval_mean_token_accuracy": 0.8154889680026622,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 89.6963,
+      "eval_samples_per_second": 15.831,
+      "eval_steps_per_second": 1.984,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5327721010893584,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.8992080092430115,
+      "learning_rate": 0.0003144959490538604,
+      "loss": 0.5068036556243897,
+      "mean_token_accuracy": 0.8420542575418949,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5734540803378887,
+      "eval_loss": 0.6948716044425964,
+      "eval_mean_token_accuracy": 0.8137217944257715,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 89.4597,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5247377116233111,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.9291845560073853,
+      "learning_rate": 0.0003123630728876902,
+      "loss": 0.5023272037506104,
+      "mean_token_accuracy": 0.8435182586312294,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5492362935891312,
+      "eval_loss": 0.6980633735656738,
+      "eval_mean_token_accuracy": 0.8155082919624415,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 89.2124,
+      "eval_samples_per_second": 15.917,
+      "eval_steps_per_second": 1.995,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5380499072372913,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 1.0477524995803833,
+      "learning_rate": 0.00031019340770903136,
+      "loss": 0.5264591693878173,
+      "mean_token_accuracy": 0.8391959741711617,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5896503307511297,
+      "eval_loss": 0.6997144222259521,
+      "eval_mean_token_accuracy": 0.8124373062942805,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 89.1147,
+      "eval_samples_per_second": 15.935,
+      "eval_steps_per_second": 1.997,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5460653610527515,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 1.2473069429397583,
+      "learning_rate": 0.00030798756750746476,
+      "loss": 0.5319677829742432,
+      "mean_token_accuracy": 0.8364320226013661,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5659052182114526,
+      "eval_loss": 0.6924049258232117,
+      "eval_mean_token_accuracy": 0.8153053585732921,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 89.6468,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 1380
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.3850107797051802e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-140/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-140/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-140/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-140/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-140/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..7ce4b072fcbe5b636867e13937c56fedee3b68b4
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-140/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.02728394274790723,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-140/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-140/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-140/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-140/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-140/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-140/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-140/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-140/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..5b282ec27f6c30aaece0783991f4d868e9b437be
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-140/trainer_state.json
@@ -0,0 +1,181 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.33796016898008446,
+  "eval_steps": 20,
+  "global_step": 140,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.024795526266098,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.945375919342041,
+      "learning_rate": 1.6698127428345936e-05,
+      "loss": 1.7825420379638672,
+      "mean_token_accuracy": 0.6315708436071873,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.29921497588747,
+      "eval_loss": 1.169131875038147,
+      "eval_mean_token_accuracy": 0.7185706342204233,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 91.7898,
+      "eval_samples_per_second": 15.47,
+      "eval_steps_per_second": 1.939,
+      "step": 20
+    },
+    {
+      "entropy": 1.0346344470977784,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.7916979789733887,
+      "learning_rate": 3.427510366871008e-05,
+      "loss": 0.9480395317077637,
+      "mean_token_accuracy": 0.7543429024517536,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 0.9673695430326997,
+      "eval_loss": 0.8588430285453796,
+      "eval_mean_token_accuracy": 0.7737233002534073,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 89.8219,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 40
+    },
+    {
+      "entropy": 0.8977566614747048,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7972187995910645,
+      "learning_rate": 5.185207990907422e-05,
+      "loss": 0.8029604911804199,
+      "mean_token_accuracy": 0.7818016350269318,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8436836982041263,
+      "eval_loss": 0.7893033623695374,
+      "eval_mean_token_accuracy": 0.7858914984076211,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 89.9349,
+      "eval_samples_per_second": 15.789,
+      "eval_steps_per_second": 1.979,
+      "step": 60
+    },
+    {
+      "entropy": 0.8650321319699288,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.2128729820251465,
+      "learning_rate": 6.942905614943836e-05,
+      "loss": 0.77417893409729,
+      "mean_token_accuracy": 0.7875479467213153,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8209857180547179,
+      "eval_loss": 0.7572280168533325,
+      "eval_mean_token_accuracy": 0.7925970383574453,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 89.9459,
+      "eval_samples_per_second": 15.787,
+      "eval_steps_per_second": 1.979,
+      "step": 80
+    },
+    {
+      "entropy": 0.8096660938113928,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.1254287958145142,
+      "learning_rate": 8.700603238980251e-05,
+      "loss": 0.7170331001281738,
+      "mean_token_accuracy": 0.7994555257260799,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.792780208788561,
+      "eval_loss": 0.7442984580993652,
+      "eval_mean_token_accuracy": 0.7954704573984896,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 90.294,
+      "eval_samples_per_second": 15.726,
+      "eval_steps_per_second": 1.971,
+      "step": 100
+    },
+    {
+      "entropy": 0.7979829199612141,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2140507698059082,
+      "learning_rate": 0.00010458300863016666,
+      "loss": 0.7141861915588379,
+      "mean_token_accuracy": 0.8000014744699001,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.7938692486018277,
+      "eval_loss": 0.7279797196388245,
+      "eval_mean_token_accuracy": 0.7989798734027348,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 89.7189,
+      "eval_samples_per_second": 15.827,
+      "eval_steps_per_second": 1.984,
+      "step": 120
+    },
+    {
+      "entropy": 0.7916347607970238,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.0506458282470703,
+      "learning_rate": 0.0001221599848705308,
+      "loss": 0.7117055416107178,
+      "mean_token_accuracy": 0.7970046654343605,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.8052781063519167,
+      "eval_loss": 0.722576916217804,
+      "eval_mean_token_accuracy": 0.7982530523551984,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 90.0376,
+      "eval_samples_per_second": 15.771,
+      "eval_steps_per_second": 1.977,
+      "step": 140
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.4496025135521792e+16,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1400/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1400/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1400/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1400/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1400/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..7ce4b072fcbe5b636867e13937c56fedee3b68b4
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1400/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.02728394274790723,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1400/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1400/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1400/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1400/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1400/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1400/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1400/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1400/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..ebaff554059785743429b837a43e2e40054c9854
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1400/trainer_state.json
@@ -0,0 +1,1504 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 3.3741701870850935,
+  "eval_steps": 20,
+  "global_step": 1400,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.024795526266098,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.945375919342041,
+      "learning_rate": 1.6698127428345936e-05,
+      "loss": 1.7825420379638672,
+      "mean_token_accuracy": 0.6315708436071873,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.29921497588747,
+      "eval_loss": 1.169131875038147,
+      "eval_mean_token_accuracy": 0.7185706342204233,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 91.7898,
+      "eval_samples_per_second": 15.47,
+      "eval_steps_per_second": 1.939,
+      "step": 20
+    },
+    {
+      "entropy": 1.0346344470977784,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.7916979789733887,
+      "learning_rate": 3.427510366871008e-05,
+      "loss": 0.9480395317077637,
+      "mean_token_accuracy": 0.7543429024517536,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 0.9673695430326997,
+      "eval_loss": 0.8588430285453796,
+      "eval_mean_token_accuracy": 0.7737233002534073,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 89.8219,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 40
+    },
+    {
+      "entropy": 0.8977566614747048,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7972187995910645,
+      "learning_rate": 5.185207990907422e-05,
+      "loss": 0.8029604911804199,
+      "mean_token_accuracy": 0.7818016350269318,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8436836982041263,
+      "eval_loss": 0.7893033623695374,
+      "eval_mean_token_accuracy": 0.7858914984076211,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 89.9349,
+      "eval_samples_per_second": 15.789,
+      "eval_steps_per_second": 1.979,
+      "step": 60
+    },
+    {
+      "entropy": 0.8650321319699288,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.2128729820251465,
+      "learning_rate": 6.942905614943836e-05,
+      "loss": 0.77417893409729,
+      "mean_token_accuracy": 0.7875479467213153,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8209857180547179,
+      "eval_loss": 0.7572280168533325,
+      "eval_mean_token_accuracy": 0.7925970383574453,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 89.9459,
+      "eval_samples_per_second": 15.787,
+      "eval_steps_per_second": 1.979,
+      "step": 80
+    },
+    {
+      "entropy": 0.8096660938113928,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.1254287958145142,
+      "learning_rate": 8.700603238980251e-05,
+      "loss": 0.7170331001281738,
+      "mean_token_accuracy": 0.7994555257260799,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.792780208788561,
+      "eval_loss": 0.7442984580993652,
+      "eval_mean_token_accuracy": 0.7954704573984896,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 90.294,
+      "eval_samples_per_second": 15.726,
+      "eval_steps_per_second": 1.971,
+      "step": 100
+    },
+    {
+      "entropy": 0.7979829199612141,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2140507698059082,
+      "learning_rate": 0.00010458300863016666,
+      "loss": 0.7141861915588379,
+      "mean_token_accuracy": 0.8000014744699001,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.7938692486018277,
+      "eval_loss": 0.7279797196388245,
+      "eval_mean_token_accuracy": 0.7989798734027348,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 89.7189,
+      "eval_samples_per_second": 15.827,
+      "eval_steps_per_second": 1.984,
+      "step": 120
+    },
+    {
+      "entropy": 0.7916347607970238,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.0506458282470703,
+      "learning_rate": 0.0001221599848705308,
+      "loss": 0.7117055416107178,
+      "mean_token_accuracy": 0.7970046654343605,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.8052781063519167,
+      "eval_loss": 0.722576916217804,
+      "eval_mean_token_accuracy": 0.7982530523551984,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 90.0376,
+      "eval_samples_per_second": 15.771,
+      "eval_steps_per_second": 1.977,
+      "step": 140
+    },
+    {
+      "entropy": 0.7880666613578796,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8409207463264465,
+      "learning_rate": 0.00013973696111089492,
+      "loss": 0.7052523136138916,
+      "mean_token_accuracy": 0.8000123649835587,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7787443134891853,
+      "eval_loss": 0.7137772440910339,
+      "eval_mean_token_accuracy": 0.8014209287220173,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 90.2562,
+      "eval_samples_per_second": 15.733,
+      "eval_steps_per_second": 1.972,
+      "step": 160
+    },
+    {
+      "entropy": 0.7696007996797561,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.0182477235794067,
+      "learning_rate": 0.00015731393735125907,
+      "loss": 0.7008797645568847,
+      "mean_token_accuracy": 0.8007259473204613,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.752259682068664,
+      "eval_loss": 0.7171286344528198,
+      "eval_mean_token_accuracy": 0.80052122574174,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 90.0797,
+      "eval_samples_per_second": 15.764,
+      "eval_steps_per_second": 1.976,
+      "step": 180
+    },
+    {
+      "entropy": 0.7828126326203346,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9553632736206055,
+      "learning_rate": 0.0001748909135916232,
+      "loss": 0.7046853542327881,
+      "mean_token_accuracy": 0.7988562889397144,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7389638169427936,
+      "eval_loss": 0.7187935709953308,
+      "eval_mean_token_accuracy": 0.7995899999409579,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 89.9223,
+      "eval_samples_per_second": 15.791,
+      "eval_steps_per_second": 1.979,
+      "step": 200
+    },
+    {
+      "entropy": 0.7909286297857762,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 1.134459376335144,
+      "learning_rate": 0.00019246788983198735,
+      "loss": 0.6989312171936035,
+      "mean_token_accuracy": 0.7991565138101577,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7496965199374082,
+      "eval_loss": 0.7106321454048157,
+      "eval_mean_token_accuracy": 0.8046183442131857,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 89.6899,
+      "eval_samples_per_second": 15.832,
+      "eval_steps_per_second": 1.985,
+      "step": 220
+    },
+    {
+      "entropy": 0.7651963323354721,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.9557585716247559,
+      "learning_rate": 0.0002100448660723515,
+      "loss": 0.6984559059143066,
+      "mean_token_accuracy": 0.8019823372364044,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7615563079212488,
+      "eval_loss": 0.7098539471626282,
+      "eval_mean_token_accuracy": 0.8047233508544022,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 90.3752,
+      "eval_samples_per_second": 15.712,
+      "eval_steps_per_second": 1.97,
+      "step": 240
+    },
+    {
+      "entropy": 0.787507726252079,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.9279311299324036,
+      "learning_rate": 0.00022762184231271568,
+      "loss": 0.708138370513916,
+      "mean_token_accuracy": 0.8013308703899383,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7856640896100676,
+      "eval_loss": 0.7191570401191711,
+      "eval_mean_token_accuracy": 0.8024677249153008,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 93.7336,
+      "eval_samples_per_second": 15.149,
+      "eval_steps_per_second": 1.899,
+      "step": 260
+    },
+    {
+      "entropy": 0.768489234894514,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.7992776036262512,
+      "learning_rate": 0.0002451988185530798,
+      "loss": 0.7035849571228028,
+      "mean_token_accuracy": 0.8031105332076549,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7682498740346244,
+      "eval_loss": 0.7149022221565247,
+      "eval_mean_token_accuracy": 0.7973497995499814,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 89.7309,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 280
+    },
+    {
+      "entropy": 0.7720955617725849,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.1931514739990234,
+      "learning_rate": 0.0002627757947934439,
+      "loss": 0.6965402603149414,
+      "mean_token_accuracy": 0.8040348663926125,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7214708304807042,
+      "eval_loss": 0.7350823879241943,
+      "eval_mean_token_accuracy": 0.7942911925610532,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 89.6451,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 300
+    },
+    {
+      "entropy": 0.7930095426738262,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8994467258453369,
+      "learning_rate": 0.0002803527710338081,
+      "loss": 0.7331492900848389,
+      "mean_token_accuracy": 0.796185302734375,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.8142535488926963,
+      "eval_loss": 0.7237834334373474,
+      "eval_mean_token_accuracy": 0.799495273761535,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 90.6759,
+      "eval_samples_per_second": 15.66,
+      "eval_steps_per_second": 1.963,
+      "step": 320
+    },
+    {
+      "entropy": 0.7825102441012859,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 1.8224239349365234,
+      "learning_rate": 0.00029792974727417223,
+      "loss": 0.7331175804138184,
+      "mean_token_accuracy": 0.7977667152881622,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7327135443017724,
+      "eval_loss": 0.7388784289360046,
+      "eval_mean_token_accuracy": 0.8007041252730938,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 89.7692,
+      "eval_samples_per_second": 15.818,
+      "eval_steps_per_second": 1.983,
+      "step": 340
+    },
+    {
+      "entropy": 0.819331557303667,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 1.0407779216766357,
+      "learning_rate": 0.0003155067235145364,
+      "loss": 0.7456695556640625,
+      "mean_token_accuracy": 0.7936980701982975,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.8065585420372781,
+      "eval_loss": 0.735519528388977,
+      "eval_mean_token_accuracy": 0.7974206296245704,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 89.4336,
+      "eval_samples_per_second": 15.878,
+      "eval_steps_per_second": 1.99,
+      "step": 360
+    },
+    {
+      "entropy": 0.8114653021097183,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.1209490299224854,
+      "learning_rate": 0.0003330836997549005,
+      "loss": 0.7340899467468261,
+      "mean_token_accuracy": 0.7914272703230381,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.8025527285056168,
+      "eval_loss": 0.7295576930046082,
+      "eval_mean_token_accuracy": 0.7990576628218876,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 89.9579,
+      "eval_samples_per_second": 15.785,
+      "eval_steps_per_second": 1.979,
+      "step": 380
+    },
+    {
+      "entropy": 0.8042497783899307,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 1.2437798976898193,
+      "learning_rate": 0.0003506606759952647,
+      "loss": 0.7381744384765625,
+      "mean_token_accuracy": 0.7967362694442273,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7717011336530193,
+      "eval_loss": 0.7436173558235168,
+      "eval_mean_token_accuracy": 0.7975849482450592,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 89.8411,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 400
+    },
+    {
+      "entropy": 0.7987750010056929,
+      "epoch": 1.012070006035003,
+      "grad_norm": 1.2390918731689453,
+      "learning_rate": 0.000364721224843344,
+      "loss": 0.7317790031433106,
+      "mean_token_accuracy": 0.7986087551364651,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7333170604170038,
+      "eval_loss": 0.7456948757171631,
+      "eval_mean_token_accuracy": 0.7964897577682238,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 89.7288,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 420
+    },
+    {
+      "entropy": 0.7372976846992969,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.1552740335464478,
+      "learning_rate": 0.00036468510102269575,
+      "loss": 0.6734249114990234,
+      "mean_token_accuracy": 0.805834824591875,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.7130355526891987,
+      "eval_loss": 0.7555333375930786,
+      "eval_mean_token_accuracy": 0.7961652781186479,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 89.8713,
+      "eval_samples_per_second": 15.8,
+      "eval_steps_per_second": 1.981,
+      "step": 440
+    },
+    {
+      "entropy": 0.7465668715536594,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 1.1991469860076904,
+      "learning_rate": 0.0003645973816745066,
+      "loss": 0.6871879577636719,
+      "mean_token_accuracy": 0.8042450234293937,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7772979542110743,
+      "eval_loss": 0.7561826705932617,
+      "eval_mean_token_accuracy": 0.7953876427720102,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 89.8395,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 460
+    },
+    {
+      "entropy": 0.7477855801582336,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 1.325486660003662,
+      "learning_rate": 0.00036445809162231435,
+      "loss": 0.6836830139160156,
+      "mean_token_accuracy": 0.8021619468927383,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.7098110676481483,
+      "eval_loss": 0.758007287979126,
+      "eval_mean_token_accuracy": 0.7909953189030122,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 89.6101,
+      "eval_samples_per_second": 15.846,
+      "eval_steps_per_second": 1.986,
+      "step": 480
+    },
+    {
+      "entropy": 0.7632691070437432,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 1.0122262239456177,
+      "learning_rate": 0.0003642672702835562,
+      "loss": 0.7071213245391845,
+      "mean_token_accuracy": 0.8000680930912495,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7645382720432924,
+      "eval_loss": 0.7492606043815613,
+      "eval_mean_token_accuracy": 0.7976858676149604,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 89.6092,
+      "eval_samples_per_second": 15.847,
+      "eval_steps_per_second": 1.986,
+      "step": 500
+    },
+    {
+      "entropy": 0.7623726457357407,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.367774486541748,
+      "learning_rate": 0.00036402497165841384,
+      "loss": 0.6965175151824952,
+      "mean_token_accuracy": 0.8016501650214195,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7868083715438843,
+      "eval_loss": 0.7481449842453003,
+      "eval_mean_token_accuracy": 0.799330582779445,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 89.8241,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 520
+    },
+    {
+      "entropy": 0.7535207405686378,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 1.213614821434021,
+      "learning_rate": 0.00036373126431453207,
+      "loss": 0.7078310489654541,
+      "mean_token_accuracy": 0.7998907208442688,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7829091950748743,
+      "eval_loss": 0.7416072487831116,
+      "eval_mean_token_accuracy": 0.7969146250339036,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 89.4607,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 540
+    },
+    {
+      "entropy": 0.7527363449335098,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 1.0929982662200928,
+      "learning_rate": 0.00036338623136761495,
+      "loss": 0.7058369636535644,
+      "mean_token_accuracy": 0.7997087433934211,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.728419938114252,
+      "eval_loss": 0.744873046875,
+      "eval_mean_token_accuracy": 0.7966634296299366,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 89.6597,
+      "eval_samples_per_second": 15.838,
+      "eval_steps_per_second": 1.985,
+      "step": 560
+    },
+    {
+      "entropy": 0.7406099259853363,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.409364104270935,
+      "learning_rate": 0.00036298997045790515,
+      "loss": 0.6902852058410645,
+      "mean_token_accuracy": 0.805386807769537,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.6991632758231645,
+      "eval_loss": 0.7480319738388062,
+      "eval_mean_token_accuracy": 0.7970068006033308,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 89.8329,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 580
+    },
+    {
+      "entropy": 0.7459449704736472,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.7053111791610718,
+      "learning_rate": 0.00036254259372255277,
+      "loss": 0.7044640064239502,
+      "mean_token_accuracy": 0.8005018755793571,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7454566660891758,
+      "eval_loss": 0.7551484704017639,
+      "eval_mean_token_accuracy": 0.7897093731365846,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 89.8313,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 600
+    },
+    {
+      "entropy": 0.7621391348540782,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 1.174443244934082,
+      "learning_rate": 0.000362044227763882,
+      "loss": 0.7111573696136475,
+      "mean_token_accuracy": 0.798752411454916,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7664387949397055,
+      "eval_loss": 0.7365804314613342,
+      "eval_mean_token_accuracy": 0.7963101200843126,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 89.9623,
+      "eval_samples_per_second": 15.784,
+      "eval_steps_per_second": 1.979,
+      "step": 620
+    },
+    {
+      "entropy": 0.7424944408237935,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.2898184061050415,
+      "learning_rate": 0.000361495013613564,
+      "loss": 0.6913110733032226,
+      "mean_token_accuracy": 0.8009015507996082,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7255479529332579,
+      "eval_loss": 0.7345991730690002,
+      "eval_mean_token_accuracy": 0.8013592945056015,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 89.817,
+      "eval_samples_per_second": 15.81,
+      "eval_steps_per_second": 1.982,
+      "step": 640
+    },
+    {
+      "entropy": 0.7303367637097835,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.1310476064682007,
+      "learning_rate": 0.00036089510669270666,
+      "loss": 0.6875617027282714,
+      "mean_token_accuracy": 0.8035397171974182,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7101308248016271,
+      "eval_loss": 0.7394917011260986,
+      "eval_mean_token_accuracy": 0.7990792942850777,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 89.7431,
+      "eval_samples_per_second": 15.823,
+      "eval_steps_per_second": 1.983,
+      "step": 660
+    },
+    {
+      "entropy": 0.7387678384780884,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.9645389914512634,
+      "learning_rate": 0.0003602446767678725,
+      "loss": 0.6911204338073731,
+      "mean_token_accuracy": 0.8056333385407924,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7281073737010527,
+      "eval_loss": 0.7376708984375,
+      "eval_mean_token_accuracy": 0.7997356415464637,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 90.0338,
+      "eval_samples_per_second": 15.772,
+      "eval_steps_per_second": 1.977,
+      "step": 680
+    },
+    {
+      "entropy": 0.7449554048478604,
+      "epoch": 1.687990343995172,
+      "grad_norm": 1.9069629907608032,
+      "learning_rate": 0.0003595439079030364,
+      "loss": 0.7042286396026611,
+      "mean_token_accuracy": 0.8018639378249646,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7022633817088738,
+      "eval_loss": 0.7320939898490906,
+      "eval_mean_token_accuracy": 0.8006051541044471,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 89.8948,
+      "eval_samples_per_second": 15.796,
+      "eval_steps_per_second": 1.98,
+      "step": 700
+    },
+    {
+      "entropy": 0.723249051719904,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.8333877325057983,
+      "learning_rate": 0.00035879299840749777,
+      "loss": 0.7043714046478271,
+      "mean_token_accuracy": 0.8028345607221127,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.672436946898364,
+      "eval_loss": 0.713233470916748,
+      "eval_mean_token_accuracy": 0.8047760577684038,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 89.6394,
+      "eval_samples_per_second": 15.841,
+      "eval_steps_per_second": 1.986,
+      "step": 720
+    },
+    {
+      "entropy": 0.730014206469059,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.4580037593841553,
+      "learning_rate": 0.00035799216077976137,
+      "loss": 0.6985176563262939,
+      "mean_token_accuracy": 0.8015547141432762,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.6908873795123582,
+      "eval_loss": 0.7198202013969421,
+      "eval_mean_token_accuracy": 0.8024131682481659,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 89.3771,
+      "eval_samples_per_second": 15.888,
+      "eval_steps_per_second": 1.992,
+      "step": 740
+    },
+    {
+      "entropy": 0.7219605796039105,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.51682710647583,
+      "learning_rate": 0.000357141621647403,
+      "loss": 0.6844424724578857,
+      "mean_token_accuracy": 0.8032813094556331,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7113401109582922,
+      "eval_loss": 0.7195309400558472,
+      "eval_mean_token_accuracy": 0.8019187035185568,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 89.4658,
+      "eval_samples_per_second": 15.872,
+      "eval_steps_per_second": 1.99,
+      "step": 760
+    },
+    {
+      "entropy": 0.729826743900776,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 1.0838000774383545,
+      "learning_rate": 0.0003562416217029361,
+      "loss": 0.685858678817749,
+      "mean_token_accuracy": 0.8050567395985126,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.649370267484965,
+      "eval_loss": 0.7177029252052307,
+      "eval_mean_token_accuracy": 0.8050174043419656,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 89.2287,
+      "eval_samples_per_second": 15.914,
+      "eval_steps_per_second": 1.995,
+      "step": 780
+    },
+    {
+      "entropy": 0.6799364153295755,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 1.1364924907684326,
+      "learning_rate": 0.00035529241563569903,
+      "loss": 0.6714428901672364,
+      "mean_token_accuracy": 0.8087341658771038,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6603799047095052,
+      "eval_loss": 0.7201809287071228,
+      "eval_mean_token_accuracy": 0.802863868769635,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 89.2718,
+      "eval_samples_per_second": 15.906,
+      "eval_steps_per_second": 1.994,
+      "step": 800
+    },
+    {
+      "entropy": 0.7078610375523567,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 1.2191942930221558,
+      "learning_rate": 0.0003542942720597808,
+      "loss": 0.6996590614318847,
+      "mean_token_accuracy": 0.8049512051045895,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.7240545264120852,
+      "eval_loss": 0.7301309108734131,
+      "eval_mean_token_accuracy": 0.8015973166133581,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 89.3438,
+      "eval_samples_per_second": 15.894,
+      "eval_steps_per_second": 1.992,
+      "step": 820
+    },
+    {
+      "entropy": 0.6557391556826505,
+      "epoch": 2.024140012070006,
+      "grad_norm": 1.36203134059906,
+      "learning_rate": 0.0003532474734380064,
+      "loss": 0.6327113628387451,
+      "mean_token_accuracy": 0.8142829847026181,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6573808648613062,
+      "eval_loss": 0.7334365844726562,
+      "eval_mean_token_accuracy": 0.8043633893634496,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 89.3359,
+      "eval_samples_per_second": 15.895,
+      "eval_steps_per_second": 1.992,
+      "step": 840
+    },
+    {
+      "entropy": 0.6409128502011299,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 1.5804955959320068,
+      "learning_rate": 0.0003521523160020035,
+      "loss": 0.6023559093475341,
+      "mean_token_accuracy": 0.8222251623868942,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6983234788594621,
+      "eval_loss": 0.7232135534286499,
+      "eval_mean_token_accuracy": 0.8010654268639811,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 89.2615,
+      "eval_samples_per_second": 15.908,
+      "eval_steps_per_second": 1.994,
+      "step": 860
+    },
+    {
+      "entropy": 0.6151149723678827,
+      "epoch": 2.12070006035003,
+      "grad_norm": 1.1975436210632324,
+      "learning_rate": 0.00035100910966837193,
+      "loss": 0.5803564548492431,
+      "mean_token_accuracy": 0.8256325736641884,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6636555573243773,
+      "eval_loss": 0.7205662727355957,
+      "eval_mean_token_accuracy": 0.8036768369460374,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 89.2957,
+      "eval_samples_per_second": 15.902,
+      "eval_steps_per_second": 1.993,
+      "step": 880
+    },
+    {
+      "entropy": 0.6201727617532015,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9907131195068359,
+      "learning_rate": 0.0003498181779509813,
+      "loss": 0.6101872444152832,
+      "mean_token_accuracy": 0.8196512959897518,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.6473788086617931,
+      "eval_loss": 0.7062397003173828,
+      "eval_mean_token_accuracy": 0.8078362184963869,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 88.6198,
+      "eval_samples_per_second": 16.024,
+      "eval_steps_per_second": 2.009,
+      "step": 900
+    },
+    {
+      "entropy": 0.6097332935780286,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.9055633544921875,
+      "learning_rate": 0.00034857985786942026,
+      "loss": 0.5956094264984131,
+      "mean_token_accuracy": 0.8240575887262821,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.612629868341296,
+      "eval_loss": 0.7106770277023315,
+      "eval_mean_token_accuracy": 0.805324277181304,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 89.4383,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 920
+    },
+    {
+      "entropy": 0.6182701248675585,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.9513736367225647,
+      "learning_rate": 0.00034729449985362404,
+      "loss": 0.6019269466400147,
+      "mean_token_accuracy": 0.8187106013298034,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6437647456533453,
+      "eval_loss": 0.7173502445220947,
+      "eval_mean_token_accuracy": 0.8056439573175451,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.9568,
+      "eval_samples_per_second": 15.963,
+      "eval_steps_per_second": 2.001,
+      "step": 940
+    },
+    {
+      "entropy": 0.6508570514619351,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.9606207013130188,
+      "learning_rate": 0.0003459624676447067,
+      "loss": 0.6296409606933594,
+      "mean_token_accuracy": 0.8167732007801533,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6288247135248077,
+      "eval_loss": 0.7160675525665283,
+      "eval_mean_token_accuracy": 0.8066010689467527,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 89.0044,
+      "eval_samples_per_second": 15.954,
+      "eval_steps_per_second": 2.0,
+      "step": 960
+    },
+    {
+      "entropy": 0.6380081418901682,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.9677181839942932,
+      "learning_rate": 0.000344584138192027,
+      "loss": 0.6210546493530273,
+      "mean_token_accuracy": 0.8186135642230511,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6407805031604981,
+      "eval_loss": 0.7055043578147888,
+      "eval_mean_token_accuracy": 0.8070103157772107,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 89.5606,
+      "eval_samples_per_second": 15.855,
+      "eval_steps_per_second": 1.987,
+      "step": 980
+    },
+    {
+      "entropy": 0.6310251638293266,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 2.1046407222747803,
+      "learning_rate": 0.000343159901546516,
+      "loss": 0.6165830135345459,
+      "mean_token_accuracy": 0.8207855716347694,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6461507470420237,
+      "eval_loss": 0.7292136549949646,
+      "eval_mean_token_accuracy": 0.8030418261383356,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 89.315,
+      "eval_samples_per_second": 15.899,
+      "eval_steps_per_second": 1.993,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6256617344915867,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 1.1768349409103394,
+      "learning_rate": 0.0003416901607502972,
+      "loss": 0.6142709255218506,
+      "mean_token_accuracy": 0.8220669947564602,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6150523993406403,
+      "eval_loss": 0.7019563317298889,
+      "eval_mean_token_accuracy": 0.8094323099998946,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 89.419,
+      "eval_samples_per_second": 15.88,
+      "eval_steps_per_second": 1.991,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6126735735684633,
+      "epoch": 2.506940253470127,
+      "grad_norm": 1.8065496683120728,
+      "learning_rate": 0.00034017533172263055,
+      "loss": 0.6056458473205566,
+      "mean_token_accuracy": 0.8203762136399746,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6544345456562685,
+      "eval_loss": 0.7086517214775085,
+      "eval_mean_token_accuracy": 0.806709805901131,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 88.981,
+      "eval_samples_per_second": 15.958,
+      "eval_steps_per_second": 2.0,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6311754353344441,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 1.0576369762420654,
+      "learning_rate": 0.00033861584314221236,
+      "loss": 0.6176392555236816,
+      "mean_token_accuracy": 0.8177410490810871,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6273799200406235,
+      "eval_loss": 0.708163321018219,
+      "eval_mean_token_accuracy": 0.8032017099053672,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 89.7668,
+      "eval_samples_per_second": 15.819,
+      "eval_steps_per_second": 1.983,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6200249589979648,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.997065007686615,
+      "learning_rate": 0.0003370121363258637,
+      "loss": 0.6203888893127442,
+      "mean_token_accuracy": 0.8196275025606156,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6457576811983344,
+      "eval_loss": 0.6972290277481079,
+      "eval_mean_token_accuracy": 0.8095980849158898,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 89.2027,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 1080
+    },
+    {
+      "entropy": 0.626810473203659,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.865475058555603,
+      "learning_rate": 0.0003353646651036438,
+      "loss": 0.6102585792541504,
+      "mean_token_accuracy": 0.819708751142025,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.6412022431914726,
+      "eval_loss": 0.7036139965057373,
+      "eval_mean_token_accuracy": 0.8070317135098275,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 89.4733,
+      "eval_samples_per_second": 15.871,
+      "eval_steps_per_second": 1.989,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6581619590520859,
+      "epoch": 2.700060350030175,
+      "grad_norm": 1.00307035446167,
+      "learning_rate": 0.00033367389569042064,
+      "loss": 0.6397720336914062,
+      "mean_token_accuracy": 0.81413309648633,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6288736402318719,
+      "eval_loss": 0.6966825723648071,
+      "eval_mean_token_accuracy": 0.8094039669867312,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 88.7341,
+      "eval_samples_per_second": 16.003,
+      "eval_steps_per_second": 2.006,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6399412982165813,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.8589329719543457,
+      "learning_rate": 0.0003319403065539384,
+      "loss": 0.6347338199615479,
+      "mean_token_accuracy": 0.8139534957706929,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.683491913455256,
+      "eval_loss": 0.6984149217605591,
+      "eval_mean_token_accuracy": 0.8083989469522841,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 89.3929,
+      "eval_samples_per_second": 15.885,
+      "eval_steps_per_second": 1.991,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6264835461974144,
+      "epoch": 2.796620398310199,
+      "grad_norm": 0.7807853817939758,
+      "learning_rate": 0.0003301643882794163,
+      "loss": 0.6162629127502441,
+      "mean_token_accuracy": 0.8216292977333068,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6578792159476977,
+      "eval_loss": 0.6921441555023193,
+      "eval_mean_token_accuracy": 0.8094299177775223,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 89.2356,
+      "eval_samples_per_second": 15.913,
+      "eval_steps_per_second": 1.995,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6269129011780024,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 1.1675593852996826,
+      "learning_rate": 0.0003283466434307189,
+      "loss": 0.6235532760620117,
+      "mean_token_accuracy": 0.8188897788524627,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6533248447970058,
+      "eval_loss": 0.696849524974823,
+      "eval_mean_token_accuracy": 0.8102703713968898,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2602,
+      "eval_samples_per_second": 16.089,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.6259873781353236,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.9596486687660217,
+      "learning_rate": 0.00032648758640813655,
+      "loss": 0.6118728160858155,
+      "mean_token_accuracy": 0.8171859130263328,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6212078754821521,
+      "eval_loss": 0.6942870020866394,
+      "eval_mean_token_accuracy": 0.810484343365337,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 89.4375,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 1200
+    },
+    {
+      "entropy": 0.6142537571489811,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 0.9625058770179749,
+      "learning_rate": 0.00032458774330281615,
+      "loss": 0.6060261249542236,
+      "mean_token_accuracy": 0.8223113484680653,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6553665667437436,
+      "eval_loss": 0.6835269331932068,
+      "eval_mean_token_accuracy": 0.8134723968720168,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 89.5092,
+      "eval_samples_per_second": 15.864,
+      "eval_steps_per_second": 1.989,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6301972426474094,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.9657333493232727,
+      "learning_rate": 0.0003226476517478835,
+      "loss": 0.6277759075164795,
+      "mean_token_accuracy": 0.8164266042411328,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6594253118788258,
+      "eval_loss": 0.6820935606956482,
+      "eval_mean_token_accuracy": 0.8132368198941263,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 89.6211,
+      "eval_samples_per_second": 15.844,
+      "eval_steps_per_second": 1.986,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5346100655469027,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8541185259819031,
+      "learning_rate": 0.0003206678607662996,
+      "loss": 0.5261692047119141,
+      "mean_token_accuracy": 0.8382289425119177,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5626645662476507,
+      "eval_loss": 0.7028542160987854,
+      "eval_mean_token_accuracy": 0.8133395572056931,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 89.1267,
+      "eval_samples_per_second": 15.932,
+      "eval_steps_per_second": 1.997,
+      "step": 1260
+    },
+    {
+      "entropy": 0.5001101832836866,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.8208815455436707,
+      "learning_rate": 0.0003186489306154935,
+      "loss": 0.4850132942199707,
+      "mean_token_accuracy": 0.8468983843922615,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5889462046743779,
+      "eval_loss": 0.691015899181366,
+      "eval_mean_token_accuracy": 0.8135226466012805,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.9677,
+      "eval_samples_per_second": 15.961,
+      "eval_steps_per_second": 2.001,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5242766339331866,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 0.7893072962760925,
+      "learning_rate": 0.0003165914326288163,
+      "loss": 0.5061577320098877,
+      "mean_token_accuracy": 0.8425608821213245,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5543691962957382,
+      "eval_loss": 0.6913734078407288,
+      "eval_mean_token_accuracy": 0.8154889680026622,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 89.6963,
+      "eval_samples_per_second": 15.831,
+      "eval_steps_per_second": 1.984,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5327721010893584,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.8992080092430115,
+      "learning_rate": 0.0003144959490538604,
+      "loss": 0.5068036556243897,
+      "mean_token_accuracy": 0.8420542575418949,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5734540803378887,
+      "eval_loss": 0.6948716044425964,
+      "eval_mean_token_accuracy": 0.8137217944257715,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 89.4597,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5247377116233111,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.9291845560073853,
+      "learning_rate": 0.0003123630728876902,
+      "loss": 0.5023272037506104,
+      "mean_token_accuracy": 0.8435182586312294,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5492362935891312,
+      "eval_loss": 0.6980633735656738,
+      "eval_mean_token_accuracy": 0.8155082919624415,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 89.2124,
+      "eval_samples_per_second": 15.917,
+      "eval_steps_per_second": 1.995,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5380499072372913,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 1.0477524995803833,
+      "learning_rate": 0.00031019340770903136,
+      "loss": 0.5264591693878173,
+      "mean_token_accuracy": 0.8391959741711617,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5896503307511297,
+      "eval_loss": 0.6997144222259521,
+      "eval_mean_token_accuracy": 0.8124373062942805,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 89.1147,
+      "eval_samples_per_second": 15.935,
+      "eval_steps_per_second": 1.997,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5460653610527515,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 1.2473069429397583,
+      "learning_rate": 0.00030798756750746476,
+      "loss": 0.5319677829742432,
+      "mean_token_accuracy": 0.8364320226013661,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5659052182114526,
+      "eval_loss": 0.6924049258232117,
+      "eval_mean_token_accuracy": 0.8153053585732921,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 89.6468,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5367407951503992,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 1.2679712772369385,
+      "learning_rate": 0.00030574617650967485,
+      "loss": 0.5190927028656006,
+      "mean_token_accuracy": 0.8399469174444676,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5581759144081159,
+      "eval_loss": 0.6933317184448242,
+      "eval_mean_token_accuracy": 0.8139757862251796,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 89.0749,
+      "eval_samples_per_second": 15.942,
+      "eval_steps_per_second": 1.998,
+      "step": 1400
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.4040125723949466e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1420/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1420/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1420/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1420/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1420/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..7ce4b072fcbe5b636867e13937c56fedee3b68b4
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1420/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.02728394274790723,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1420/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1420/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1420/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1420/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1420/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1420/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1420/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1420/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..2e3a96db25b559fbbfd20878c9621d4d63eba4b9
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1420/trainer_state.json
@@ -0,0 +1,1525 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 3.4224502112251054,
+  "eval_steps": 20,
+  "global_step": 1420,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.024795526266098,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.945375919342041,
+      "learning_rate": 1.6698127428345936e-05,
+      "loss": 1.7825420379638672,
+      "mean_token_accuracy": 0.6315708436071873,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.29921497588747,
+      "eval_loss": 1.169131875038147,
+      "eval_mean_token_accuracy": 0.7185706342204233,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 91.7898,
+      "eval_samples_per_second": 15.47,
+      "eval_steps_per_second": 1.939,
+      "step": 20
+    },
+    {
+      "entropy": 1.0346344470977784,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.7916979789733887,
+      "learning_rate": 3.427510366871008e-05,
+      "loss": 0.9480395317077637,
+      "mean_token_accuracy": 0.7543429024517536,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 0.9673695430326997,
+      "eval_loss": 0.8588430285453796,
+      "eval_mean_token_accuracy": 0.7737233002534073,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 89.8219,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 40
+    },
+    {
+      "entropy": 0.8977566614747048,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7972187995910645,
+      "learning_rate": 5.185207990907422e-05,
+      "loss": 0.8029604911804199,
+      "mean_token_accuracy": 0.7818016350269318,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8436836982041263,
+      "eval_loss": 0.7893033623695374,
+      "eval_mean_token_accuracy": 0.7858914984076211,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 89.9349,
+      "eval_samples_per_second": 15.789,
+      "eval_steps_per_second": 1.979,
+      "step": 60
+    },
+    {
+      "entropy": 0.8650321319699288,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.2128729820251465,
+      "learning_rate": 6.942905614943836e-05,
+      "loss": 0.77417893409729,
+      "mean_token_accuracy": 0.7875479467213153,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8209857180547179,
+      "eval_loss": 0.7572280168533325,
+      "eval_mean_token_accuracy": 0.7925970383574453,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 89.9459,
+      "eval_samples_per_second": 15.787,
+      "eval_steps_per_second": 1.979,
+      "step": 80
+    },
+    {
+      "entropy": 0.8096660938113928,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.1254287958145142,
+      "learning_rate": 8.700603238980251e-05,
+      "loss": 0.7170331001281738,
+      "mean_token_accuracy": 0.7994555257260799,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.792780208788561,
+      "eval_loss": 0.7442984580993652,
+      "eval_mean_token_accuracy": 0.7954704573984896,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 90.294,
+      "eval_samples_per_second": 15.726,
+      "eval_steps_per_second": 1.971,
+      "step": 100
+    },
+    {
+      "entropy": 0.7979829199612141,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2140507698059082,
+      "learning_rate": 0.00010458300863016666,
+      "loss": 0.7141861915588379,
+      "mean_token_accuracy": 0.8000014744699001,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.7938692486018277,
+      "eval_loss": 0.7279797196388245,
+      "eval_mean_token_accuracy": 0.7989798734027348,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 89.7189,
+      "eval_samples_per_second": 15.827,
+      "eval_steps_per_second": 1.984,
+      "step": 120
+    },
+    {
+      "entropy": 0.7916347607970238,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.0506458282470703,
+      "learning_rate": 0.0001221599848705308,
+      "loss": 0.7117055416107178,
+      "mean_token_accuracy": 0.7970046654343605,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.8052781063519167,
+      "eval_loss": 0.722576916217804,
+      "eval_mean_token_accuracy": 0.7982530523551984,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 90.0376,
+      "eval_samples_per_second": 15.771,
+      "eval_steps_per_second": 1.977,
+      "step": 140
+    },
+    {
+      "entropy": 0.7880666613578796,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8409207463264465,
+      "learning_rate": 0.00013973696111089492,
+      "loss": 0.7052523136138916,
+      "mean_token_accuracy": 0.8000123649835587,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7787443134891853,
+      "eval_loss": 0.7137772440910339,
+      "eval_mean_token_accuracy": 0.8014209287220173,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 90.2562,
+      "eval_samples_per_second": 15.733,
+      "eval_steps_per_second": 1.972,
+      "step": 160
+    },
+    {
+      "entropy": 0.7696007996797561,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.0182477235794067,
+      "learning_rate": 0.00015731393735125907,
+      "loss": 0.7008797645568847,
+      "mean_token_accuracy": 0.8007259473204613,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.752259682068664,
+      "eval_loss": 0.7171286344528198,
+      "eval_mean_token_accuracy": 0.80052122574174,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 90.0797,
+      "eval_samples_per_second": 15.764,
+      "eval_steps_per_second": 1.976,
+      "step": 180
+    },
+    {
+      "entropy": 0.7828126326203346,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9553632736206055,
+      "learning_rate": 0.0001748909135916232,
+      "loss": 0.7046853542327881,
+      "mean_token_accuracy": 0.7988562889397144,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7389638169427936,
+      "eval_loss": 0.7187935709953308,
+      "eval_mean_token_accuracy": 0.7995899999409579,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 89.9223,
+      "eval_samples_per_second": 15.791,
+      "eval_steps_per_second": 1.979,
+      "step": 200
+    },
+    {
+      "entropy": 0.7909286297857762,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 1.134459376335144,
+      "learning_rate": 0.00019246788983198735,
+      "loss": 0.6989312171936035,
+      "mean_token_accuracy": 0.7991565138101577,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7496965199374082,
+      "eval_loss": 0.7106321454048157,
+      "eval_mean_token_accuracy": 0.8046183442131857,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 89.6899,
+      "eval_samples_per_second": 15.832,
+      "eval_steps_per_second": 1.985,
+      "step": 220
+    },
+    {
+      "entropy": 0.7651963323354721,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.9557585716247559,
+      "learning_rate": 0.0002100448660723515,
+      "loss": 0.6984559059143066,
+      "mean_token_accuracy": 0.8019823372364044,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7615563079212488,
+      "eval_loss": 0.7098539471626282,
+      "eval_mean_token_accuracy": 0.8047233508544022,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 90.3752,
+      "eval_samples_per_second": 15.712,
+      "eval_steps_per_second": 1.97,
+      "step": 240
+    },
+    {
+      "entropy": 0.787507726252079,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.9279311299324036,
+      "learning_rate": 0.00022762184231271568,
+      "loss": 0.708138370513916,
+      "mean_token_accuracy": 0.8013308703899383,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7856640896100676,
+      "eval_loss": 0.7191570401191711,
+      "eval_mean_token_accuracy": 0.8024677249153008,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 93.7336,
+      "eval_samples_per_second": 15.149,
+      "eval_steps_per_second": 1.899,
+      "step": 260
+    },
+    {
+      "entropy": 0.768489234894514,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.7992776036262512,
+      "learning_rate": 0.0002451988185530798,
+      "loss": 0.7035849571228028,
+      "mean_token_accuracy": 0.8031105332076549,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7682498740346244,
+      "eval_loss": 0.7149022221565247,
+      "eval_mean_token_accuracy": 0.7973497995499814,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 89.7309,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 280
+    },
+    {
+      "entropy": 0.7720955617725849,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.1931514739990234,
+      "learning_rate": 0.0002627757947934439,
+      "loss": 0.6965402603149414,
+      "mean_token_accuracy": 0.8040348663926125,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7214708304807042,
+      "eval_loss": 0.7350823879241943,
+      "eval_mean_token_accuracy": 0.7942911925610532,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 89.6451,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 300
+    },
+    {
+      "entropy": 0.7930095426738262,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8994467258453369,
+      "learning_rate": 0.0002803527710338081,
+      "loss": 0.7331492900848389,
+      "mean_token_accuracy": 0.796185302734375,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.8142535488926963,
+      "eval_loss": 0.7237834334373474,
+      "eval_mean_token_accuracy": 0.799495273761535,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 90.6759,
+      "eval_samples_per_second": 15.66,
+      "eval_steps_per_second": 1.963,
+      "step": 320
+    },
+    {
+      "entropy": 0.7825102441012859,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 1.8224239349365234,
+      "learning_rate": 0.00029792974727417223,
+      "loss": 0.7331175804138184,
+      "mean_token_accuracy": 0.7977667152881622,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7327135443017724,
+      "eval_loss": 0.7388784289360046,
+      "eval_mean_token_accuracy": 0.8007041252730938,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 89.7692,
+      "eval_samples_per_second": 15.818,
+      "eval_steps_per_second": 1.983,
+      "step": 340
+    },
+    {
+      "entropy": 0.819331557303667,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 1.0407779216766357,
+      "learning_rate": 0.0003155067235145364,
+      "loss": 0.7456695556640625,
+      "mean_token_accuracy": 0.7936980701982975,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.8065585420372781,
+      "eval_loss": 0.735519528388977,
+      "eval_mean_token_accuracy": 0.7974206296245704,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 89.4336,
+      "eval_samples_per_second": 15.878,
+      "eval_steps_per_second": 1.99,
+      "step": 360
+    },
+    {
+      "entropy": 0.8114653021097183,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.1209490299224854,
+      "learning_rate": 0.0003330836997549005,
+      "loss": 0.7340899467468261,
+      "mean_token_accuracy": 0.7914272703230381,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.8025527285056168,
+      "eval_loss": 0.7295576930046082,
+      "eval_mean_token_accuracy": 0.7990576628218876,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 89.9579,
+      "eval_samples_per_second": 15.785,
+      "eval_steps_per_second": 1.979,
+      "step": 380
+    },
+    {
+      "entropy": 0.8042497783899307,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 1.2437798976898193,
+      "learning_rate": 0.0003506606759952647,
+      "loss": 0.7381744384765625,
+      "mean_token_accuracy": 0.7967362694442273,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7717011336530193,
+      "eval_loss": 0.7436173558235168,
+      "eval_mean_token_accuracy": 0.7975849482450592,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 89.8411,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 400
+    },
+    {
+      "entropy": 0.7987750010056929,
+      "epoch": 1.012070006035003,
+      "grad_norm": 1.2390918731689453,
+      "learning_rate": 0.000364721224843344,
+      "loss": 0.7317790031433106,
+      "mean_token_accuracy": 0.7986087551364651,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7333170604170038,
+      "eval_loss": 0.7456948757171631,
+      "eval_mean_token_accuracy": 0.7964897577682238,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 89.7288,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 420
+    },
+    {
+      "entropy": 0.7372976846992969,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.1552740335464478,
+      "learning_rate": 0.00036468510102269575,
+      "loss": 0.6734249114990234,
+      "mean_token_accuracy": 0.805834824591875,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.7130355526891987,
+      "eval_loss": 0.7555333375930786,
+      "eval_mean_token_accuracy": 0.7961652781186479,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 89.8713,
+      "eval_samples_per_second": 15.8,
+      "eval_steps_per_second": 1.981,
+      "step": 440
+    },
+    {
+      "entropy": 0.7465668715536594,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 1.1991469860076904,
+      "learning_rate": 0.0003645973816745066,
+      "loss": 0.6871879577636719,
+      "mean_token_accuracy": 0.8042450234293937,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7772979542110743,
+      "eval_loss": 0.7561826705932617,
+      "eval_mean_token_accuracy": 0.7953876427720102,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 89.8395,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 460
+    },
+    {
+      "entropy": 0.7477855801582336,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 1.325486660003662,
+      "learning_rate": 0.00036445809162231435,
+      "loss": 0.6836830139160156,
+      "mean_token_accuracy": 0.8021619468927383,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.7098110676481483,
+      "eval_loss": 0.758007287979126,
+      "eval_mean_token_accuracy": 0.7909953189030122,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 89.6101,
+      "eval_samples_per_second": 15.846,
+      "eval_steps_per_second": 1.986,
+      "step": 480
+    },
+    {
+      "entropy": 0.7632691070437432,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 1.0122262239456177,
+      "learning_rate": 0.0003642672702835562,
+      "loss": 0.7071213245391845,
+      "mean_token_accuracy": 0.8000680930912495,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7645382720432924,
+      "eval_loss": 0.7492606043815613,
+      "eval_mean_token_accuracy": 0.7976858676149604,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 89.6092,
+      "eval_samples_per_second": 15.847,
+      "eval_steps_per_second": 1.986,
+      "step": 500
+    },
+    {
+      "entropy": 0.7623726457357407,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.367774486541748,
+      "learning_rate": 0.00036402497165841384,
+      "loss": 0.6965175151824952,
+      "mean_token_accuracy": 0.8016501650214195,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7868083715438843,
+      "eval_loss": 0.7481449842453003,
+      "eval_mean_token_accuracy": 0.799330582779445,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 89.8241,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 520
+    },
+    {
+      "entropy": 0.7535207405686378,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 1.213614821434021,
+      "learning_rate": 0.00036373126431453207,
+      "loss": 0.7078310489654541,
+      "mean_token_accuracy": 0.7998907208442688,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7829091950748743,
+      "eval_loss": 0.7416072487831116,
+      "eval_mean_token_accuracy": 0.7969146250339036,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 89.4607,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 540
+    },
+    {
+      "entropy": 0.7527363449335098,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 1.0929982662200928,
+      "learning_rate": 0.00036338623136761495,
+      "loss": 0.7058369636535644,
+      "mean_token_accuracy": 0.7997087433934211,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.728419938114252,
+      "eval_loss": 0.744873046875,
+      "eval_mean_token_accuracy": 0.7966634296299366,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 89.6597,
+      "eval_samples_per_second": 15.838,
+      "eval_steps_per_second": 1.985,
+      "step": 560
+    },
+    {
+      "entropy": 0.7406099259853363,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.409364104270935,
+      "learning_rate": 0.00036298997045790515,
+      "loss": 0.6902852058410645,
+      "mean_token_accuracy": 0.805386807769537,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.6991632758231645,
+      "eval_loss": 0.7480319738388062,
+      "eval_mean_token_accuracy": 0.7970068006033308,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 89.8329,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 580
+    },
+    {
+      "entropy": 0.7459449704736472,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.7053111791610718,
+      "learning_rate": 0.00036254259372255277,
+      "loss": 0.7044640064239502,
+      "mean_token_accuracy": 0.8005018755793571,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7454566660891758,
+      "eval_loss": 0.7551484704017639,
+      "eval_mean_token_accuracy": 0.7897093731365846,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 89.8313,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 600
+    },
+    {
+      "entropy": 0.7621391348540782,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 1.174443244934082,
+      "learning_rate": 0.000362044227763882,
+      "loss": 0.7111573696136475,
+      "mean_token_accuracy": 0.798752411454916,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7664387949397055,
+      "eval_loss": 0.7365804314613342,
+      "eval_mean_token_accuracy": 0.7963101200843126,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 89.9623,
+      "eval_samples_per_second": 15.784,
+      "eval_steps_per_second": 1.979,
+      "step": 620
+    },
+    {
+      "entropy": 0.7424944408237935,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.2898184061050415,
+      "learning_rate": 0.000361495013613564,
+      "loss": 0.6913110733032226,
+      "mean_token_accuracy": 0.8009015507996082,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7255479529332579,
+      "eval_loss": 0.7345991730690002,
+      "eval_mean_token_accuracy": 0.8013592945056015,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 89.817,
+      "eval_samples_per_second": 15.81,
+      "eval_steps_per_second": 1.982,
+      "step": 640
+    },
+    {
+      "entropy": 0.7303367637097835,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.1310476064682007,
+      "learning_rate": 0.00036089510669270666,
+      "loss": 0.6875617027282714,
+      "mean_token_accuracy": 0.8035397171974182,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7101308248016271,
+      "eval_loss": 0.7394917011260986,
+      "eval_mean_token_accuracy": 0.7990792942850777,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 89.7431,
+      "eval_samples_per_second": 15.823,
+      "eval_steps_per_second": 1.983,
+      "step": 660
+    },
+    {
+      "entropy": 0.7387678384780884,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.9645389914512634,
+      "learning_rate": 0.0003602446767678725,
+      "loss": 0.6911204338073731,
+      "mean_token_accuracy": 0.8056333385407924,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7281073737010527,
+      "eval_loss": 0.7376708984375,
+      "eval_mean_token_accuracy": 0.7997356415464637,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 90.0338,
+      "eval_samples_per_second": 15.772,
+      "eval_steps_per_second": 1.977,
+      "step": 680
+    },
+    {
+      "entropy": 0.7449554048478604,
+      "epoch": 1.687990343995172,
+      "grad_norm": 1.9069629907608032,
+      "learning_rate": 0.0003595439079030364,
+      "loss": 0.7042286396026611,
+      "mean_token_accuracy": 0.8018639378249646,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7022633817088738,
+      "eval_loss": 0.7320939898490906,
+      "eval_mean_token_accuracy": 0.8006051541044471,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 89.8948,
+      "eval_samples_per_second": 15.796,
+      "eval_steps_per_second": 1.98,
+      "step": 700
+    },
+    {
+      "entropy": 0.723249051719904,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.8333877325057983,
+      "learning_rate": 0.00035879299840749777,
+      "loss": 0.7043714046478271,
+      "mean_token_accuracy": 0.8028345607221127,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.672436946898364,
+      "eval_loss": 0.713233470916748,
+      "eval_mean_token_accuracy": 0.8047760577684038,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 89.6394,
+      "eval_samples_per_second": 15.841,
+      "eval_steps_per_second": 1.986,
+      "step": 720
+    },
+    {
+      "entropy": 0.730014206469059,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.4580037593841553,
+      "learning_rate": 0.00035799216077976137,
+      "loss": 0.6985176563262939,
+      "mean_token_accuracy": 0.8015547141432762,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.6908873795123582,
+      "eval_loss": 0.7198202013969421,
+      "eval_mean_token_accuracy": 0.8024131682481659,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 89.3771,
+      "eval_samples_per_second": 15.888,
+      "eval_steps_per_second": 1.992,
+      "step": 740
+    },
+    {
+      "entropy": 0.7219605796039105,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.51682710647583,
+      "learning_rate": 0.000357141621647403,
+      "loss": 0.6844424724578857,
+      "mean_token_accuracy": 0.8032813094556331,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7113401109582922,
+      "eval_loss": 0.7195309400558472,
+      "eval_mean_token_accuracy": 0.8019187035185568,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 89.4658,
+      "eval_samples_per_second": 15.872,
+      "eval_steps_per_second": 1.99,
+      "step": 760
+    },
+    {
+      "entropy": 0.729826743900776,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 1.0838000774383545,
+      "learning_rate": 0.0003562416217029361,
+      "loss": 0.685858678817749,
+      "mean_token_accuracy": 0.8050567395985126,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.649370267484965,
+      "eval_loss": 0.7177029252052307,
+      "eval_mean_token_accuracy": 0.8050174043419656,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 89.2287,
+      "eval_samples_per_second": 15.914,
+      "eval_steps_per_second": 1.995,
+      "step": 780
+    },
+    {
+      "entropy": 0.6799364153295755,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 1.1364924907684326,
+      "learning_rate": 0.00035529241563569903,
+      "loss": 0.6714428901672364,
+      "mean_token_accuracy": 0.8087341658771038,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6603799047095052,
+      "eval_loss": 0.7201809287071228,
+      "eval_mean_token_accuracy": 0.802863868769635,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 89.2718,
+      "eval_samples_per_second": 15.906,
+      "eval_steps_per_second": 1.994,
+      "step": 800
+    },
+    {
+      "entropy": 0.7078610375523567,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 1.2191942930221558,
+      "learning_rate": 0.0003542942720597808,
+      "loss": 0.6996590614318847,
+      "mean_token_accuracy": 0.8049512051045895,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.7240545264120852,
+      "eval_loss": 0.7301309108734131,
+      "eval_mean_token_accuracy": 0.8015973166133581,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 89.3438,
+      "eval_samples_per_second": 15.894,
+      "eval_steps_per_second": 1.992,
+      "step": 820
+    },
+    {
+      "entropy": 0.6557391556826505,
+      "epoch": 2.024140012070006,
+      "grad_norm": 1.36203134059906,
+      "learning_rate": 0.0003532474734380064,
+      "loss": 0.6327113628387451,
+      "mean_token_accuracy": 0.8142829847026181,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6573808648613062,
+      "eval_loss": 0.7334365844726562,
+      "eval_mean_token_accuracy": 0.8043633893634496,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 89.3359,
+      "eval_samples_per_second": 15.895,
+      "eval_steps_per_second": 1.992,
+      "step": 840
+    },
+    {
+      "entropy": 0.6409128502011299,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 1.5804955959320068,
+      "learning_rate": 0.0003521523160020035,
+      "loss": 0.6023559093475341,
+      "mean_token_accuracy": 0.8222251623868942,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6983234788594621,
+      "eval_loss": 0.7232135534286499,
+      "eval_mean_token_accuracy": 0.8010654268639811,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 89.2615,
+      "eval_samples_per_second": 15.908,
+      "eval_steps_per_second": 1.994,
+      "step": 860
+    },
+    {
+      "entropy": 0.6151149723678827,
+      "epoch": 2.12070006035003,
+      "grad_norm": 1.1975436210632324,
+      "learning_rate": 0.00035100910966837193,
+      "loss": 0.5803564548492431,
+      "mean_token_accuracy": 0.8256325736641884,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6636555573243773,
+      "eval_loss": 0.7205662727355957,
+      "eval_mean_token_accuracy": 0.8036768369460374,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 89.2957,
+      "eval_samples_per_second": 15.902,
+      "eval_steps_per_second": 1.993,
+      "step": 880
+    },
+    {
+      "entropy": 0.6201727617532015,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9907131195068359,
+      "learning_rate": 0.0003498181779509813,
+      "loss": 0.6101872444152832,
+      "mean_token_accuracy": 0.8196512959897518,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.6473788086617931,
+      "eval_loss": 0.7062397003173828,
+      "eval_mean_token_accuracy": 0.8078362184963869,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 88.6198,
+      "eval_samples_per_second": 16.024,
+      "eval_steps_per_second": 2.009,
+      "step": 900
+    },
+    {
+      "entropy": 0.6097332935780286,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.9055633544921875,
+      "learning_rate": 0.00034857985786942026,
+      "loss": 0.5956094264984131,
+      "mean_token_accuracy": 0.8240575887262821,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.612629868341296,
+      "eval_loss": 0.7106770277023315,
+      "eval_mean_token_accuracy": 0.805324277181304,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 89.4383,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 920
+    },
+    {
+      "entropy": 0.6182701248675585,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.9513736367225647,
+      "learning_rate": 0.00034729449985362404,
+      "loss": 0.6019269466400147,
+      "mean_token_accuracy": 0.8187106013298034,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6437647456533453,
+      "eval_loss": 0.7173502445220947,
+      "eval_mean_token_accuracy": 0.8056439573175451,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.9568,
+      "eval_samples_per_second": 15.963,
+      "eval_steps_per_second": 2.001,
+      "step": 940
+    },
+    {
+      "entropy": 0.6508570514619351,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.9606207013130188,
+      "learning_rate": 0.0003459624676447067,
+      "loss": 0.6296409606933594,
+      "mean_token_accuracy": 0.8167732007801533,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6288247135248077,
+      "eval_loss": 0.7160675525665283,
+      "eval_mean_token_accuracy": 0.8066010689467527,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 89.0044,
+      "eval_samples_per_second": 15.954,
+      "eval_steps_per_second": 2.0,
+      "step": 960
+    },
+    {
+      "entropy": 0.6380081418901682,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.9677181839942932,
+      "learning_rate": 0.000344584138192027,
+      "loss": 0.6210546493530273,
+      "mean_token_accuracy": 0.8186135642230511,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6407805031604981,
+      "eval_loss": 0.7055043578147888,
+      "eval_mean_token_accuracy": 0.8070103157772107,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 89.5606,
+      "eval_samples_per_second": 15.855,
+      "eval_steps_per_second": 1.987,
+      "step": 980
+    },
+    {
+      "entropy": 0.6310251638293266,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 2.1046407222747803,
+      "learning_rate": 0.000343159901546516,
+      "loss": 0.6165830135345459,
+      "mean_token_accuracy": 0.8207855716347694,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6461507470420237,
+      "eval_loss": 0.7292136549949646,
+      "eval_mean_token_accuracy": 0.8030418261383356,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 89.315,
+      "eval_samples_per_second": 15.899,
+      "eval_steps_per_second": 1.993,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6256617344915867,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 1.1768349409103394,
+      "learning_rate": 0.0003416901607502972,
+      "loss": 0.6142709255218506,
+      "mean_token_accuracy": 0.8220669947564602,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6150523993406403,
+      "eval_loss": 0.7019563317298889,
+      "eval_mean_token_accuracy": 0.8094323099998946,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 89.419,
+      "eval_samples_per_second": 15.88,
+      "eval_steps_per_second": 1.991,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6126735735684633,
+      "epoch": 2.506940253470127,
+      "grad_norm": 1.8065496683120728,
+      "learning_rate": 0.00034017533172263055,
+      "loss": 0.6056458473205566,
+      "mean_token_accuracy": 0.8203762136399746,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6544345456562685,
+      "eval_loss": 0.7086517214775085,
+      "eval_mean_token_accuracy": 0.806709805901131,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 88.981,
+      "eval_samples_per_second": 15.958,
+      "eval_steps_per_second": 2.0,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6311754353344441,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 1.0576369762420654,
+      "learning_rate": 0.00033861584314221236,
+      "loss": 0.6176392555236816,
+      "mean_token_accuracy": 0.8177410490810871,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6273799200406235,
+      "eval_loss": 0.708163321018219,
+      "eval_mean_token_accuracy": 0.8032017099053672,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 89.7668,
+      "eval_samples_per_second": 15.819,
+      "eval_steps_per_second": 1.983,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6200249589979648,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.997065007686615,
+      "learning_rate": 0.0003370121363258637,
+      "loss": 0.6203888893127442,
+      "mean_token_accuracy": 0.8196275025606156,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6457576811983344,
+      "eval_loss": 0.6972290277481079,
+      "eval_mean_token_accuracy": 0.8095980849158898,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 89.2027,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 1080
+    },
+    {
+      "entropy": 0.626810473203659,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.865475058555603,
+      "learning_rate": 0.0003353646651036438,
+      "loss": 0.6102585792541504,
+      "mean_token_accuracy": 0.819708751142025,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.6412022431914726,
+      "eval_loss": 0.7036139965057373,
+      "eval_mean_token_accuracy": 0.8070317135098275,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 89.4733,
+      "eval_samples_per_second": 15.871,
+      "eval_steps_per_second": 1.989,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6581619590520859,
+      "epoch": 2.700060350030175,
+      "grad_norm": 1.00307035446167,
+      "learning_rate": 0.00033367389569042064,
+      "loss": 0.6397720336914062,
+      "mean_token_accuracy": 0.81413309648633,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6288736402318719,
+      "eval_loss": 0.6966825723648071,
+      "eval_mean_token_accuracy": 0.8094039669867312,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 88.7341,
+      "eval_samples_per_second": 16.003,
+      "eval_steps_per_second": 2.006,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6399412982165813,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.8589329719543457,
+      "learning_rate": 0.0003319403065539384,
+      "loss": 0.6347338199615479,
+      "mean_token_accuracy": 0.8139534957706929,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.683491913455256,
+      "eval_loss": 0.6984149217605591,
+      "eval_mean_token_accuracy": 0.8083989469522841,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 89.3929,
+      "eval_samples_per_second": 15.885,
+      "eval_steps_per_second": 1.991,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6264835461974144,
+      "epoch": 2.796620398310199,
+      "grad_norm": 0.7807853817939758,
+      "learning_rate": 0.0003301643882794163,
+      "loss": 0.6162629127502441,
+      "mean_token_accuracy": 0.8216292977333068,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6578792159476977,
+      "eval_loss": 0.6921441555023193,
+      "eval_mean_token_accuracy": 0.8094299177775223,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 89.2356,
+      "eval_samples_per_second": 15.913,
+      "eval_steps_per_second": 1.995,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6269129011780024,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 1.1675593852996826,
+      "learning_rate": 0.0003283466434307189,
+      "loss": 0.6235532760620117,
+      "mean_token_accuracy": 0.8188897788524627,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6533248447970058,
+      "eval_loss": 0.696849524974823,
+      "eval_mean_token_accuracy": 0.8102703713968898,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2602,
+      "eval_samples_per_second": 16.089,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.6259873781353236,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.9596486687660217,
+      "learning_rate": 0.00032648758640813655,
+      "loss": 0.6118728160858155,
+      "mean_token_accuracy": 0.8171859130263328,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6212078754821521,
+      "eval_loss": 0.6942870020866394,
+      "eval_mean_token_accuracy": 0.810484343365337,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 89.4375,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 1200
+    },
+    {
+      "entropy": 0.6142537571489811,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 0.9625058770179749,
+      "learning_rate": 0.00032458774330281615,
+      "loss": 0.6060261249542236,
+      "mean_token_accuracy": 0.8223113484680653,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6553665667437436,
+      "eval_loss": 0.6835269331932068,
+      "eval_mean_token_accuracy": 0.8134723968720168,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 89.5092,
+      "eval_samples_per_second": 15.864,
+      "eval_steps_per_second": 1.989,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6301972426474094,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.9657333493232727,
+      "learning_rate": 0.0003226476517478835,
+      "loss": 0.6277759075164795,
+      "mean_token_accuracy": 0.8164266042411328,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6594253118788258,
+      "eval_loss": 0.6820935606956482,
+      "eval_mean_token_accuracy": 0.8132368198941263,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 89.6211,
+      "eval_samples_per_second": 15.844,
+      "eval_steps_per_second": 1.986,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5346100655469027,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8541185259819031,
+      "learning_rate": 0.0003206678607662996,
+      "loss": 0.5261692047119141,
+      "mean_token_accuracy": 0.8382289425119177,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5626645662476507,
+      "eval_loss": 0.7028542160987854,
+      "eval_mean_token_accuracy": 0.8133395572056931,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 89.1267,
+      "eval_samples_per_second": 15.932,
+      "eval_steps_per_second": 1.997,
+      "step": 1260
+    },
+    {
+      "entropy": 0.5001101832836866,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.8208815455436707,
+      "learning_rate": 0.0003186489306154935,
+      "loss": 0.4850132942199707,
+      "mean_token_accuracy": 0.8468983843922615,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5889462046743779,
+      "eval_loss": 0.691015899181366,
+      "eval_mean_token_accuracy": 0.8135226466012805,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.9677,
+      "eval_samples_per_second": 15.961,
+      "eval_steps_per_second": 2.001,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5242766339331866,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 0.7893072962760925,
+      "learning_rate": 0.0003165914326288163,
+      "loss": 0.5061577320098877,
+      "mean_token_accuracy": 0.8425608821213245,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5543691962957382,
+      "eval_loss": 0.6913734078407288,
+      "eval_mean_token_accuracy": 0.8154889680026622,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 89.6963,
+      "eval_samples_per_second": 15.831,
+      "eval_steps_per_second": 1.984,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5327721010893584,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.8992080092430115,
+      "learning_rate": 0.0003144959490538604,
+      "loss": 0.5068036556243897,
+      "mean_token_accuracy": 0.8420542575418949,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5734540803378887,
+      "eval_loss": 0.6948716044425964,
+      "eval_mean_token_accuracy": 0.8137217944257715,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 89.4597,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5247377116233111,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.9291845560073853,
+      "learning_rate": 0.0003123630728876902,
+      "loss": 0.5023272037506104,
+      "mean_token_accuracy": 0.8435182586312294,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5492362935891312,
+      "eval_loss": 0.6980633735656738,
+      "eval_mean_token_accuracy": 0.8155082919624415,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 89.2124,
+      "eval_samples_per_second": 15.917,
+      "eval_steps_per_second": 1.995,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5380499072372913,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 1.0477524995803833,
+      "learning_rate": 0.00031019340770903136,
+      "loss": 0.5264591693878173,
+      "mean_token_accuracy": 0.8391959741711617,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5896503307511297,
+      "eval_loss": 0.6997144222259521,
+      "eval_mean_token_accuracy": 0.8124373062942805,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 89.1147,
+      "eval_samples_per_second": 15.935,
+      "eval_steps_per_second": 1.997,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5460653610527515,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 1.2473069429397583,
+      "learning_rate": 0.00030798756750746476,
+      "loss": 0.5319677829742432,
+      "mean_token_accuracy": 0.8364320226013661,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5659052182114526,
+      "eval_loss": 0.6924049258232117,
+      "eval_mean_token_accuracy": 0.8153053585732921,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 89.6468,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5367407951503992,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 1.2679712772369385,
+      "learning_rate": 0.00030574617650967485,
+      "loss": 0.5190927028656006,
+      "mean_token_accuracy": 0.8399469174444676,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5581759144081159,
+      "eval_loss": 0.6933317184448242,
+      "eval_mean_token_accuracy": 0.8139757862251796,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 89.0749,
+      "eval_samples_per_second": 15.942,
+      "eval_steps_per_second": 1.998,
+      "step": 1400
+    },
+    {
+      "entropy": 0.530736118927598,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.9717612862586975,
+      "learning_rate": 0.0003034698690028009,
+      "loss": 0.512363052368164,
+      "mean_token_accuracy": 0.8419127956032753,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5233335836549823,
+      "eval_loss": 0.6958550214767456,
+      "eval_mean_token_accuracy": 0.8158778501360604,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 88.8664,
+      "eval_samples_per_second": 15.979,
+      "eval_steps_per_second": 2.003,
+      "step": 1420
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.4238859163696333e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1440/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1440/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1440/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1440/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1440/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..7ce4b072fcbe5b636867e13937c56fedee3b68b4
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1440/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.02728394274790723,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1440/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1440/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1440/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1440/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1440/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1440/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1440/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1440/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..e1f02caa202e9bd6910bd5befa5f3b552788b5f4
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1440/trainer_state.json
@@ -0,0 +1,1546 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 3.4707302353651177,
+  "eval_steps": 20,
+  "global_step": 1440,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.024795526266098,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.945375919342041,
+      "learning_rate": 1.6698127428345936e-05,
+      "loss": 1.7825420379638672,
+      "mean_token_accuracy": 0.6315708436071873,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.29921497588747,
+      "eval_loss": 1.169131875038147,
+      "eval_mean_token_accuracy": 0.7185706342204233,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 91.7898,
+      "eval_samples_per_second": 15.47,
+      "eval_steps_per_second": 1.939,
+      "step": 20
+    },
+    {
+      "entropy": 1.0346344470977784,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.7916979789733887,
+      "learning_rate": 3.427510366871008e-05,
+      "loss": 0.9480395317077637,
+      "mean_token_accuracy": 0.7543429024517536,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 0.9673695430326997,
+      "eval_loss": 0.8588430285453796,
+      "eval_mean_token_accuracy": 0.7737233002534073,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 89.8219,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 40
+    },
+    {
+      "entropy": 0.8977566614747048,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7972187995910645,
+      "learning_rate": 5.185207990907422e-05,
+      "loss": 0.8029604911804199,
+      "mean_token_accuracy": 0.7818016350269318,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8436836982041263,
+      "eval_loss": 0.7893033623695374,
+      "eval_mean_token_accuracy": 0.7858914984076211,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 89.9349,
+      "eval_samples_per_second": 15.789,
+      "eval_steps_per_second": 1.979,
+      "step": 60
+    },
+    {
+      "entropy": 0.8650321319699288,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.2128729820251465,
+      "learning_rate": 6.942905614943836e-05,
+      "loss": 0.77417893409729,
+      "mean_token_accuracy": 0.7875479467213153,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8209857180547179,
+      "eval_loss": 0.7572280168533325,
+      "eval_mean_token_accuracy": 0.7925970383574453,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 89.9459,
+      "eval_samples_per_second": 15.787,
+      "eval_steps_per_second": 1.979,
+      "step": 80
+    },
+    {
+      "entropy": 0.8096660938113928,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.1254287958145142,
+      "learning_rate": 8.700603238980251e-05,
+      "loss": 0.7170331001281738,
+      "mean_token_accuracy": 0.7994555257260799,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.792780208788561,
+      "eval_loss": 0.7442984580993652,
+      "eval_mean_token_accuracy": 0.7954704573984896,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 90.294,
+      "eval_samples_per_second": 15.726,
+      "eval_steps_per_second": 1.971,
+      "step": 100
+    },
+    {
+      "entropy": 0.7979829199612141,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2140507698059082,
+      "learning_rate": 0.00010458300863016666,
+      "loss": 0.7141861915588379,
+      "mean_token_accuracy": 0.8000014744699001,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.7938692486018277,
+      "eval_loss": 0.7279797196388245,
+      "eval_mean_token_accuracy": 0.7989798734027348,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 89.7189,
+      "eval_samples_per_second": 15.827,
+      "eval_steps_per_second": 1.984,
+      "step": 120
+    },
+    {
+      "entropy": 0.7916347607970238,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.0506458282470703,
+      "learning_rate": 0.0001221599848705308,
+      "loss": 0.7117055416107178,
+      "mean_token_accuracy": 0.7970046654343605,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.8052781063519167,
+      "eval_loss": 0.722576916217804,
+      "eval_mean_token_accuracy": 0.7982530523551984,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 90.0376,
+      "eval_samples_per_second": 15.771,
+      "eval_steps_per_second": 1.977,
+      "step": 140
+    },
+    {
+      "entropy": 0.7880666613578796,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8409207463264465,
+      "learning_rate": 0.00013973696111089492,
+      "loss": 0.7052523136138916,
+      "mean_token_accuracy": 0.8000123649835587,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7787443134891853,
+      "eval_loss": 0.7137772440910339,
+      "eval_mean_token_accuracy": 0.8014209287220173,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 90.2562,
+      "eval_samples_per_second": 15.733,
+      "eval_steps_per_second": 1.972,
+      "step": 160
+    },
+    {
+      "entropy": 0.7696007996797561,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.0182477235794067,
+      "learning_rate": 0.00015731393735125907,
+      "loss": 0.7008797645568847,
+      "mean_token_accuracy": 0.8007259473204613,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.752259682068664,
+      "eval_loss": 0.7171286344528198,
+      "eval_mean_token_accuracy": 0.80052122574174,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 90.0797,
+      "eval_samples_per_second": 15.764,
+      "eval_steps_per_second": 1.976,
+      "step": 180
+    },
+    {
+      "entropy": 0.7828126326203346,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9553632736206055,
+      "learning_rate": 0.0001748909135916232,
+      "loss": 0.7046853542327881,
+      "mean_token_accuracy": 0.7988562889397144,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7389638169427936,
+      "eval_loss": 0.7187935709953308,
+      "eval_mean_token_accuracy": 0.7995899999409579,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 89.9223,
+      "eval_samples_per_second": 15.791,
+      "eval_steps_per_second": 1.979,
+      "step": 200
+    },
+    {
+      "entropy": 0.7909286297857762,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 1.134459376335144,
+      "learning_rate": 0.00019246788983198735,
+      "loss": 0.6989312171936035,
+      "mean_token_accuracy": 0.7991565138101577,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7496965199374082,
+      "eval_loss": 0.7106321454048157,
+      "eval_mean_token_accuracy": 0.8046183442131857,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 89.6899,
+      "eval_samples_per_second": 15.832,
+      "eval_steps_per_second": 1.985,
+      "step": 220
+    },
+    {
+      "entropy": 0.7651963323354721,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.9557585716247559,
+      "learning_rate": 0.0002100448660723515,
+      "loss": 0.6984559059143066,
+      "mean_token_accuracy": 0.8019823372364044,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7615563079212488,
+      "eval_loss": 0.7098539471626282,
+      "eval_mean_token_accuracy": 0.8047233508544022,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 90.3752,
+      "eval_samples_per_second": 15.712,
+      "eval_steps_per_second": 1.97,
+      "step": 240
+    },
+    {
+      "entropy": 0.787507726252079,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.9279311299324036,
+      "learning_rate": 0.00022762184231271568,
+      "loss": 0.708138370513916,
+      "mean_token_accuracy": 0.8013308703899383,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7856640896100676,
+      "eval_loss": 0.7191570401191711,
+      "eval_mean_token_accuracy": 0.8024677249153008,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 93.7336,
+      "eval_samples_per_second": 15.149,
+      "eval_steps_per_second": 1.899,
+      "step": 260
+    },
+    {
+      "entropy": 0.768489234894514,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.7992776036262512,
+      "learning_rate": 0.0002451988185530798,
+      "loss": 0.7035849571228028,
+      "mean_token_accuracy": 0.8031105332076549,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7682498740346244,
+      "eval_loss": 0.7149022221565247,
+      "eval_mean_token_accuracy": 0.7973497995499814,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 89.7309,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 280
+    },
+    {
+      "entropy": 0.7720955617725849,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.1931514739990234,
+      "learning_rate": 0.0002627757947934439,
+      "loss": 0.6965402603149414,
+      "mean_token_accuracy": 0.8040348663926125,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7214708304807042,
+      "eval_loss": 0.7350823879241943,
+      "eval_mean_token_accuracy": 0.7942911925610532,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 89.6451,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 300
+    },
+    {
+      "entropy": 0.7930095426738262,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8994467258453369,
+      "learning_rate": 0.0002803527710338081,
+      "loss": 0.7331492900848389,
+      "mean_token_accuracy": 0.796185302734375,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.8142535488926963,
+      "eval_loss": 0.7237834334373474,
+      "eval_mean_token_accuracy": 0.799495273761535,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 90.6759,
+      "eval_samples_per_second": 15.66,
+      "eval_steps_per_second": 1.963,
+      "step": 320
+    },
+    {
+      "entropy": 0.7825102441012859,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 1.8224239349365234,
+      "learning_rate": 0.00029792974727417223,
+      "loss": 0.7331175804138184,
+      "mean_token_accuracy": 0.7977667152881622,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7327135443017724,
+      "eval_loss": 0.7388784289360046,
+      "eval_mean_token_accuracy": 0.8007041252730938,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 89.7692,
+      "eval_samples_per_second": 15.818,
+      "eval_steps_per_second": 1.983,
+      "step": 340
+    },
+    {
+      "entropy": 0.819331557303667,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 1.0407779216766357,
+      "learning_rate": 0.0003155067235145364,
+      "loss": 0.7456695556640625,
+      "mean_token_accuracy": 0.7936980701982975,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.8065585420372781,
+      "eval_loss": 0.735519528388977,
+      "eval_mean_token_accuracy": 0.7974206296245704,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 89.4336,
+      "eval_samples_per_second": 15.878,
+      "eval_steps_per_second": 1.99,
+      "step": 360
+    },
+    {
+      "entropy": 0.8114653021097183,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.1209490299224854,
+      "learning_rate": 0.0003330836997549005,
+      "loss": 0.7340899467468261,
+      "mean_token_accuracy": 0.7914272703230381,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.8025527285056168,
+      "eval_loss": 0.7295576930046082,
+      "eval_mean_token_accuracy": 0.7990576628218876,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 89.9579,
+      "eval_samples_per_second": 15.785,
+      "eval_steps_per_second": 1.979,
+      "step": 380
+    },
+    {
+      "entropy": 0.8042497783899307,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 1.2437798976898193,
+      "learning_rate": 0.0003506606759952647,
+      "loss": 0.7381744384765625,
+      "mean_token_accuracy": 0.7967362694442273,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7717011336530193,
+      "eval_loss": 0.7436173558235168,
+      "eval_mean_token_accuracy": 0.7975849482450592,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 89.8411,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 400
+    },
+    {
+      "entropy": 0.7987750010056929,
+      "epoch": 1.012070006035003,
+      "grad_norm": 1.2390918731689453,
+      "learning_rate": 0.000364721224843344,
+      "loss": 0.7317790031433106,
+      "mean_token_accuracy": 0.7986087551364651,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7333170604170038,
+      "eval_loss": 0.7456948757171631,
+      "eval_mean_token_accuracy": 0.7964897577682238,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 89.7288,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 420
+    },
+    {
+      "entropy": 0.7372976846992969,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.1552740335464478,
+      "learning_rate": 0.00036468510102269575,
+      "loss": 0.6734249114990234,
+      "mean_token_accuracy": 0.805834824591875,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.7130355526891987,
+      "eval_loss": 0.7555333375930786,
+      "eval_mean_token_accuracy": 0.7961652781186479,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 89.8713,
+      "eval_samples_per_second": 15.8,
+      "eval_steps_per_second": 1.981,
+      "step": 440
+    },
+    {
+      "entropy": 0.7465668715536594,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 1.1991469860076904,
+      "learning_rate": 0.0003645973816745066,
+      "loss": 0.6871879577636719,
+      "mean_token_accuracy": 0.8042450234293937,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7772979542110743,
+      "eval_loss": 0.7561826705932617,
+      "eval_mean_token_accuracy": 0.7953876427720102,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 89.8395,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 460
+    },
+    {
+      "entropy": 0.7477855801582336,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 1.325486660003662,
+      "learning_rate": 0.00036445809162231435,
+      "loss": 0.6836830139160156,
+      "mean_token_accuracy": 0.8021619468927383,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.7098110676481483,
+      "eval_loss": 0.758007287979126,
+      "eval_mean_token_accuracy": 0.7909953189030122,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 89.6101,
+      "eval_samples_per_second": 15.846,
+      "eval_steps_per_second": 1.986,
+      "step": 480
+    },
+    {
+      "entropy": 0.7632691070437432,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 1.0122262239456177,
+      "learning_rate": 0.0003642672702835562,
+      "loss": 0.7071213245391845,
+      "mean_token_accuracy": 0.8000680930912495,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7645382720432924,
+      "eval_loss": 0.7492606043815613,
+      "eval_mean_token_accuracy": 0.7976858676149604,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 89.6092,
+      "eval_samples_per_second": 15.847,
+      "eval_steps_per_second": 1.986,
+      "step": 500
+    },
+    {
+      "entropy": 0.7623726457357407,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.367774486541748,
+      "learning_rate": 0.00036402497165841384,
+      "loss": 0.6965175151824952,
+      "mean_token_accuracy": 0.8016501650214195,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7868083715438843,
+      "eval_loss": 0.7481449842453003,
+      "eval_mean_token_accuracy": 0.799330582779445,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 89.8241,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 520
+    },
+    {
+      "entropy": 0.7535207405686378,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 1.213614821434021,
+      "learning_rate": 0.00036373126431453207,
+      "loss": 0.7078310489654541,
+      "mean_token_accuracy": 0.7998907208442688,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7829091950748743,
+      "eval_loss": 0.7416072487831116,
+      "eval_mean_token_accuracy": 0.7969146250339036,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 89.4607,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 540
+    },
+    {
+      "entropy": 0.7527363449335098,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 1.0929982662200928,
+      "learning_rate": 0.00036338623136761495,
+      "loss": 0.7058369636535644,
+      "mean_token_accuracy": 0.7997087433934211,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.728419938114252,
+      "eval_loss": 0.744873046875,
+      "eval_mean_token_accuracy": 0.7966634296299366,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 89.6597,
+      "eval_samples_per_second": 15.838,
+      "eval_steps_per_second": 1.985,
+      "step": 560
+    },
+    {
+      "entropy": 0.7406099259853363,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.409364104270935,
+      "learning_rate": 0.00036298997045790515,
+      "loss": 0.6902852058410645,
+      "mean_token_accuracy": 0.805386807769537,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.6991632758231645,
+      "eval_loss": 0.7480319738388062,
+      "eval_mean_token_accuracy": 0.7970068006033308,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 89.8329,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 580
+    },
+    {
+      "entropy": 0.7459449704736472,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.7053111791610718,
+      "learning_rate": 0.00036254259372255277,
+      "loss": 0.7044640064239502,
+      "mean_token_accuracy": 0.8005018755793571,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7454566660891758,
+      "eval_loss": 0.7551484704017639,
+      "eval_mean_token_accuracy": 0.7897093731365846,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 89.8313,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 600
+    },
+    {
+      "entropy": 0.7621391348540782,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 1.174443244934082,
+      "learning_rate": 0.000362044227763882,
+      "loss": 0.7111573696136475,
+      "mean_token_accuracy": 0.798752411454916,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7664387949397055,
+      "eval_loss": 0.7365804314613342,
+      "eval_mean_token_accuracy": 0.7963101200843126,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 89.9623,
+      "eval_samples_per_second": 15.784,
+      "eval_steps_per_second": 1.979,
+      "step": 620
+    },
+    {
+      "entropy": 0.7424944408237935,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.2898184061050415,
+      "learning_rate": 0.000361495013613564,
+      "loss": 0.6913110733032226,
+      "mean_token_accuracy": 0.8009015507996082,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7255479529332579,
+      "eval_loss": 0.7345991730690002,
+      "eval_mean_token_accuracy": 0.8013592945056015,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 89.817,
+      "eval_samples_per_second": 15.81,
+      "eval_steps_per_second": 1.982,
+      "step": 640
+    },
+    {
+      "entropy": 0.7303367637097835,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.1310476064682007,
+      "learning_rate": 0.00036089510669270666,
+      "loss": 0.6875617027282714,
+      "mean_token_accuracy": 0.8035397171974182,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7101308248016271,
+      "eval_loss": 0.7394917011260986,
+      "eval_mean_token_accuracy": 0.7990792942850777,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 89.7431,
+      "eval_samples_per_second": 15.823,
+      "eval_steps_per_second": 1.983,
+      "step": 660
+    },
+    {
+      "entropy": 0.7387678384780884,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.9645389914512634,
+      "learning_rate": 0.0003602446767678725,
+      "loss": 0.6911204338073731,
+      "mean_token_accuracy": 0.8056333385407924,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7281073737010527,
+      "eval_loss": 0.7376708984375,
+      "eval_mean_token_accuracy": 0.7997356415464637,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 90.0338,
+      "eval_samples_per_second": 15.772,
+      "eval_steps_per_second": 1.977,
+      "step": 680
+    },
+    {
+      "entropy": 0.7449554048478604,
+      "epoch": 1.687990343995172,
+      "grad_norm": 1.9069629907608032,
+      "learning_rate": 0.0003595439079030364,
+      "loss": 0.7042286396026611,
+      "mean_token_accuracy": 0.8018639378249646,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7022633817088738,
+      "eval_loss": 0.7320939898490906,
+      "eval_mean_token_accuracy": 0.8006051541044471,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 89.8948,
+      "eval_samples_per_second": 15.796,
+      "eval_steps_per_second": 1.98,
+      "step": 700
+    },
+    {
+      "entropy": 0.723249051719904,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.8333877325057983,
+      "learning_rate": 0.00035879299840749777,
+      "loss": 0.7043714046478271,
+      "mean_token_accuracy": 0.8028345607221127,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.672436946898364,
+      "eval_loss": 0.713233470916748,
+      "eval_mean_token_accuracy": 0.8047760577684038,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 89.6394,
+      "eval_samples_per_second": 15.841,
+      "eval_steps_per_second": 1.986,
+      "step": 720
+    },
+    {
+      "entropy": 0.730014206469059,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.4580037593841553,
+      "learning_rate": 0.00035799216077976137,
+      "loss": 0.6985176563262939,
+      "mean_token_accuracy": 0.8015547141432762,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.6908873795123582,
+      "eval_loss": 0.7198202013969421,
+      "eval_mean_token_accuracy": 0.8024131682481659,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 89.3771,
+      "eval_samples_per_second": 15.888,
+      "eval_steps_per_second": 1.992,
+      "step": 740
+    },
+    {
+      "entropy": 0.7219605796039105,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.51682710647583,
+      "learning_rate": 0.000357141621647403,
+      "loss": 0.6844424724578857,
+      "mean_token_accuracy": 0.8032813094556331,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7113401109582922,
+      "eval_loss": 0.7195309400558472,
+      "eval_mean_token_accuracy": 0.8019187035185568,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 89.4658,
+      "eval_samples_per_second": 15.872,
+      "eval_steps_per_second": 1.99,
+      "step": 760
+    },
+    {
+      "entropy": 0.729826743900776,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 1.0838000774383545,
+      "learning_rate": 0.0003562416217029361,
+      "loss": 0.685858678817749,
+      "mean_token_accuracy": 0.8050567395985126,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.649370267484965,
+      "eval_loss": 0.7177029252052307,
+      "eval_mean_token_accuracy": 0.8050174043419656,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 89.2287,
+      "eval_samples_per_second": 15.914,
+      "eval_steps_per_second": 1.995,
+      "step": 780
+    },
+    {
+      "entropy": 0.6799364153295755,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 1.1364924907684326,
+      "learning_rate": 0.00035529241563569903,
+      "loss": 0.6714428901672364,
+      "mean_token_accuracy": 0.8087341658771038,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6603799047095052,
+      "eval_loss": 0.7201809287071228,
+      "eval_mean_token_accuracy": 0.802863868769635,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 89.2718,
+      "eval_samples_per_second": 15.906,
+      "eval_steps_per_second": 1.994,
+      "step": 800
+    },
+    {
+      "entropy": 0.7078610375523567,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 1.2191942930221558,
+      "learning_rate": 0.0003542942720597808,
+      "loss": 0.6996590614318847,
+      "mean_token_accuracy": 0.8049512051045895,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.7240545264120852,
+      "eval_loss": 0.7301309108734131,
+      "eval_mean_token_accuracy": 0.8015973166133581,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 89.3438,
+      "eval_samples_per_second": 15.894,
+      "eval_steps_per_second": 1.992,
+      "step": 820
+    },
+    {
+      "entropy": 0.6557391556826505,
+      "epoch": 2.024140012070006,
+      "grad_norm": 1.36203134059906,
+      "learning_rate": 0.0003532474734380064,
+      "loss": 0.6327113628387451,
+      "mean_token_accuracy": 0.8142829847026181,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6573808648613062,
+      "eval_loss": 0.7334365844726562,
+      "eval_mean_token_accuracy": 0.8043633893634496,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 89.3359,
+      "eval_samples_per_second": 15.895,
+      "eval_steps_per_second": 1.992,
+      "step": 840
+    },
+    {
+      "entropy": 0.6409128502011299,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 1.5804955959320068,
+      "learning_rate": 0.0003521523160020035,
+      "loss": 0.6023559093475341,
+      "mean_token_accuracy": 0.8222251623868942,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6983234788594621,
+      "eval_loss": 0.7232135534286499,
+      "eval_mean_token_accuracy": 0.8010654268639811,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 89.2615,
+      "eval_samples_per_second": 15.908,
+      "eval_steps_per_second": 1.994,
+      "step": 860
+    },
+    {
+      "entropy": 0.6151149723678827,
+      "epoch": 2.12070006035003,
+      "grad_norm": 1.1975436210632324,
+      "learning_rate": 0.00035100910966837193,
+      "loss": 0.5803564548492431,
+      "mean_token_accuracy": 0.8256325736641884,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6636555573243773,
+      "eval_loss": 0.7205662727355957,
+      "eval_mean_token_accuracy": 0.8036768369460374,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 89.2957,
+      "eval_samples_per_second": 15.902,
+      "eval_steps_per_second": 1.993,
+      "step": 880
+    },
+    {
+      "entropy": 0.6201727617532015,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9907131195068359,
+      "learning_rate": 0.0003498181779509813,
+      "loss": 0.6101872444152832,
+      "mean_token_accuracy": 0.8196512959897518,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.6473788086617931,
+      "eval_loss": 0.7062397003173828,
+      "eval_mean_token_accuracy": 0.8078362184963869,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 88.6198,
+      "eval_samples_per_second": 16.024,
+      "eval_steps_per_second": 2.009,
+      "step": 900
+    },
+    {
+      "entropy": 0.6097332935780286,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.9055633544921875,
+      "learning_rate": 0.00034857985786942026,
+      "loss": 0.5956094264984131,
+      "mean_token_accuracy": 0.8240575887262821,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.612629868341296,
+      "eval_loss": 0.7106770277023315,
+      "eval_mean_token_accuracy": 0.805324277181304,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 89.4383,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 920
+    },
+    {
+      "entropy": 0.6182701248675585,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.9513736367225647,
+      "learning_rate": 0.00034729449985362404,
+      "loss": 0.6019269466400147,
+      "mean_token_accuracy": 0.8187106013298034,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6437647456533453,
+      "eval_loss": 0.7173502445220947,
+      "eval_mean_token_accuracy": 0.8056439573175451,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.9568,
+      "eval_samples_per_second": 15.963,
+      "eval_steps_per_second": 2.001,
+      "step": 940
+    },
+    {
+      "entropy": 0.6508570514619351,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.9606207013130188,
+      "learning_rate": 0.0003459624676447067,
+      "loss": 0.6296409606933594,
+      "mean_token_accuracy": 0.8167732007801533,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6288247135248077,
+      "eval_loss": 0.7160675525665283,
+      "eval_mean_token_accuracy": 0.8066010689467527,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 89.0044,
+      "eval_samples_per_second": 15.954,
+      "eval_steps_per_second": 2.0,
+      "step": 960
+    },
+    {
+      "entropy": 0.6380081418901682,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.9677181839942932,
+      "learning_rate": 0.000344584138192027,
+      "loss": 0.6210546493530273,
+      "mean_token_accuracy": 0.8186135642230511,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6407805031604981,
+      "eval_loss": 0.7055043578147888,
+      "eval_mean_token_accuracy": 0.8070103157772107,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 89.5606,
+      "eval_samples_per_second": 15.855,
+      "eval_steps_per_second": 1.987,
+      "step": 980
+    },
+    {
+      "entropy": 0.6310251638293266,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 2.1046407222747803,
+      "learning_rate": 0.000343159901546516,
+      "loss": 0.6165830135345459,
+      "mean_token_accuracy": 0.8207855716347694,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6461507470420237,
+      "eval_loss": 0.7292136549949646,
+      "eval_mean_token_accuracy": 0.8030418261383356,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 89.315,
+      "eval_samples_per_second": 15.899,
+      "eval_steps_per_second": 1.993,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6256617344915867,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 1.1768349409103394,
+      "learning_rate": 0.0003416901607502972,
+      "loss": 0.6142709255218506,
+      "mean_token_accuracy": 0.8220669947564602,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6150523993406403,
+      "eval_loss": 0.7019563317298889,
+      "eval_mean_token_accuracy": 0.8094323099998946,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 89.419,
+      "eval_samples_per_second": 15.88,
+      "eval_steps_per_second": 1.991,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6126735735684633,
+      "epoch": 2.506940253470127,
+      "grad_norm": 1.8065496683120728,
+      "learning_rate": 0.00034017533172263055,
+      "loss": 0.6056458473205566,
+      "mean_token_accuracy": 0.8203762136399746,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6544345456562685,
+      "eval_loss": 0.7086517214775085,
+      "eval_mean_token_accuracy": 0.806709805901131,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 88.981,
+      "eval_samples_per_second": 15.958,
+      "eval_steps_per_second": 2.0,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6311754353344441,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 1.0576369762420654,
+      "learning_rate": 0.00033861584314221236,
+      "loss": 0.6176392555236816,
+      "mean_token_accuracy": 0.8177410490810871,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6273799200406235,
+      "eval_loss": 0.708163321018219,
+      "eval_mean_token_accuracy": 0.8032017099053672,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 89.7668,
+      "eval_samples_per_second": 15.819,
+      "eval_steps_per_second": 1.983,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6200249589979648,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.997065007686615,
+      "learning_rate": 0.0003370121363258637,
+      "loss": 0.6203888893127442,
+      "mean_token_accuracy": 0.8196275025606156,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6457576811983344,
+      "eval_loss": 0.6972290277481079,
+      "eval_mean_token_accuracy": 0.8095980849158898,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 89.2027,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 1080
+    },
+    {
+      "entropy": 0.626810473203659,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.865475058555603,
+      "learning_rate": 0.0003353646651036438,
+      "loss": 0.6102585792541504,
+      "mean_token_accuracy": 0.819708751142025,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.6412022431914726,
+      "eval_loss": 0.7036139965057373,
+      "eval_mean_token_accuracy": 0.8070317135098275,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 89.4733,
+      "eval_samples_per_second": 15.871,
+      "eval_steps_per_second": 1.989,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6581619590520859,
+      "epoch": 2.700060350030175,
+      "grad_norm": 1.00307035446167,
+      "learning_rate": 0.00033367389569042064,
+      "loss": 0.6397720336914062,
+      "mean_token_accuracy": 0.81413309648633,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6288736402318719,
+      "eval_loss": 0.6966825723648071,
+      "eval_mean_token_accuracy": 0.8094039669867312,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 88.7341,
+      "eval_samples_per_second": 16.003,
+      "eval_steps_per_second": 2.006,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6399412982165813,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.8589329719543457,
+      "learning_rate": 0.0003319403065539384,
+      "loss": 0.6347338199615479,
+      "mean_token_accuracy": 0.8139534957706929,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.683491913455256,
+      "eval_loss": 0.6984149217605591,
+      "eval_mean_token_accuracy": 0.8083989469522841,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 89.3929,
+      "eval_samples_per_second": 15.885,
+      "eval_steps_per_second": 1.991,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6264835461974144,
+      "epoch": 2.796620398310199,
+      "grad_norm": 0.7807853817939758,
+      "learning_rate": 0.0003301643882794163,
+      "loss": 0.6162629127502441,
+      "mean_token_accuracy": 0.8216292977333068,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6578792159476977,
+      "eval_loss": 0.6921441555023193,
+      "eval_mean_token_accuracy": 0.8094299177775223,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 89.2356,
+      "eval_samples_per_second": 15.913,
+      "eval_steps_per_second": 1.995,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6269129011780024,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 1.1675593852996826,
+      "learning_rate": 0.0003283466434307189,
+      "loss": 0.6235532760620117,
+      "mean_token_accuracy": 0.8188897788524627,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6533248447970058,
+      "eval_loss": 0.696849524974823,
+      "eval_mean_token_accuracy": 0.8102703713968898,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2602,
+      "eval_samples_per_second": 16.089,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.6259873781353236,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.9596486687660217,
+      "learning_rate": 0.00032648758640813655,
+      "loss": 0.6118728160858155,
+      "mean_token_accuracy": 0.8171859130263328,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6212078754821521,
+      "eval_loss": 0.6942870020866394,
+      "eval_mean_token_accuracy": 0.810484343365337,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 89.4375,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 1200
+    },
+    {
+      "entropy": 0.6142537571489811,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 0.9625058770179749,
+      "learning_rate": 0.00032458774330281615,
+      "loss": 0.6060261249542236,
+      "mean_token_accuracy": 0.8223113484680653,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6553665667437436,
+      "eval_loss": 0.6835269331932068,
+      "eval_mean_token_accuracy": 0.8134723968720168,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 89.5092,
+      "eval_samples_per_second": 15.864,
+      "eval_steps_per_second": 1.989,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6301972426474094,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.9657333493232727,
+      "learning_rate": 0.0003226476517478835,
+      "loss": 0.6277759075164795,
+      "mean_token_accuracy": 0.8164266042411328,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6594253118788258,
+      "eval_loss": 0.6820935606956482,
+      "eval_mean_token_accuracy": 0.8132368198941263,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 89.6211,
+      "eval_samples_per_second": 15.844,
+      "eval_steps_per_second": 1.986,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5346100655469027,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8541185259819031,
+      "learning_rate": 0.0003206678607662996,
+      "loss": 0.5261692047119141,
+      "mean_token_accuracy": 0.8382289425119177,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5626645662476507,
+      "eval_loss": 0.7028542160987854,
+      "eval_mean_token_accuracy": 0.8133395572056931,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 89.1267,
+      "eval_samples_per_second": 15.932,
+      "eval_steps_per_second": 1.997,
+      "step": 1260
+    },
+    {
+      "entropy": 0.5001101832836866,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.8208815455436707,
+      "learning_rate": 0.0003186489306154935,
+      "loss": 0.4850132942199707,
+      "mean_token_accuracy": 0.8468983843922615,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5889462046743779,
+      "eval_loss": 0.691015899181366,
+      "eval_mean_token_accuracy": 0.8135226466012805,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.9677,
+      "eval_samples_per_second": 15.961,
+      "eval_steps_per_second": 2.001,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5242766339331866,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 0.7893072962760925,
+      "learning_rate": 0.0003165914326288163,
+      "loss": 0.5061577320098877,
+      "mean_token_accuracy": 0.8425608821213245,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5543691962957382,
+      "eval_loss": 0.6913734078407288,
+      "eval_mean_token_accuracy": 0.8154889680026622,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 89.6963,
+      "eval_samples_per_second": 15.831,
+      "eval_steps_per_second": 1.984,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5327721010893584,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.8992080092430115,
+      "learning_rate": 0.0003144959490538604,
+      "loss": 0.5068036556243897,
+      "mean_token_accuracy": 0.8420542575418949,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5734540803378887,
+      "eval_loss": 0.6948716044425964,
+      "eval_mean_token_accuracy": 0.8137217944257715,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 89.4597,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5247377116233111,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.9291845560073853,
+      "learning_rate": 0.0003123630728876902,
+      "loss": 0.5023272037506104,
+      "mean_token_accuracy": 0.8435182586312294,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5492362935891312,
+      "eval_loss": 0.6980633735656738,
+      "eval_mean_token_accuracy": 0.8155082919624415,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 89.2124,
+      "eval_samples_per_second": 15.917,
+      "eval_steps_per_second": 1.995,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5380499072372913,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 1.0477524995803833,
+      "learning_rate": 0.00031019340770903136,
+      "loss": 0.5264591693878173,
+      "mean_token_accuracy": 0.8391959741711617,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5896503307511297,
+      "eval_loss": 0.6997144222259521,
+      "eval_mean_token_accuracy": 0.8124373062942805,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 89.1147,
+      "eval_samples_per_second": 15.935,
+      "eval_steps_per_second": 1.997,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5460653610527515,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 1.2473069429397583,
+      "learning_rate": 0.00030798756750746476,
+      "loss": 0.5319677829742432,
+      "mean_token_accuracy": 0.8364320226013661,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5659052182114526,
+      "eval_loss": 0.6924049258232117,
+      "eval_mean_token_accuracy": 0.8153053585732921,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 89.6468,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5367407951503992,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 1.2679712772369385,
+      "learning_rate": 0.00030574617650967485,
+      "loss": 0.5190927028656006,
+      "mean_token_accuracy": 0.8399469174444676,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5581759144081159,
+      "eval_loss": 0.6933317184448242,
+      "eval_mean_token_accuracy": 0.8139757862251796,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 89.0749,
+      "eval_samples_per_second": 15.942,
+      "eval_steps_per_second": 1.998,
+      "step": 1400
+    },
+    {
+      "entropy": 0.530736118927598,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.9717612862586975,
+      "learning_rate": 0.0003034698690028009,
+      "loss": 0.512363052368164,
+      "mean_token_accuracy": 0.8419127956032753,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5233335836549823,
+      "eval_loss": 0.6958550214767456,
+      "eval_mean_token_accuracy": 0.8158778501360604,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 88.8664,
+      "eval_samples_per_second": 15.979,
+      "eval_steps_per_second": 2.003,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5327631575986743,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7649208903312683,
+      "learning_rate": 0.00030115928915494116,
+      "loss": 0.5202179431915284,
+      "mean_token_accuracy": 0.8384683802723885,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.5960972657364406,
+      "eval_loss": 0.6984783411026001,
+      "eval_mean_token_accuracy": 0.8095930597085631,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 89.5168,
+      "eval_samples_per_second": 15.863,
+      "eval_steps_per_second": 1.988,
+      "step": 1440
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.4433804811041178e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1460/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1460/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1460/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1460/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1460/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..7ce4b072fcbe5b636867e13937c56fedee3b68b4
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1460/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.02728394274790723,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1460/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1460/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1460/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1460/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1460/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1460/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1460/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1460/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..11d416f646df5ae460a1f9adf36551579293eb1d
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1460/trainer_state.json
@@ -0,0 +1,1567 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 3.51901025950513,
+  "eval_steps": 20,
+  "global_step": 1460,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.024795526266098,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.945375919342041,
+      "learning_rate": 1.6698127428345936e-05,
+      "loss": 1.7825420379638672,
+      "mean_token_accuracy": 0.6315708436071873,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.29921497588747,
+      "eval_loss": 1.169131875038147,
+      "eval_mean_token_accuracy": 0.7185706342204233,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 91.7898,
+      "eval_samples_per_second": 15.47,
+      "eval_steps_per_second": 1.939,
+      "step": 20
+    },
+    {
+      "entropy": 1.0346344470977784,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.7916979789733887,
+      "learning_rate": 3.427510366871008e-05,
+      "loss": 0.9480395317077637,
+      "mean_token_accuracy": 0.7543429024517536,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 0.9673695430326997,
+      "eval_loss": 0.8588430285453796,
+      "eval_mean_token_accuracy": 0.7737233002534073,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 89.8219,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 40
+    },
+    {
+      "entropy": 0.8977566614747048,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7972187995910645,
+      "learning_rate": 5.185207990907422e-05,
+      "loss": 0.8029604911804199,
+      "mean_token_accuracy": 0.7818016350269318,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8436836982041263,
+      "eval_loss": 0.7893033623695374,
+      "eval_mean_token_accuracy": 0.7858914984076211,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 89.9349,
+      "eval_samples_per_second": 15.789,
+      "eval_steps_per_second": 1.979,
+      "step": 60
+    },
+    {
+      "entropy": 0.8650321319699288,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.2128729820251465,
+      "learning_rate": 6.942905614943836e-05,
+      "loss": 0.77417893409729,
+      "mean_token_accuracy": 0.7875479467213153,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8209857180547179,
+      "eval_loss": 0.7572280168533325,
+      "eval_mean_token_accuracy": 0.7925970383574453,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 89.9459,
+      "eval_samples_per_second": 15.787,
+      "eval_steps_per_second": 1.979,
+      "step": 80
+    },
+    {
+      "entropy": 0.8096660938113928,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.1254287958145142,
+      "learning_rate": 8.700603238980251e-05,
+      "loss": 0.7170331001281738,
+      "mean_token_accuracy": 0.7994555257260799,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.792780208788561,
+      "eval_loss": 0.7442984580993652,
+      "eval_mean_token_accuracy": 0.7954704573984896,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 90.294,
+      "eval_samples_per_second": 15.726,
+      "eval_steps_per_second": 1.971,
+      "step": 100
+    },
+    {
+      "entropy": 0.7979829199612141,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2140507698059082,
+      "learning_rate": 0.00010458300863016666,
+      "loss": 0.7141861915588379,
+      "mean_token_accuracy": 0.8000014744699001,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.7938692486018277,
+      "eval_loss": 0.7279797196388245,
+      "eval_mean_token_accuracy": 0.7989798734027348,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 89.7189,
+      "eval_samples_per_second": 15.827,
+      "eval_steps_per_second": 1.984,
+      "step": 120
+    },
+    {
+      "entropy": 0.7916347607970238,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.0506458282470703,
+      "learning_rate": 0.0001221599848705308,
+      "loss": 0.7117055416107178,
+      "mean_token_accuracy": 0.7970046654343605,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.8052781063519167,
+      "eval_loss": 0.722576916217804,
+      "eval_mean_token_accuracy": 0.7982530523551984,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 90.0376,
+      "eval_samples_per_second": 15.771,
+      "eval_steps_per_second": 1.977,
+      "step": 140
+    },
+    {
+      "entropy": 0.7880666613578796,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8409207463264465,
+      "learning_rate": 0.00013973696111089492,
+      "loss": 0.7052523136138916,
+      "mean_token_accuracy": 0.8000123649835587,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7787443134891853,
+      "eval_loss": 0.7137772440910339,
+      "eval_mean_token_accuracy": 0.8014209287220173,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 90.2562,
+      "eval_samples_per_second": 15.733,
+      "eval_steps_per_second": 1.972,
+      "step": 160
+    },
+    {
+      "entropy": 0.7696007996797561,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.0182477235794067,
+      "learning_rate": 0.00015731393735125907,
+      "loss": 0.7008797645568847,
+      "mean_token_accuracy": 0.8007259473204613,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.752259682068664,
+      "eval_loss": 0.7171286344528198,
+      "eval_mean_token_accuracy": 0.80052122574174,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 90.0797,
+      "eval_samples_per_second": 15.764,
+      "eval_steps_per_second": 1.976,
+      "step": 180
+    },
+    {
+      "entropy": 0.7828126326203346,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9553632736206055,
+      "learning_rate": 0.0001748909135916232,
+      "loss": 0.7046853542327881,
+      "mean_token_accuracy": 0.7988562889397144,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7389638169427936,
+      "eval_loss": 0.7187935709953308,
+      "eval_mean_token_accuracy": 0.7995899999409579,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 89.9223,
+      "eval_samples_per_second": 15.791,
+      "eval_steps_per_second": 1.979,
+      "step": 200
+    },
+    {
+      "entropy": 0.7909286297857762,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 1.134459376335144,
+      "learning_rate": 0.00019246788983198735,
+      "loss": 0.6989312171936035,
+      "mean_token_accuracy": 0.7991565138101577,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7496965199374082,
+      "eval_loss": 0.7106321454048157,
+      "eval_mean_token_accuracy": 0.8046183442131857,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 89.6899,
+      "eval_samples_per_second": 15.832,
+      "eval_steps_per_second": 1.985,
+      "step": 220
+    },
+    {
+      "entropy": 0.7651963323354721,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.9557585716247559,
+      "learning_rate": 0.0002100448660723515,
+      "loss": 0.6984559059143066,
+      "mean_token_accuracy": 0.8019823372364044,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7615563079212488,
+      "eval_loss": 0.7098539471626282,
+      "eval_mean_token_accuracy": 0.8047233508544022,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 90.3752,
+      "eval_samples_per_second": 15.712,
+      "eval_steps_per_second": 1.97,
+      "step": 240
+    },
+    {
+      "entropy": 0.787507726252079,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.9279311299324036,
+      "learning_rate": 0.00022762184231271568,
+      "loss": 0.708138370513916,
+      "mean_token_accuracy": 0.8013308703899383,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7856640896100676,
+      "eval_loss": 0.7191570401191711,
+      "eval_mean_token_accuracy": 0.8024677249153008,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 93.7336,
+      "eval_samples_per_second": 15.149,
+      "eval_steps_per_second": 1.899,
+      "step": 260
+    },
+    {
+      "entropy": 0.768489234894514,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.7992776036262512,
+      "learning_rate": 0.0002451988185530798,
+      "loss": 0.7035849571228028,
+      "mean_token_accuracy": 0.8031105332076549,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7682498740346244,
+      "eval_loss": 0.7149022221565247,
+      "eval_mean_token_accuracy": 0.7973497995499814,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 89.7309,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 280
+    },
+    {
+      "entropy": 0.7720955617725849,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.1931514739990234,
+      "learning_rate": 0.0002627757947934439,
+      "loss": 0.6965402603149414,
+      "mean_token_accuracy": 0.8040348663926125,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7214708304807042,
+      "eval_loss": 0.7350823879241943,
+      "eval_mean_token_accuracy": 0.7942911925610532,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 89.6451,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 300
+    },
+    {
+      "entropy": 0.7930095426738262,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8994467258453369,
+      "learning_rate": 0.0002803527710338081,
+      "loss": 0.7331492900848389,
+      "mean_token_accuracy": 0.796185302734375,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.8142535488926963,
+      "eval_loss": 0.7237834334373474,
+      "eval_mean_token_accuracy": 0.799495273761535,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 90.6759,
+      "eval_samples_per_second": 15.66,
+      "eval_steps_per_second": 1.963,
+      "step": 320
+    },
+    {
+      "entropy": 0.7825102441012859,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 1.8224239349365234,
+      "learning_rate": 0.00029792974727417223,
+      "loss": 0.7331175804138184,
+      "mean_token_accuracy": 0.7977667152881622,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7327135443017724,
+      "eval_loss": 0.7388784289360046,
+      "eval_mean_token_accuracy": 0.8007041252730938,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 89.7692,
+      "eval_samples_per_second": 15.818,
+      "eval_steps_per_second": 1.983,
+      "step": 340
+    },
+    {
+      "entropy": 0.819331557303667,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 1.0407779216766357,
+      "learning_rate": 0.0003155067235145364,
+      "loss": 0.7456695556640625,
+      "mean_token_accuracy": 0.7936980701982975,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.8065585420372781,
+      "eval_loss": 0.735519528388977,
+      "eval_mean_token_accuracy": 0.7974206296245704,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 89.4336,
+      "eval_samples_per_second": 15.878,
+      "eval_steps_per_second": 1.99,
+      "step": 360
+    },
+    {
+      "entropy": 0.8114653021097183,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.1209490299224854,
+      "learning_rate": 0.0003330836997549005,
+      "loss": 0.7340899467468261,
+      "mean_token_accuracy": 0.7914272703230381,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.8025527285056168,
+      "eval_loss": 0.7295576930046082,
+      "eval_mean_token_accuracy": 0.7990576628218876,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 89.9579,
+      "eval_samples_per_second": 15.785,
+      "eval_steps_per_second": 1.979,
+      "step": 380
+    },
+    {
+      "entropy": 0.8042497783899307,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 1.2437798976898193,
+      "learning_rate": 0.0003506606759952647,
+      "loss": 0.7381744384765625,
+      "mean_token_accuracy": 0.7967362694442273,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7717011336530193,
+      "eval_loss": 0.7436173558235168,
+      "eval_mean_token_accuracy": 0.7975849482450592,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 89.8411,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 400
+    },
+    {
+      "entropy": 0.7987750010056929,
+      "epoch": 1.012070006035003,
+      "grad_norm": 1.2390918731689453,
+      "learning_rate": 0.000364721224843344,
+      "loss": 0.7317790031433106,
+      "mean_token_accuracy": 0.7986087551364651,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7333170604170038,
+      "eval_loss": 0.7456948757171631,
+      "eval_mean_token_accuracy": 0.7964897577682238,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 89.7288,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 420
+    },
+    {
+      "entropy": 0.7372976846992969,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.1552740335464478,
+      "learning_rate": 0.00036468510102269575,
+      "loss": 0.6734249114990234,
+      "mean_token_accuracy": 0.805834824591875,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.7130355526891987,
+      "eval_loss": 0.7555333375930786,
+      "eval_mean_token_accuracy": 0.7961652781186479,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 89.8713,
+      "eval_samples_per_second": 15.8,
+      "eval_steps_per_second": 1.981,
+      "step": 440
+    },
+    {
+      "entropy": 0.7465668715536594,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 1.1991469860076904,
+      "learning_rate": 0.0003645973816745066,
+      "loss": 0.6871879577636719,
+      "mean_token_accuracy": 0.8042450234293937,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7772979542110743,
+      "eval_loss": 0.7561826705932617,
+      "eval_mean_token_accuracy": 0.7953876427720102,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 89.8395,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 460
+    },
+    {
+      "entropy": 0.7477855801582336,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 1.325486660003662,
+      "learning_rate": 0.00036445809162231435,
+      "loss": 0.6836830139160156,
+      "mean_token_accuracy": 0.8021619468927383,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.7098110676481483,
+      "eval_loss": 0.758007287979126,
+      "eval_mean_token_accuracy": 0.7909953189030122,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 89.6101,
+      "eval_samples_per_second": 15.846,
+      "eval_steps_per_second": 1.986,
+      "step": 480
+    },
+    {
+      "entropy": 0.7632691070437432,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 1.0122262239456177,
+      "learning_rate": 0.0003642672702835562,
+      "loss": 0.7071213245391845,
+      "mean_token_accuracy": 0.8000680930912495,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7645382720432924,
+      "eval_loss": 0.7492606043815613,
+      "eval_mean_token_accuracy": 0.7976858676149604,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 89.6092,
+      "eval_samples_per_second": 15.847,
+      "eval_steps_per_second": 1.986,
+      "step": 500
+    },
+    {
+      "entropy": 0.7623726457357407,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.367774486541748,
+      "learning_rate": 0.00036402497165841384,
+      "loss": 0.6965175151824952,
+      "mean_token_accuracy": 0.8016501650214195,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7868083715438843,
+      "eval_loss": 0.7481449842453003,
+      "eval_mean_token_accuracy": 0.799330582779445,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 89.8241,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 520
+    },
+    {
+      "entropy": 0.7535207405686378,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 1.213614821434021,
+      "learning_rate": 0.00036373126431453207,
+      "loss": 0.7078310489654541,
+      "mean_token_accuracy": 0.7998907208442688,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7829091950748743,
+      "eval_loss": 0.7416072487831116,
+      "eval_mean_token_accuracy": 0.7969146250339036,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 89.4607,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 540
+    },
+    {
+      "entropy": 0.7527363449335098,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 1.0929982662200928,
+      "learning_rate": 0.00036338623136761495,
+      "loss": 0.7058369636535644,
+      "mean_token_accuracy": 0.7997087433934211,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.728419938114252,
+      "eval_loss": 0.744873046875,
+      "eval_mean_token_accuracy": 0.7966634296299366,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 89.6597,
+      "eval_samples_per_second": 15.838,
+      "eval_steps_per_second": 1.985,
+      "step": 560
+    },
+    {
+      "entropy": 0.7406099259853363,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.409364104270935,
+      "learning_rate": 0.00036298997045790515,
+      "loss": 0.6902852058410645,
+      "mean_token_accuracy": 0.805386807769537,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.6991632758231645,
+      "eval_loss": 0.7480319738388062,
+      "eval_mean_token_accuracy": 0.7970068006033308,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 89.8329,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 580
+    },
+    {
+      "entropy": 0.7459449704736472,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.7053111791610718,
+      "learning_rate": 0.00036254259372255277,
+      "loss": 0.7044640064239502,
+      "mean_token_accuracy": 0.8005018755793571,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7454566660891758,
+      "eval_loss": 0.7551484704017639,
+      "eval_mean_token_accuracy": 0.7897093731365846,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 89.8313,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 600
+    },
+    {
+      "entropy": 0.7621391348540782,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 1.174443244934082,
+      "learning_rate": 0.000362044227763882,
+      "loss": 0.7111573696136475,
+      "mean_token_accuracy": 0.798752411454916,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7664387949397055,
+      "eval_loss": 0.7365804314613342,
+      "eval_mean_token_accuracy": 0.7963101200843126,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 89.9623,
+      "eval_samples_per_second": 15.784,
+      "eval_steps_per_second": 1.979,
+      "step": 620
+    },
+    {
+      "entropy": 0.7424944408237935,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.2898184061050415,
+      "learning_rate": 0.000361495013613564,
+      "loss": 0.6913110733032226,
+      "mean_token_accuracy": 0.8009015507996082,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7255479529332579,
+      "eval_loss": 0.7345991730690002,
+      "eval_mean_token_accuracy": 0.8013592945056015,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 89.817,
+      "eval_samples_per_second": 15.81,
+      "eval_steps_per_second": 1.982,
+      "step": 640
+    },
+    {
+      "entropy": 0.7303367637097835,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.1310476064682007,
+      "learning_rate": 0.00036089510669270666,
+      "loss": 0.6875617027282714,
+      "mean_token_accuracy": 0.8035397171974182,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7101308248016271,
+      "eval_loss": 0.7394917011260986,
+      "eval_mean_token_accuracy": 0.7990792942850777,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 89.7431,
+      "eval_samples_per_second": 15.823,
+      "eval_steps_per_second": 1.983,
+      "step": 660
+    },
+    {
+      "entropy": 0.7387678384780884,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.9645389914512634,
+      "learning_rate": 0.0003602446767678725,
+      "loss": 0.6911204338073731,
+      "mean_token_accuracy": 0.8056333385407924,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7281073737010527,
+      "eval_loss": 0.7376708984375,
+      "eval_mean_token_accuracy": 0.7997356415464637,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 90.0338,
+      "eval_samples_per_second": 15.772,
+      "eval_steps_per_second": 1.977,
+      "step": 680
+    },
+    {
+      "entropy": 0.7449554048478604,
+      "epoch": 1.687990343995172,
+      "grad_norm": 1.9069629907608032,
+      "learning_rate": 0.0003595439079030364,
+      "loss": 0.7042286396026611,
+      "mean_token_accuracy": 0.8018639378249646,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7022633817088738,
+      "eval_loss": 0.7320939898490906,
+      "eval_mean_token_accuracy": 0.8006051541044471,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 89.8948,
+      "eval_samples_per_second": 15.796,
+      "eval_steps_per_second": 1.98,
+      "step": 700
+    },
+    {
+      "entropy": 0.723249051719904,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.8333877325057983,
+      "learning_rate": 0.00035879299840749777,
+      "loss": 0.7043714046478271,
+      "mean_token_accuracy": 0.8028345607221127,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.672436946898364,
+      "eval_loss": 0.713233470916748,
+      "eval_mean_token_accuracy": 0.8047760577684038,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 89.6394,
+      "eval_samples_per_second": 15.841,
+      "eval_steps_per_second": 1.986,
+      "step": 720
+    },
+    {
+      "entropy": 0.730014206469059,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.4580037593841553,
+      "learning_rate": 0.00035799216077976137,
+      "loss": 0.6985176563262939,
+      "mean_token_accuracy": 0.8015547141432762,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.6908873795123582,
+      "eval_loss": 0.7198202013969421,
+      "eval_mean_token_accuracy": 0.8024131682481659,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 89.3771,
+      "eval_samples_per_second": 15.888,
+      "eval_steps_per_second": 1.992,
+      "step": 740
+    },
+    {
+      "entropy": 0.7219605796039105,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.51682710647583,
+      "learning_rate": 0.000357141621647403,
+      "loss": 0.6844424724578857,
+      "mean_token_accuracy": 0.8032813094556331,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7113401109582922,
+      "eval_loss": 0.7195309400558472,
+      "eval_mean_token_accuracy": 0.8019187035185568,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 89.4658,
+      "eval_samples_per_second": 15.872,
+      "eval_steps_per_second": 1.99,
+      "step": 760
+    },
+    {
+      "entropy": 0.729826743900776,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 1.0838000774383545,
+      "learning_rate": 0.0003562416217029361,
+      "loss": 0.685858678817749,
+      "mean_token_accuracy": 0.8050567395985126,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.649370267484965,
+      "eval_loss": 0.7177029252052307,
+      "eval_mean_token_accuracy": 0.8050174043419656,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 89.2287,
+      "eval_samples_per_second": 15.914,
+      "eval_steps_per_second": 1.995,
+      "step": 780
+    },
+    {
+      "entropy": 0.6799364153295755,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 1.1364924907684326,
+      "learning_rate": 0.00035529241563569903,
+      "loss": 0.6714428901672364,
+      "mean_token_accuracy": 0.8087341658771038,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6603799047095052,
+      "eval_loss": 0.7201809287071228,
+      "eval_mean_token_accuracy": 0.802863868769635,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 89.2718,
+      "eval_samples_per_second": 15.906,
+      "eval_steps_per_second": 1.994,
+      "step": 800
+    },
+    {
+      "entropy": 0.7078610375523567,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 1.2191942930221558,
+      "learning_rate": 0.0003542942720597808,
+      "loss": 0.6996590614318847,
+      "mean_token_accuracy": 0.8049512051045895,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.7240545264120852,
+      "eval_loss": 0.7301309108734131,
+      "eval_mean_token_accuracy": 0.8015973166133581,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 89.3438,
+      "eval_samples_per_second": 15.894,
+      "eval_steps_per_second": 1.992,
+      "step": 820
+    },
+    {
+      "entropy": 0.6557391556826505,
+      "epoch": 2.024140012070006,
+      "grad_norm": 1.36203134059906,
+      "learning_rate": 0.0003532474734380064,
+      "loss": 0.6327113628387451,
+      "mean_token_accuracy": 0.8142829847026181,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6573808648613062,
+      "eval_loss": 0.7334365844726562,
+      "eval_mean_token_accuracy": 0.8043633893634496,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 89.3359,
+      "eval_samples_per_second": 15.895,
+      "eval_steps_per_second": 1.992,
+      "step": 840
+    },
+    {
+      "entropy": 0.6409128502011299,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 1.5804955959320068,
+      "learning_rate": 0.0003521523160020035,
+      "loss": 0.6023559093475341,
+      "mean_token_accuracy": 0.8222251623868942,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6983234788594621,
+      "eval_loss": 0.7232135534286499,
+      "eval_mean_token_accuracy": 0.8010654268639811,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 89.2615,
+      "eval_samples_per_second": 15.908,
+      "eval_steps_per_second": 1.994,
+      "step": 860
+    },
+    {
+      "entropy": 0.6151149723678827,
+      "epoch": 2.12070006035003,
+      "grad_norm": 1.1975436210632324,
+      "learning_rate": 0.00035100910966837193,
+      "loss": 0.5803564548492431,
+      "mean_token_accuracy": 0.8256325736641884,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6636555573243773,
+      "eval_loss": 0.7205662727355957,
+      "eval_mean_token_accuracy": 0.8036768369460374,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 89.2957,
+      "eval_samples_per_second": 15.902,
+      "eval_steps_per_second": 1.993,
+      "step": 880
+    },
+    {
+      "entropy": 0.6201727617532015,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9907131195068359,
+      "learning_rate": 0.0003498181779509813,
+      "loss": 0.6101872444152832,
+      "mean_token_accuracy": 0.8196512959897518,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.6473788086617931,
+      "eval_loss": 0.7062397003173828,
+      "eval_mean_token_accuracy": 0.8078362184963869,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 88.6198,
+      "eval_samples_per_second": 16.024,
+      "eval_steps_per_second": 2.009,
+      "step": 900
+    },
+    {
+      "entropy": 0.6097332935780286,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.9055633544921875,
+      "learning_rate": 0.00034857985786942026,
+      "loss": 0.5956094264984131,
+      "mean_token_accuracy": 0.8240575887262821,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.612629868341296,
+      "eval_loss": 0.7106770277023315,
+      "eval_mean_token_accuracy": 0.805324277181304,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 89.4383,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 920
+    },
+    {
+      "entropy": 0.6182701248675585,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.9513736367225647,
+      "learning_rate": 0.00034729449985362404,
+      "loss": 0.6019269466400147,
+      "mean_token_accuracy": 0.8187106013298034,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6437647456533453,
+      "eval_loss": 0.7173502445220947,
+      "eval_mean_token_accuracy": 0.8056439573175451,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.9568,
+      "eval_samples_per_second": 15.963,
+      "eval_steps_per_second": 2.001,
+      "step": 940
+    },
+    {
+      "entropy": 0.6508570514619351,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.9606207013130188,
+      "learning_rate": 0.0003459624676447067,
+      "loss": 0.6296409606933594,
+      "mean_token_accuracy": 0.8167732007801533,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6288247135248077,
+      "eval_loss": 0.7160675525665283,
+      "eval_mean_token_accuracy": 0.8066010689467527,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 89.0044,
+      "eval_samples_per_second": 15.954,
+      "eval_steps_per_second": 2.0,
+      "step": 960
+    },
+    {
+      "entropy": 0.6380081418901682,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.9677181839942932,
+      "learning_rate": 0.000344584138192027,
+      "loss": 0.6210546493530273,
+      "mean_token_accuracy": 0.8186135642230511,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6407805031604981,
+      "eval_loss": 0.7055043578147888,
+      "eval_mean_token_accuracy": 0.8070103157772107,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 89.5606,
+      "eval_samples_per_second": 15.855,
+      "eval_steps_per_second": 1.987,
+      "step": 980
+    },
+    {
+      "entropy": 0.6310251638293266,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 2.1046407222747803,
+      "learning_rate": 0.000343159901546516,
+      "loss": 0.6165830135345459,
+      "mean_token_accuracy": 0.8207855716347694,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6461507470420237,
+      "eval_loss": 0.7292136549949646,
+      "eval_mean_token_accuracy": 0.8030418261383356,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 89.315,
+      "eval_samples_per_second": 15.899,
+      "eval_steps_per_second": 1.993,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6256617344915867,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 1.1768349409103394,
+      "learning_rate": 0.0003416901607502972,
+      "loss": 0.6142709255218506,
+      "mean_token_accuracy": 0.8220669947564602,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6150523993406403,
+      "eval_loss": 0.7019563317298889,
+      "eval_mean_token_accuracy": 0.8094323099998946,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 89.419,
+      "eval_samples_per_second": 15.88,
+      "eval_steps_per_second": 1.991,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6126735735684633,
+      "epoch": 2.506940253470127,
+      "grad_norm": 1.8065496683120728,
+      "learning_rate": 0.00034017533172263055,
+      "loss": 0.6056458473205566,
+      "mean_token_accuracy": 0.8203762136399746,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6544345456562685,
+      "eval_loss": 0.7086517214775085,
+      "eval_mean_token_accuracy": 0.806709805901131,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 88.981,
+      "eval_samples_per_second": 15.958,
+      "eval_steps_per_second": 2.0,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6311754353344441,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 1.0576369762420654,
+      "learning_rate": 0.00033861584314221236,
+      "loss": 0.6176392555236816,
+      "mean_token_accuracy": 0.8177410490810871,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6273799200406235,
+      "eval_loss": 0.708163321018219,
+      "eval_mean_token_accuracy": 0.8032017099053672,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 89.7668,
+      "eval_samples_per_second": 15.819,
+      "eval_steps_per_second": 1.983,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6200249589979648,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.997065007686615,
+      "learning_rate": 0.0003370121363258637,
+      "loss": 0.6203888893127442,
+      "mean_token_accuracy": 0.8196275025606156,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6457576811983344,
+      "eval_loss": 0.6972290277481079,
+      "eval_mean_token_accuracy": 0.8095980849158898,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 89.2027,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 1080
+    },
+    {
+      "entropy": 0.626810473203659,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.865475058555603,
+      "learning_rate": 0.0003353646651036438,
+      "loss": 0.6102585792541504,
+      "mean_token_accuracy": 0.819708751142025,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.6412022431914726,
+      "eval_loss": 0.7036139965057373,
+      "eval_mean_token_accuracy": 0.8070317135098275,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 89.4733,
+      "eval_samples_per_second": 15.871,
+      "eval_steps_per_second": 1.989,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6581619590520859,
+      "epoch": 2.700060350030175,
+      "grad_norm": 1.00307035446167,
+      "learning_rate": 0.00033367389569042064,
+      "loss": 0.6397720336914062,
+      "mean_token_accuracy": 0.81413309648633,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6288736402318719,
+      "eval_loss": 0.6966825723648071,
+      "eval_mean_token_accuracy": 0.8094039669867312,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 88.7341,
+      "eval_samples_per_second": 16.003,
+      "eval_steps_per_second": 2.006,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6399412982165813,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.8589329719543457,
+      "learning_rate": 0.0003319403065539384,
+      "loss": 0.6347338199615479,
+      "mean_token_accuracy": 0.8139534957706929,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.683491913455256,
+      "eval_loss": 0.6984149217605591,
+      "eval_mean_token_accuracy": 0.8083989469522841,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 89.3929,
+      "eval_samples_per_second": 15.885,
+      "eval_steps_per_second": 1.991,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6264835461974144,
+      "epoch": 2.796620398310199,
+      "grad_norm": 0.7807853817939758,
+      "learning_rate": 0.0003301643882794163,
+      "loss": 0.6162629127502441,
+      "mean_token_accuracy": 0.8216292977333068,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6578792159476977,
+      "eval_loss": 0.6921441555023193,
+      "eval_mean_token_accuracy": 0.8094299177775223,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 89.2356,
+      "eval_samples_per_second": 15.913,
+      "eval_steps_per_second": 1.995,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6269129011780024,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 1.1675593852996826,
+      "learning_rate": 0.0003283466434307189,
+      "loss": 0.6235532760620117,
+      "mean_token_accuracy": 0.8188897788524627,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6533248447970058,
+      "eval_loss": 0.696849524974823,
+      "eval_mean_token_accuracy": 0.8102703713968898,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2602,
+      "eval_samples_per_second": 16.089,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.6259873781353236,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.9596486687660217,
+      "learning_rate": 0.00032648758640813655,
+      "loss": 0.6118728160858155,
+      "mean_token_accuracy": 0.8171859130263328,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6212078754821521,
+      "eval_loss": 0.6942870020866394,
+      "eval_mean_token_accuracy": 0.810484343365337,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 89.4375,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 1200
+    },
+    {
+      "entropy": 0.6142537571489811,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 0.9625058770179749,
+      "learning_rate": 0.00032458774330281615,
+      "loss": 0.6060261249542236,
+      "mean_token_accuracy": 0.8223113484680653,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6553665667437436,
+      "eval_loss": 0.6835269331932068,
+      "eval_mean_token_accuracy": 0.8134723968720168,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 89.5092,
+      "eval_samples_per_second": 15.864,
+      "eval_steps_per_second": 1.989,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6301972426474094,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.9657333493232727,
+      "learning_rate": 0.0003226476517478835,
+      "loss": 0.6277759075164795,
+      "mean_token_accuracy": 0.8164266042411328,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6594253118788258,
+      "eval_loss": 0.6820935606956482,
+      "eval_mean_token_accuracy": 0.8132368198941263,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 89.6211,
+      "eval_samples_per_second": 15.844,
+      "eval_steps_per_second": 1.986,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5346100655469027,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8541185259819031,
+      "learning_rate": 0.0003206678607662996,
+      "loss": 0.5261692047119141,
+      "mean_token_accuracy": 0.8382289425119177,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5626645662476507,
+      "eval_loss": 0.7028542160987854,
+      "eval_mean_token_accuracy": 0.8133395572056931,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 89.1267,
+      "eval_samples_per_second": 15.932,
+      "eval_steps_per_second": 1.997,
+      "step": 1260
+    },
+    {
+      "entropy": 0.5001101832836866,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.8208815455436707,
+      "learning_rate": 0.0003186489306154935,
+      "loss": 0.4850132942199707,
+      "mean_token_accuracy": 0.8468983843922615,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5889462046743779,
+      "eval_loss": 0.691015899181366,
+      "eval_mean_token_accuracy": 0.8135226466012805,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.9677,
+      "eval_samples_per_second": 15.961,
+      "eval_steps_per_second": 2.001,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5242766339331866,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 0.7893072962760925,
+      "learning_rate": 0.0003165914326288163,
+      "loss": 0.5061577320098877,
+      "mean_token_accuracy": 0.8425608821213245,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5543691962957382,
+      "eval_loss": 0.6913734078407288,
+      "eval_mean_token_accuracy": 0.8154889680026622,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 89.6963,
+      "eval_samples_per_second": 15.831,
+      "eval_steps_per_second": 1.984,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5327721010893584,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.8992080092430115,
+      "learning_rate": 0.0003144959490538604,
+      "loss": 0.5068036556243897,
+      "mean_token_accuracy": 0.8420542575418949,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5734540803378887,
+      "eval_loss": 0.6948716044425964,
+      "eval_mean_token_accuracy": 0.8137217944257715,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 89.4597,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5247377116233111,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.9291845560073853,
+      "learning_rate": 0.0003123630728876902,
+      "loss": 0.5023272037506104,
+      "mean_token_accuracy": 0.8435182586312294,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5492362935891312,
+      "eval_loss": 0.6980633735656738,
+      "eval_mean_token_accuracy": 0.8155082919624415,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 89.2124,
+      "eval_samples_per_second": 15.917,
+      "eval_steps_per_second": 1.995,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5380499072372913,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 1.0477524995803833,
+      "learning_rate": 0.00031019340770903136,
+      "loss": 0.5264591693878173,
+      "mean_token_accuracy": 0.8391959741711617,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5896503307511297,
+      "eval_loss": 0.6997144222259521,
+      "eval_mean_token_accuracy": 0.8124373062942805,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 89.1147,
+      "eval_samples_per_second": 15.935,
+      "eval_steps_per_second": 1.997,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5460653610527515,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 1.2473069429397583,
+      "learning_rate": 0.00030798756750746476,
+      "loss": 0.5319677829742432,
+      "mean_token_accuracy": 0.8364320226013661,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5659052182114526,
+      "eval_loss": 0.6924049258232117,
+      "eval_mean_token_accuracy": 0.8153053585732921,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 89.6468,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5367407951503992,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 1.2679712772369385,
+      "learning_rate": 0.00030574617650967485,
+      "loss": 0.5190927028656006,
+      "mean_token_accuracy": 0.8399469174444676,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5581759144081159,
+      "eval_loss": 0.6933317184448242,
+      "eval_mean_token_accuracy": 0.8139757862251796,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 89.0749,
+      "eval_samples_per_second": 15.942,
+      "eval_steps_per_second": 1.998,
+      "step": 1400
+    },
+    {
+      "entropy": 0.530736118927598,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.9717612862586975,
+      "learning_rate": 0.0003034698690028009,
+      "loss": 0.512363052368164,
+      "mean_token_accuracy": 0.8419127956032753,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5233335836549823,
+      "eval_loss": 0.6958550214767456,
+      "eval_mean_token_accuracy": 0.8158778501360604,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 88.8664,
+      "eval_samples_per_second": 15.979,
+      "eval_steps_per_second": 2.003,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5327631575986743,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7649208903312683,
+      "learning_rate": 0.00030115928915494116,
+      "loss": 0.5202179431915284,
+      "mean_token_accuracy": 0.8384683802723885,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.5960972657364406,
+      "eval_loss": 0.6984783411026001,
+      "eval_mean_token_accuracy": 0.8095930597085631,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 89.5168,
+      "eval_samples_per_second": 15.863,
+      "eval_steps_per_second": 1.988,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5428155034780502,
+      "epoch": 3.51901025950513,
+      "grad_norm": 1.2790874242782593,
+      "learning_rate": 0.00029881509083286116,
+      "loss": 0.524717378616333,
+      "mean_token_accuracy": 0.8346644505858422,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.550318562917495,
+      "eval_loss": 0.6988471746444702,
+      "eval_mean_token_accuracy": 0.8132969897784544,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 89.2598,
+      "eval_samples_per_second": 15.909,
+      "eval_steps_per_second": 1.994,
+      "step": 1460
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.463979070638244e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1480/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1480/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1480/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1480/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1480/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..7ce4b072fcbe5b636867e13937c56fedee3b68b4
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1480/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.02728394274790723,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1480/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1480/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1480/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1480/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1480/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1480/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1480/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1480/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..70d1d09b0ae9c0b9a0c79e64cf64569035fc29b5
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1480/trainer_state.json
@@ -0,0 +1,1588 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 3.567290283645142,
+  "eval_steps": 20,
+  "global_step": 1480,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.024795526266098,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.945375919342041,
+      "learning_rate": 1.6698127428345936e-05,
+      "loss": 1.7825420379638672,
+      "mean_token_accuracy": 0.6315708436071873,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.29921497588747,
+      "eval_loss": 1.169131875038147,
+      "eval_mean_token_accuracy": 0.7185706342204233,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 91.7898,
+      "eval_samples_per_second": 15.47,
+      "eval_steps_per_second": 1.939,
+      "step": 20
+    },
+    {
+      "entropy": 1.0346344470977784,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.7916979789733887,
+      "learning_rate": 3.427510366871008e-05,
+      "loss": 0.9480395317077637,
+      "mean_token_accuracy": 0.7543429024517536,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 0.9673695430326997,
+      "eval_loss": 0.8588430285453796,
+      "eval_mean_token_accuracy": 0.7737233002534073,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 89.8219,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 40
+    },
+    {
+      "entropy": 0.8977566614747048,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7972187995910645,
+      "learning_rate": 5.185207990907422e-05,
+      "loss": 0.8029604911804199,
+      "mean_token_accuracy": 0.7818016350269318,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8436836982041263,
+      "eval_loss": 0.7893033623695374,
+      "eval_mean_token_accuracy": 0.7858914984076211,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 89.9349,
+      "eval_samples_per_second": 15.789,
+      "eval_steps_per_second": 1.979,
+      "step": 60
+    },
+    {
+      "entropy": 0.8650321319699288,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.2128729820251465,
+      "learning_rate": 6.942905614943836e-05,
+      "loss": 0.77417893409729,
+      "mean_token_accuracy": 0.7875479467213153,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8209857180547179,
+      "eval_loss": 0.7572280168533325,
+      "eval_mean_token_accuracy": 0.7925970383574453,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 89.9459,
+      "eval_samples_per_second": 15.787,
+      "eval_steps_per_second": 1.979,
+      "step": 80
+    },
+    {
+      "entropy": 0.8096660938113928,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.1254287958145142,
+      "learning_rate": 8.700603238980251e-05,
+      "loss": 0.7170331001281738,
+      "mean_token_accuracy": 0.7994555257260799,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.792780208788561,
+      "eval_loss": 0.7442984580993652,
+      "eval_mean_token_accuracy": 0.7954704573984896,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 90.294,
+      "eval_samples_per_second": 15.726,
+      "eval_steps_per_second": 1.971,
+      "step": 100
+    },
+    {
+      "entropy": 0.7979829199612141,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2140507698059082,
+      "learning_rate": 0.00010458300863016666,
+      "loss": 0.7141861915588379,
+      "mean_token_accuracy": 0.8000014744699001,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.7938692486018277,
+      "eval_loss": 0.7279797196388245,
+      "eval_mean_token_accuracy": 0.7989798734027348,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 89.7189,
+      "eval_samples_per_second": 15.827,
+      "eval_steps_per_second": 1.984,
+      "step": 120
+    },
+    {
+      "entropy": 0.7916347607970238,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.0506458282470703,
+      "learning_rate": 0.0001221599848705308,
+      "loss": 0.7117055416107178,
+      "mean_token_accuracy": 0.7970046654343605,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.8052781063519167,
+      "eval_loss": 0.722576916217804,
+      "eval_mean_token_accuracy": 0.7982530523551984,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 90.0376,
+      "eval_samples_per_second": 15.771,
+      "eval_steps_per_second": 1.977,
+      "step": 140
+    },
+    {
+      "entropy": 0.7880666613578796,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8409207463264465,
+      "learning_rate": 0.00013973696111089492,
+      "loss": 0.7052523136138916,
+      "mean_token_accuracy": 0.8000123649835587,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7787443134891853,
+      "eval_loss": 0.7137772440910339,
+      "eval_mean_token_accuracy": 0.8014209287220173,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 90.2562,
+      "eval_samples_per_second": 15.733,
+      "eval_steps_per_second": 1.972,
+      "step": 160
+    },
+    {
+      "entropy": 0.7696007996797561,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.0182477235794067,
+      "learning_rate": 0.00015731393735125907,
+      "loss": 0.7008797645568847,
+      "mean_token_accuracy": 0.8007259473204613,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.752259682068664,
+      "eval_loss": 0.7171286344528198,
+      "eval_mean_token_accuracy": 0.80052122574174,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 90.0797,
+      "eval_samples_per_second": 15.764,
+      "eval_steps_per_second": 1.976,
+      "step": 180
+    },
+    {
+      "entropy": 0.7828126326203346,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9553632736206055,
+      "learning_rate": 0.0001748909135916232,
+      "loss": 0.7046853542327881,
+      "mean_token_accuracy": 0.7988562889397144,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7389638169427936,
+      "eval_loss": 0.7187935709953308,
+      "eval_mean_token_accuracy": 0.7995899999409579,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 89.9223,
+      "eval_samples_per_second": 15.791,
+      "eval_steps_per_second": 1.979,
+      "step": 200
+    },
+    {
+      "entropy": 0.7909286297857762,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 1.134459376335144,
+      "learning_rate": 0.00019246788983198735,
+      "loss": 0.6989312171936035,
+      "mean_token_accuracy": 0.7991565138101577,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7496965199374082,
+      "eval_loss": 0.7106321454048157,
+      "eval_mean_token_accuracy": 0.8046183442131857,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 89.6899,
+      "eval_samples_per_second": 15.832,
+      "eval_steps_per_second": 1.985,
+      "step": 220
+    },
+    {
+      "entropy": 0.7651963323354721,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.9557585716247559,
+      "learning_rate": 0.0002100448660723515,
+      "loss": 0.6984559059143066,
+      "mean_token_accuracy": 0.8019823372364044,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7615563079212488,
+      "eval_loss": 0.7098539471626282,
+      "eval_mean_token_accuracy": 0.8047233508544022,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 90.3752,
+      "eval_samples_per_second": 15.712,
+      "eval_steps_per_second": 1.97,
+      "step": 240
+    },
+    {
+      "entropy": 0.787507726252079,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.9279311299324036,
+      "learning_rate": 0.00022762184231271568,
+      "loss": 0.708138370513916,
+      "mean_token_accuracy": 0.8013308703899383,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7856640896100676,
+      "eval_loss": 0.7191570401191711,
+      "eval_mean_token_accuracy": 0.8024677249153008,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 93.7336,
+      "eval_samples_per_second": 15.149,
+      "eval_steps_per_second": 1.899,
+      "step": 260
+    },
+    {
+      "entropy": 0.768489234894514,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.7992776036262512,
+      "learning_rate": 0.0002451988185530798,
+      "loss": 0.7035849571228028,
+      "mean_token_accuracy": 0.8031105332076549,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7682498740346244,
+      "eval_loss": 0.7149022221565247,
+      "eval_mean_token_accuracy": 0.7973497995499814,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 89.7309,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 280
+    },
+    {
+      "entropy": 0.7720955617725849,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.1931514739990234,
+      "learning_rate": 0.0002627757947934439,
+      "loss": 0.6965402603149414,
+      "mean_token_accuracy": 0.8040348663926125,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7214708304807042,
+      "eval_loss": 0.7350823879241943,
+      "eval_mean_token_accuracy": 0.7942911925610532,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 89.6451,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 300
+    },
+    {
+      "entropy": 0.7930095426738262,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8994467258453369,
+      "learning_rate": 0.0002803527710338081,
+      "loss": 0.7331492900848389,
+      "mean_token_accuracy": 0.796185302734375,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.8142535488926963,
+      "eval_loss": 0.7237834334373474,
+      "eval_mean_token_accuracy": 0.799495273761535,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 90.6759,
+      "eval_samples_per_second": 15.66,
+      "eval_steps_per_second": 1.963,
+      "step": 320
+    },
+    {
+      "entropy": 0.7825102441012859,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 1.8224239349365234,
+      "learning_rate": 0.00029792974727417223,
+      "loss": 0.7331175804138184,
+      "mean_token_accuracy": 0.7977667152881622,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7327135443017724,
+      "eval_loss": 0.7388784289360046,
+      "eval_mean_token_accuracy": 0.8007041252730938,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 89.7692,
+      "eval_samples_per_second": 15.818,
+      "eval_steps_per_second": 1.983,
+      "step": 340
+    },
+    {
+      "entropy": 0.819331557303667,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 1.0407779216766357,
+      "learning_rate": 0.0003155067235145364,
+      "loss": 0.7456695556640625,
+      "mean_token_accuracy": 0.7936980701982975,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.8065585420372781,
+      "eval_loss": 0.735519528388977,
+      "eval_mean_token_accuracy": 0.7974206296245704,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 89.4336,
+      "eval_samples_per_second": 15.878,
+      "eval_steps_per_second": 1.99,
+      "step": 360
+    },
+    {
+      "entropy": 0.8114653021097183,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.1209490299224854,
+      "learning_rate": 0.0003330836997549005,
+      "loss": 0.7340899467468261,
+      "mean_token_accuracy": 0.7914272703230381,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.8025527285056168,
+      "eval_loss": 0.7295576930046082,
+      "eval_mean_token_accuracy": 0.7990576628218876,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 89.9579,
+      "eval_samples_per_second": 15.785,
+      "eval_steps_per_second": 1.979,
+      "step": 380
+    },
+    {
+      "entropy": 0.8042497783899307,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 1.2437798976898193,
+      "learning_rate": 0.0003506606759952647,
+      "loss": 0.7381744384765625,
+      "mean_token_accuracy": 0.7967362694442273,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7717011336530193,
+      "eval_loss": 0.7436173558235168,
+      "eval_mean_token_accuracy": 0.7975849482450592,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 89.8411,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 400
+    },
+    {
+      "entropy": 0.7987750010056929,
+      "epoch": 1.012070006035003,
+      "grad_norm": 1.2390918731689453,
+      "learning_rate": 0.000364721224843344,
+      "loss": 0.7317790031433106,
+      "mean_token_accuracy": 0.7986087551364651,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7333170604170038,
+      "eval_loss": 0.7456948757171631,
+      "eval_mean_token_accuracy": 0.7964897577682238,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 89.7288,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 420
+    },
+    {
+      "entropy": 0.7372976846992969,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.1552740335464478,
+      "learning_rate": 0.00036468510102269575,
+      "loss": 0.6734249114990234,
+      "mean_token_accuracy": 0.805834824591875,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.7130355526891987,
+      "eval_loss": 0.7555333375930786,
+      "eval_mean_token_accuracy": 0.7961652781186479,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 89.8713,
+      "eval_samples_per_second": 15.8,
+      "eval_steps_per_second": 1.981,
+      "step": 440
+    },
+    {
+      "entropy": 0.7465668715536594,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 1.1991469860076904,
+      "learning_rate": 0.0003645973816745066,
+      "loss": 0.6871879577636719,
+      "mean_token_accuracy": 0.8042450234293937,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7772979542110743,
+      "eval_loss": 0.7561826705932617,
+      "eval_mean_token_accuracy": 0.7953876427720102,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 89.8395,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 460
+    },
+    {
+      "entropy": 0.7477855801582336,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 1.325486660003662,
+      "learning_rate": 0.00036445809162231435,
+      "loss": 0.6836830139160156,
+      "mean_token_accuracy": 0.8021619468927383,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.7098110676481483,
+      "eval_loss": 0.758007287979126,
+      "eval_mean_token_accuracy": 0.7909953189030122,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 89.6101,
+      "eval_samples_per_second": 15.846,
+      "eval_steps_per_second": 1.986,
+      "step": 480
+    },
+    {
+      "entropy": 0.7632691070437432,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 1.0122262239456177,
+      "learning_rate": 0.0003642672702835562,
+      "loss": 0.7071213245391845,
+      "mean_token_accuracy": 0.8000680930912495,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7645382720432924,
+      "eval_loss": 0.7492606043815613,
+      "eval_mean_token_accuracy": 0.7976858676149604,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 89.6092,
+      "eval_samples_per_second": 15.847,
+      "eval_steps_per_second": 1.986,
+      "step": 500
+    },
+    {
+      "entropy": 0.7623726457357407,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.367774486541748,
+      "learning_rate": 0.00036402497165841384,
+      "loss": 0.6965175151824952,
+      "mean_token_accuracy": 0.8016501650214195,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7868083715438843,
+      "eval_loss": 0.7481449842453003,
+      "eval_mean_token_accuracy": 0.799330582779445,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 89.8241,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 520
+    },
+    {
+      "entropy": 0.7535207405686378,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 1.213614821434021,
+      "learning_rate": 0.00036373126431453207,
+      "loss": 0.7078310489654541,
+      "mean_token_accuracy": 0.7998907208442688,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7829091950748743,
+      "eval_loss": 0.7416072487831116,
+      "eval_mean_token_accuracy": 0.7969146250339036,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 89.4607,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 540
+    },
+    {
+      "entropy": 0.7527363449335098,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 1.0929982662200928,
+      "learning_rate": 0.00036338623136761495,
+      "loss": 0.7058369636535644,
+      "mean_token_accuracy": 0.7997087433934211,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.728419938114252,
+      "eval_loss": 0.744873046875,
+      "eval_mean_token_accuracy": 0.7966634296299366,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 89.6597,
+      "eval_samples_per_second": 15.838,
+      "eval_steps_per_second": 1.985,
+      "step": 560
+    },
+    {
+      "entropy": 0.7406099259853363,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.409364104270935,
+      "learning_rate": 0.00036298997045790515,
+      "loss": 0.6902852058410645,
+      "mean_token_accuracy": 0.805386807769537,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.6991632758231645,
+      "eval_loss": 0.7480319738388062,
+      "eval_mean_token_accuracy": 0.7970068006033308,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 89.8329,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 580
+    },
+    {
+      "entropy": 0.7459449704736472,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.7053111791610718,
+      "learning_rate": 0.00036254259372255277,
+      "loss": 0.7044640064239502,
+      "mean_token_accuracy": 0.8005018755793571,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7454566660891758,
+      "eval_loss": 0.7551484704017639,
+      "eval_mean_token_accuracy": 0.7897093731365846,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 89.8313,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 600
+    },
+    {
+      "entropy": 0.7621391348540782,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 1.174443244934082,
+      "learning_rate": 0.000362044227763882,
+      "loss": 0.7111573696136475,
+      "mean_token_accuracy": 0.798752411454916,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7664387949397055,
+      "eval_loss": 0.7365804314613342,
+      "eval_mean_token_accuracy": 0.7963101200843126,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 89.9623,
+      "eval_samples_per_second": 15.784,
+      "eval_steps_per_second": 1.979,
+      "step": 620
+    },
+    {
+      "entropy": 0.7424944408237935,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.2898184061050415,
+      "learning_rate": 0.000361495013613564,
+      "loss": 0.6913110733032226,
+      "mean_token_accuracy": 0.8009015507996082,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7255479529332579,
+      "eval_loss": 0.7345991730690002,
+      "eval_mean_token_accuracy": 0.8013592945056015,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 89.817,
+      "eval_samples_per_second": 15.81,
+      "eval_steps_per_second": 1.982,
+      "step": 640
+    },
+    {
+      "entropy": 0.7303367637097835,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.1310476064682007,
+      "learning_rate": 0.00036089510669270666,
+      "loss": 0.6875617027282714,
+      "mean_token_accuracy": 0.8035397171974182,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7101308248016271,
+      "eval_loss": 0.7394917011260986,
+      "eval_mean_token_accuracy": 0.7990792942850777,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 89.7431,
+      "eval_samples_per_second": 15.823,
+      "eval_steps_per_second": 1.983,
+      "step": 660
+    },
+    {
+      "entropy": 0.7387678384780884,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.9645389914512634,
+      "learning_rate": 0.0003602446767678725,
+      "loss": 0.6911204338073731,
+      "mean_token_accuracy": 0.8056333385407924,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7281073737010527,
+      "eval_loss": 0.7376708984375,
+      "eval_mean_token_accuracy": 0.7997356415464637,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 90.0338,
+      "eval_samples_per_second": 15.772,
+      "eval_steps_per_second": 1.977,
+      "step": 680
+    },
+    {
+      "entropy": 0.7449554048478604,
+      "epoch": 1.687990343995172,
+      "grad_norm": 1.9069629907608032,
+      "learning_rate": 0.0003595439079030364,
+      "loss": 0.7042286396026611,
+      "mean_token_accuracy": 0.8018639378249646,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7022633817088738,
+      "eval_loss": 0.7320939898490906,
+      "eval_mean_token_accuracy": 0.8006051541044471,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 89.8948,
+      "eval_samples_per_second": 15.796,
+      "eval_steps_per_second": 1.98,
+      "step": 700
+    },
+    {
+      "entropy": 0.723249051719904,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.8333877325057983,
+      "learning_rate": 0.00035879299840749777,
+      "loss": 0.7043714046478271,
+      "mean_token_accuracy": 0.8028345607221127,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.672436946898364,
+      "eval_loss": 0.713233470916748,
+      "eval_mean_token_accuracy": 0.8047760577684038,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 89.6394,
+      "eval_samples_per_second": 15.841,
+      "eval_steps_per_second": 1.986,
+      "step": 720
+    },
+    {
+      "entropy": 0.730014206469059,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.4580037593841553,
+      "learning_rate": 0.00035799216077976137,
+      "loss": 0.6985176563262939,
+      "mean_token_accuracy": 0.8015547141432762,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.6908873795123582,
+      "eval_loss": 0.7198202013969421,
+      "eval_mean_token_accuracy": 0.8024131682481659,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 89.3771,
+      "eval_samples_per_second": 15.888,
+      "eval_steps_per_second": 1.992,
+      "step": 740
+    },
+    {
+      "entropy": 0.7219605796039105,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.51682710647583,
+      "learning_rate": 0.000357141621647403,
+      "loss": 0.6844424724578857,
+      "mean_token_accuracy": 0.8032813094556331,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7113401109582922,
+      "eval_loss": 0.7195309400558472,
+      "eval_mean_token_accuracy": 0.8019187035185568,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 89.4658,
+      "eval_samples_per_second": 15.872,
+      "eval_steps_per_second": 1.99,
+      "step": 760
+    },
+    {
+      "entropy": 0.729826743900776,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 1.0838000774383545,
+      "learning_rate": 0.0003562416217029361,
+      "loss": 0.685858678817749,
+      "mean_token_accuracy": 0.8050567395985126,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.649370267484965,
+      "eval_loss": 0.7177029252052307,
+      "eval_mean_token_accuracy": 0.8050174043419656,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 89.2287,
+      "eval_samples_per_second": 15.914,
+      "eval_steps_per_second": 1.995,
+      "step": 780
+    },
+    {
+      "entropy": 0.6799364153295755,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 1.1364924907684326,
+      "learning_rate": 0.00035529241563569903,
+      "loss": 0.6714428901672364,
+      "mean_token_accuracy": 0.8087341658771038,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6603799047095052,
+      "eval_loss": 0.7201809287071228,
+      "eval_mean_token_accuracy": 0.802863868769635,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 89.2718,
+      "eval_samples_per_second": 15.906,
+      "eval_steps_per_second": 1.994,
+      "step": 800
+    },
+    {
+      "entropy": 0.7078610375523567,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 1.2191942930221558,
+      "learning_rate": 0.0003542942720597808,
+      "loss": 0.6996590614318847,
+      "mean_token_accuracy": 0.8049512051045895,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.7240545264120852,
+      "eval_loss": 0.7301309108734131,
+      "eval_mean_token_accuracy": 0.8015973166133581,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 89.3438,
+      "eval_samples_per_second": 15.894,
+      "eval_steps_per_second": 1.992,
+      "step": 820
+    },
+    {
+      "entropy": 0.6557391556826505,
+      "epoch": 2.024140012070006,
+      "grad_norm": 1.36203134059906,
+      "learning_rate": 0.0003532474734380064,
+      "loss": 0.6327113628387451,
+      "mean_token_accuracy": 0.8142829847026181,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6573808648613062,
+      "eval_loss": 0.7334365844726562,
+      "eval_mean_token_accuracy": 0.8043633893634496,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 89.3359,
+      "eval_samples_per_second": 15.895,
+      "eval_steps_per_second": 1.992,
+      "step": 840
+    },
+    {
+      "entropy": 0.6409128502011299,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 1.5804955959320068,
+      "learning_rate": 0.0003521523160020035,
+      "loss": 0.6023559093475341,
+      "mean_token_accuracy": 0.8222251623868942,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6983234788594621,
+      "eval_loss": 0.7232135534286499,
+      "eval_mean_token_accuracy": 0.8010654268639811,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 89.2615,
+      "eval_samples_per_second": 15.908,
+      "eval_steps_per_second": 1.994,
+      "step": 860
+    },
+    {
+      "entropy": 0.6151149723678827,
+      "epoch": 2.12070006035003,
+      "grad_norm": 1.1975436210632324,
+      "learning_rate": 0.00035100910966837193,
+      "loss": 0.5803564548492431,
+      "mean_token_accuracy": 0.8256325736641884,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6636555573243773,
+      "eval_loss": 0.7205662727355957,
+      "eval_mean_token_accuracy": 0.8036768369460374,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 89.2957,
+      "eval_samples_per_second": 15.902,
+      "eval_steps_per_second": 1.993,
+      "step": 880
+    },
+    {
+      "entropy": 0.6201727617532015,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9907131195068359,
+      "learning_rate": 0.0003498181779509813,
+      "loss": 0.6101872444152832,
+      "mean_token_accuracy": 0.8196512959897518,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.6473788086617931,
+      "eval_loss": 0.7062397003173828,
+      "eval_mean_token_accuracy": 0.8078362184963869,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 88.6198,
+      "eval_samples_per_second": 16.024,
+      "eval_steps_per_second": 2.009,
+      "step": 900
+    },
+    {
+      "entropy": 0.6097332935780286,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.9055633544921875,
+      "learning_rate": 0.00034857985786942026,
+      "loss": 0.5956094264984131,
+      "mean_token_accuracy": 0.8240575887262821,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.612629868341296,
+      "eval_loss": 0.7106770277023315,
+      "eval_mean_token_accuracy": 0.805324277181304,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 89.4383,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 920
+    },
+    {
+      "entropy": 0.6182701248675585,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.9513736367225647,
+      "learning_rate": 0.00034729449985362404,
+      "loss": 0.6019269466400147,
+      "mean_token_accuracy": 0.8187106013298034,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6437647456533453,
+      "eval_loss": 0.7173502445220947,
+      "eval_mean_token_accuracy": 0.8056439573175451,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.9568,
+      "eval_samples_per_second": 15.963,
+      "eval_steps_per_second": 2.001,
+      "step": 940
+    },
+    {
+      "entropy": 0.6508570514619351,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.9606207013130188,
+      "learning_rate": 0.0003459624676447067,
+      "loss": 0.6296409606933594,
+      "mean_token_accuracy": 0.8167732007801533,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6288247135248077,
+      "eval_loss": 0.7160675525665283,
+      "eval_mean_token_accuracy": 0.8066010689467527,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 89.0044,
+      "eval_samples_per_second": 15.954,
+      "eval_steps_per_second": 2.0,
+      "step": 960
+    },
+    {
+      "entropy": 0.6380081418901682,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.9677181839942932,
+      "learning_rate": 0.000344584138192027,
+      "loss": 0.6210546493530273,
+      "mean_token_accuracy": 0.8186135642230511,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6407805031604981,
+      "eval_loss": 0.7055043578147888,
+      "eval_mean_token_accuracy": 0.8070103157772107,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 89.5606,
+      "eval_samples_per_second": 15.855,
+      "eval_steps_per_second": 1.987,
+      "step": 980
+    },
+    {
+      "entropy": 0.6310251638293266,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 2.1046407222747803,
+      "learning_rate": 0.000343159901546516,
+      "loss": 0.6165830135345459,
+      "mean_token_accuracy": 0.8207855716347694,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6461507470420237,
+      "eval_loss": 0.7292136549949646,
+      "eval_mean_token_accuracy": 0.8030418261383356,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 89.315,
+      "eval_samples_per_second": 15.899,
+      "eval_steps_per_second": 1.993,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6256617344915867,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 1.1768349409103394,
+      "learning_rate": 0.0003416901607502972,
+      "loss": 0.6142709255218506,
+      "mean_token_accuracy": 0.8220669947564602,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6150523993406403,
+      "eval_loss": 0.7019563317298889,
+      "eval_mean_token_accuracy": 0.8094323099998946,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 89.419,
+      "eval_samples_per_second": 15.88,
+      "eval_steps_per_second": 1.991,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6126735735684633,
+      "epoch": 2.506940253470127,
+      "grad_norm": 1.8065496683120728,
+      "learning_rate": 0.00034017533172263055,
+      "loss": 0.6056458473205566,
+      "mean_token_accuracy": 0.8203762136399746,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6544345456562685,
+      "eval_loss": 0.7086517214775085,
+      "eval_mean_token_accuracy": 0.806709805901131,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 88.981,
+      "eval_samples_per_second": 15.958,
+      "eval_steps_per_second": 2.0,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6311754353344441,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 1.0576369762420654,
+      "learning_rate": 0.00033861584314221236,
+      "loss": 0.6176392555236816,
+      "mean_token_accuracy": 0.8177410490810871,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6273799200406235,
+      "eval_loss": 0.708163321018219,
+      "eval_mean_token_accuracy": 0.8032017099053672,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 89.7668,
+      "eval_samples_per_second": 15.819,
+      "eval_steps_per_second": 1.983,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6200249589979648,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.997065007686615,
+      "learning_rate": 0.0003370121363258637,
+      "loss": 0.6203888893127442,
+      "mean_token_accuracy": 0.8196275025606156,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6457576811983344,
+      "eval_loss": 0.6972290277481079,
+      "eval_mean_token_accuracy": 0.8095980849158898,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 89.2027,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 1080
+    },
+    {
+      "entropy": 0.626810473203659,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.865475058555603,
+      "learning_rate": 0.0003353646651036438,
+      "loss": 0.6102585792541504,
+      "mean_token_accuracy": 0.819708751142025,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.6412022431914726,
+      "eval_loss": 0.7036139965057373,
+      "eval_mean_token_accuracy": 0.8070317135098275,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 89.4733,
+      "eval_samples_per_second": 15.871,
+      "eval_steps_per_second": 1.989,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6581619590520859,
+      "epoch": 2.700060350030175,
+      "grad_norm": 1.00307035446167,
+      "learning_rate": 0.00033367389569042064,
+      "loss": 0.6397720336914062,
+      "mean_token_accuracy": 0.81413309648633,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6288736402318719,
+      "eval_loss": 0.6966825723648071,
+      "eval_mean_token_accuracy": 0.8094039669867312,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 88.7341,
+      "eval_samples_per_second": 16.003,
+      "eval_steps_per_second": 2.006,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6399412982165813,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.8589329719543457,
+      "learning_rate": 0.0003319403065539384,
+      "loss": 0.6347338199615479,
+      "mean_token_accuracy": 0.8139534957706929,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.683491913455256,
+      "eval_loss": 0.6984149217605591,
+      "eval_mean_token_accuracy": 0.8083989469522841,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 89.3929,
+      "eval_samples_per_second": 15.885,
+      "eval_steps_per_second": 1.991,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6264835461974144,
+      "epoch": 2.796620398310199,
+      "grad_norm": 0.7807853817939758,
+      "learning_rate": 0.0003301643882794163,
+      "loss": 0.6162629127502441,
+      "mean_token_accuracy": 0.8216292977333068,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6578792159476977,
+      "eval_loss": 0.6921441555023193,
+      "eval_mean_token_accuracy": 0.8094299177775223,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 89.2356,
+      "eval_samples_per_second": 15.913,
+      "eval_steps_per_second": 1.995,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6269129011780024,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 1.1675593852996826,
+      "learning_rate": 0.0003283466434307189,
+      "loss": 0.6235532760620117,
+      "mean_token_accuracy": 0.8188897788524627,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6533248447970058,
+      "eval_loss": 0.696849524974823,
+      "eval_mean_token_accuracy": 0.8102703713968898,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2602,
+      "eval_samples_per_second": 16.089,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.6259873781353236,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.9596486687660217,
+      "learning_rate": 0.00032648758640813655,
+      "loss": 0.6118728160858155,
+      "mean_token_accuracy": 0.8171859130263328,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6212078754821521,
+      "eval_loss": 0.6942870020866394,
+      "eval_mean_token_accuracy": 0.810484343365337,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 89.4375,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 1200
+    },
+    {
+      "entropy": 0.6142537571489811,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 0.9625058770179749,
+      "learning_rate": 0.00032458774330281615,
+      "loss": 0.6060261249542236,
+      "mean_token_accuracy": 0.8223113484680653,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6553665667437436,
+      "eval_loss": 0.6835269331932068,
+      "eval_mean_token_accuracy": 0.8134723968720168,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 89.5092,
+      "eval_samples_per_second": 15.864,
+      "eval_steps_per_second": 1.989,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6301972426474094,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.9657333493232727,
+      "learning_rate": 0.0003226476517478835,
+      "loss": 0.6277759075164795,
+      "mean_token_accuracy": 0.8164266042411328,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6594253118788258,
+      "eval_loss": 0.6820935606956482,
+      "eval_mean_token_accuracy": 0.8132368198941263,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 89.6211,
+      "eval_samples_per_second": 15.844,
+      "eval_steps_per_second": 1.986,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5346100655469027,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8541185259819031,
+      "learning_rate": 0.0003206678607662996,
+      "loss": 0.5261692047119141,
+      "mean_token_accuracy": 0.8382289425119177,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5626645662476507,
+      "eval_loss": 0.7028542160987854,
+      "eval_mean_token_accuracy": 0.8133395572056931,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 89.1267,
+      "eval_samples_per_second": 15.932,
+      "eval_steps_per_second": 1.997,
+      "step": 1260
+    },
+    {
+      "entropy": 0.5001101832836866,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.8208815455436707,
+      "learning_rate": 0.0003186489306154935,
+      "loss": 0.4850132942199707,
+      "mean_token_accuracy": 0.8468983843922615,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5889462046743779,
+      "eval_loss": 0.691015899181366,
+      "eval_mean_token_accuracy": 0.8135226466012805,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.9677,
+      "eval_samples_per_second": 15.961,
+      "eval_steps_per_second": 2.001,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5242766339331866,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 0.7893072962760925,
+      "learning_rate": 0.0003165914326288163,
+      "loss": 0.5061577320098877,
+      "mean_token_accuracy": 0.8425608821213245,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5543691962957382,
+      "eval_loss": 0.6913734078407288,
+      "eval_mean_token_accuracy": 0.8154889680026622,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 89.6963,
+      "eval_samples_per_second": 15.831,
+      "eval_steps_per_second": 1.984,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5327721010893584,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.8992080092430115,
+      "learning_rate": 0.0003144959490538604,
+      "loss": 0.5068036556243897,
+      "mean_token_accuracy": 0.8420542575418949,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5734540803378887,
+      "eval_loss": 0.6948716044425964,
+      "eval_mean_token_accuracy": 0.8137217944257715,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 89.4597,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5247377116233111,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.9291845560073853,
+      "learning_rate": 0.0003123630728876902,
+      "loss": 0.5023272037506104,
+      "mean_token_accuracy": 0.8435182586312294,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5492362935891312,
+      "eval_loss": 0.6980633735656738,
+      "eval_mean_token_accuracy": 0.8155082919624415,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 89.2124,
+      "eval_samples_per_second": 15.917,
+      "eval_steps_per_second": 1.995,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5380499072372913,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 1.0477524995803833,
+      "learning_rate": 0.00031019340770903136,
+      "loss": 0.5264591693878173,
+      "mean_token_accuracy": 0.8391959741711617,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5896503307511297,
+      "eval_loss": 0.6997144222259521,
+      "eval_mean_token_accuracy": 0.8124373062942805,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 89.1147,
+      "eval_samples_per_second": 15.935,
+      "eval_steps_per_second": 1.997,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5460653610527515,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 1.2473069429397583,
+      "learning_rate": 0.00030798756750746476,
+      "loss": 0.5319677829742432,
+      "mean_token_accuracy": 0.8364320226013661,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5659052182114526,
+      "eval_loss": 0.6924049258232117,
+      "eval_mean_token_accuracy": 0.8153053585732921,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 89.6468,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5367407951503992,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 1.2679712772369385,
+      "learning_rate": 0.00030574617650967485,
+      "loss": 0.5190927028656006,
+      "mean_token_accuracy": 0.8399469174444676,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5581759144081159,
+      "eval_loss": 0.6933317184448242,
+      "eval_mean_token_accuracy": 0.8139757862251796,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 89.0749,
+      "eval_samples_per_second": 15.942,
+      "eval_steps_per_second": 1.998,
+      "step": 1400
+    },
+    {
+      "entropy": 0.530736118927598,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.9717612862586975,
+      "learning_rate": 0.0003034698690028009,
+      "loss": 0.512363052368164,
+      "mean_token_accuracy": 0.8419127956032753,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5233335836549823,
+      "eval_loss": 0.6958550214767456,
+      "eval_mean_token_accuracy": 0.8158778501360604,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 88.8664,
+      "eval_samples_per_second": 15.979,
+      "eval_steps_per_second": 2.003,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5327631575986743,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7649208903312683,
+      "learning_rate": 0.00030115928915494116,
+      "loss": 0.5202179431915284,
+      "mean_token_accuracy": 0.8384683802723885,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.5960972657364406,
+      "eval_loss": 0.6984783411026001,
+      "eval_mean_token_accuracy": 0.8095930597085631,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 89.5168,
+      "eval_samples_per_second": 15.863,
+      "eval_steps_per_second": 1.988,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5428155034780502,
+      "epoch": 3.51901025950513,
+      "grad_norm": 1.2790874242782593,
+      "learning_rate": 0.00029881509083286116,
+      "loss": 0.524717378616333,
+      "mean_token_accuracy": 0.8346644505858422,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.550318562917495,
+      "eval_loss": 0.6988471746444702,
+      "eval_mean_token_accuracy": 0.8132969897784544,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 89.2598,
+      "eval_samples_per_second": 15.909,
+      "eval_steps_per_second": 1.994,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5489993002265692,
+      "epoch": 3.567290283645142,
+      "grad_norm": 1.055821418762207,
+      "learning_rate": 0.00029643793741695687,
+      "loss": 0.5421150207519532,
+      "mean_token_accuracy": 0.8345710933208466,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5816821718818685,
+      "eval_loss": 0.6853668689727783,
+      "eval_mean_token_accuracy": 0.8142332360985574,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 89.0424,
+      "eval_samples_per_second": 15.947,
+      "eval_steps_per_second": 1.999,
+      "step": 1480
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.4846844723277824e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1500/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1500/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1500/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1500/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1500/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..7ce4b072fcbe5b636867e13937c56fedee3b68b4
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1500/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.02728394274790723,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1500/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1500/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1500/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1500/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1500/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1500/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1500/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1500/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..4388d2e8803dc4c690037a01a10761ff8ca98509
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1500/trainer_state.json
@@ -0,0 +1,1609 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 3.6155703077851538,
+  "eval_steps": 20,
+  "global_step": 1500,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.024795526266098,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.945375919342041,
+      "learning_rate": 1.6698127428345936e-05,
+      "loss": 1.7825420379638672,
+      "mean_token_accuracy": 0.6315708436071873,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.29921497588747,
+      "eval_loss": 1.169131875038147,
+      "eval_mean_token_accuracy": 0.7185706342204233,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 91.7898,
+      "eval_samples_per_second": 15.47,
+      "eval_steps_per_second": 1.939,
+      "step": 20
+    },
+    {
+      "entropy": 1.0346344470977784,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.7916979789733887,
+      "learning_rate": 3.427510366871008e-05,
+      "loss": 0.9480395317077637,
+      "mean_token_accuracy": 0.7543429024517536,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 0.9673695430326997,
+      "eval_loss": 0.8588430285453796,
+      "eval_mean_token_accuracy": 0.7737233002534073,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 89.8219,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 40
+    },
+    {
+      "entropy": 0.8977566614747048,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7972187995910645,
+      "learning_rate": 5.185207990907422e-05,
+      "loss": 0.8029604911804199,
+      "mean_token_accuracy": 0.7818016350269318,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8436836982041263,
+      "eval_loss": 0.7893033623695374,
+      "eval_mean_token_accuracy": 0.7858914984076211,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 89.9349,
+      "eval_samples_per_second": 15.789,
+      "eval_steps_per_second": 1.979,
+      "step": 60
+    },
+    {
+      "entropy": 0.8650321319699288,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.2128729820251465,
+      "learning_rate": 6.942905614943836e-05,
+      "loss": 0.77417893409729,
+      "mean_token_accuracy": 0.7875479467213153,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8209857180547179,
+      "eval_loss": 0.7572280168533325,
+      "eval_mean_token_accuracy": 0.7925970383574453,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 89.9459,
+      "eval_samples_per_second": 15.787,
+      "eval_steps_per_second": 1.979,
+      "step": 80
+    },
+    {
+      "entropy": 0.8096660938113928,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.1254287958145142,
+      "learning_rate": 8.700603238980251e-05,
+      "loss": 0.7170331001281738,
+      "mean_token_accuracy": 0.7994555257260799,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.792780208788561,
+      "eval_loss": 0.7442984580993652,
+      "eval_mean_token_accuracy": 0.7954704573984896,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 90.294,
+      "eval_samples_per_second": 15.726,
+      "eval_steps_per_second": 1.971,
+      "step": 100
+    },
+    {
+      "entropy": 0.7979829199612141,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2140507698059082,
+      "learning_rate": 0.00010458300863016666,
+      "loss": 0.7141861915588379,
+      "mean_token_accuracy": 0.8000014744699001,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.7938692486018277,
+      "eval_loss": 0.7279797196388245,
+      "eval_mean_token_accuracy": 0.7989798734027348,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 89.7189,
+      "eval_samples_per_second": 15.827,
+      "eval_steps_per_second": 1.984,
+      "step": 120
+    },
+    {
+      "entropy": 0.7916347607970238,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.0506458282470703,
+      "learning_rate": 0.0001221599848705308,
+      "loss": 0.7117055416107178,
+      "mean_token_accuracy": 0.7970046654343605,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.8052781063519167,
+      "eval_loss": 0.722576916217804,
+      "eval_mean_token_accuracy": 0.7982530523551984,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 90.0376,
+      "eval_samples_per_second": 15.771,
+      "eval_steps_per_second": 1.977,
+      "step": 140
+    },
+    {
+      "entropy": 0.7880666613578796,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8409207463264465,
+      "learning_rate": 0.00013973696111089492,
+      "loss": 0.7052523136138916,
+      "mean_token_accuracy": 0.8000123649835587,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7787443134891853,
+      "eval_loss": 0.7137772440910339,
+      "eval_mean_token_accuracy": 0.8014209287220173,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 90.2562,
+      "eval_samples_per_second": 15.733,
+      "eval_steps_per_second": 1.972,
+      "step": 160
+    },
+    {
+      "entropy": 0.7696007996797561,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.0182477235794067,
+      "learning_rate": 0.00015731393735125907,
+      "loss": 0.7008797645568847,
+      "mean_token_accuracy": 0.8007259473204613,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.752259682068664,
+      "eval_loss": 0.7171286344528198,
+      "eval_mean_token_accuracy": 0.80052122574174,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 90.0797,
+      "eval_samples_per_second": 15.764,
+      "eval_steps_per_second": 1.976,
+      "step": 180
+    },
+    {
+      "entropy": 0.7828126326203346,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9553632736206055,
+      "learning_rate": 0.0001748909135916232,
+      "loss": 0.7046853542327881,
+      "mean_token_accuracy": 0.7988562889397144,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7389638169427936,
+      "eval_loss": 0.7187935709953308,
+      "eval_mean_token_accuracy": 0.7995899999409579,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 89.9223,
+      "eval_samples_per_second": 15.791,
+      "eval_steps_per_second": 1.979,
+      "step": 200
+    },
+    {
+      "entropy": 0.7909286297857762,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 1.134459376335144,
+      "learning_rate": 0.00019246788983198735,
+      "loss": 0.6989312171936035,
+      "mean_token_accuracy": 0.7991565138101577,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7496965199374082,
+      "eval_loss": 0.7106321454048157,
+      "eval_mean_token_accuracy": 0.8046183442131857,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 89.6899,
+      "eval_samples_per_second": 15.832,
+      "eval_steps_per_second": 1.985,
+      "step": 220
+    },
+    {
+      "entropy": 0.7651963323354721,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.9557585716247559,
+      "learning_rate": 0.0002100448660723515,
+      "loss": 0.6984559059143066,
+      "mean_token_accuracy": 0.8019823372364044,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7615563079212488,
+      "eval_loss": 0.7098539471626282,
+      "eval_mean_token_accuracy": 0.8047233508544022,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 90.3752,
+      "eval_samples_per_second": 15.712,
+      "eval_steps_per_second": 1.97,
+      "step": 240
+    },
+    {
+      "entropy": 0.787507726252079,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.9279311299324036,
+      "learning_rate": 0.00022762184231271568,
+      "loss": 0.708138370513916,
+      "mean_token_accuracy": 0.8013308703899383,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7856640896100676,
+      "eval_loss": 0.7191570401191711,
+      "eval_mean_token_accuracy": 0.8024677249153008,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 93.7336,
+      "eval_samples_per_second": 15.149,
+      "eval_steps_per_second": 1.899,
+      "step": 260
+    },
+    {
+      "entropy": 0.768489234894514,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.7992776036262512,
+      "learning_rate": 0.0002451988185530798,
+      "loss": 0.7035849571228028,
+      "mean_token_accuracy": 0.8031105332076549,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7682498740346244,
+      "eval_loss": 0.7149022221565247,
+      "eval_mean_token_accuracy": 0.7973497995499814,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 89.7309,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 280
+    },
+    {
+      "entropy": 0.7720955617725849,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.1931514739990234,
+      "learning_rate": 0.0002627757947934439,
+      "loss": 0.6965402603149414,
+      "mean_token_accuracy": 0.8040348663926125,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7214708304807042,
+      "eval_loss": 0.7350823879241943,
+      "eval_mean_token_accuracy": 0.7942911925610532,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 89.6451,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 300
+    },
+    {
+      "entropy": 0.7930095426738262,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8994467258453369,
+      "learning_rate": 0.0002803527710338081,
+      "loss": 0.7331492900848389,
+      "mean_token_accuracy": 0.796185302734375,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.8142535488926963,
+      "eval_loss": 0.7237834334373474,
+      "eval_mean_token_accuracy": 0.799495273761535,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 90.6759,
+      "eval_samples_per_second": 15.66,
+      "eval_steps_per_second": 1.963,
+      "step": 320
+    },
+    {
+      "entropy": 0.7825102441012859,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 1.8224239349365234,
+      "learning_rate": 0.00029792974727417223,
+      "loss": 0.7331175804138184,
+      "mean_token_accuracy": 0.7977667152881622,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7327135443017724,
+      "eval_loss": 0.7388784289360046,
+      "eval_mean_token_accuracy": 0.8007041252730938,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 89.7692,
+      "eval_samples_per_second": 15.818,
+      "eval_steps_per_second": 1.983,
+      "step": 340
+    },
+    {
+      "entropy": 0.819331557303667,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 1.0407779216766357,
+      "learning_rate": 0.0003155067235145364,
+      "loss": 0.7456695556640625,
+      "mean_token_accuracy": 0.7936980701982975,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.8065585420372781,
+      "eval_loss": 0.735519528388977,
+      "eval_mean_token_accuracy": 0.7974206296245704,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 89.4336,
+      "eval_samples_per_second": 15.878,
+      "eval_steps_per_second": 1.99,
+      "step": 360
+    },
+    {
+      "entropy": 0.8114653021097183,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.1209490299224854,
+      "learning_rate": 0.0003330836997549005,
+      "loss": 0.7340899467468261,
+      "mean_token_accuracy": 0.7914272703230381,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.8025527285056168,
+      "eval_loss": 0.7295576930046082,
+      "eval_mean_token_accuracy": 0.7990576628218876,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 89.9579,
+      "eval_samples_per_second": 15.785,
+      "eval_steps_per_second": 1.979,
+      "step": 380
+    },
+    {
+      "entropy": 0.8042497783899307,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 1.2437798976898193,
+      "learning_rate": 0.0003506606759952647,
+      "loss": 0.7381744384765625,
+      "mean_token_accuracy": 0.7967362694442273,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7717011336530193,
+      "eval_loss": 0.7436173558235168,
+      "eval_mean_token_accuracy": 0.7975849482450592,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 89.8411,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 400
+    },
+    {
+      "entropy": 0.7987750010056929,
+      "epoch": 1.012070006035003,
+      "grad_norm": 1.2390918731689453,
+      "learning_rate": 0.000364721224843344,
+      "loss": 0.7317790031433106,
+      "mean_token_accuracy": 0.7986087551364651,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7333170604170038,
+      "eval_loss": 0.7456948757171631,
+      "eval_mean_token_accuracy": 0.7964897577682238,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 89.7288,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 420
+    },
+    {
+      "entropy": 0.7372976846992969,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.1552740335464478,
+      "learning_rate": 0.00036468510102269575,
+      "loss": 0.6734249114990234,
+      "mean_token_accuracy": 0.805834824591875,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.7130355526891987,
+      "eval_loss": 0.7555333375930786,
+      "eval_mean_token_accuracy": 0.7961652781186479,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 89.8713,
+      "eval_samples_per_second": 15.8,
+      "eval_steps_per_second": 1.981,
+      "step": 440
+    },
+    {
+      "entropy": 0.7465668715536594,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 1.1991469860076904,
+      "learning_rate": 0.0003645973816745066,
+      "loss": 0.6871879577636719,
+      "mean_token_accuracy": 0.8042450234293937,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7772979542110743,
+      "eval_loss": 0.7561826705932617,
+      "eval_mean_token_accuracy": 0.7953876427720102,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 89.8395,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 460
+    },
+    {
+      "entropy": 0.7477855801582336,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 1.325486660003662,
+      "learning_rate": 0.00036445809162231435,
+      "loss": 0.6836830139160156,
+      "mean_token_accuracy": 0.8021619468927383,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.7098110676481483,
+      "eval_loss": 0.758007287979126,
+      "eval_mean_token_accuracy": 0.7909953189030122,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 89.6101,
+      "eval_samples_per_second": 15.846,
+      "eval_steps_per_second": 1.986,
+      "step": 480
+    },
+    {
+      "entropy": 0.7632691070437432,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 1.0122262239456177,
+      "learning_rate": 0.0003642672702835562,
+      "loss": 0.7071213245391845,
+      "mean_token_accuracy": 0.8000680930912495,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7645382720432924,
+      "eval_loss": 0.7492606043815613,
+      "eval_mean_token_accuracy": 0.7976858676149604,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 89.6092,
+      "eval_samples_per_second": 15.847,
+      "eval_steps_per_second": 1.986,
+      "step": 500
+    },
+    {
+      "entropy": 0.7623726457357407,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.367774486541748,
+      "learning_rate": 0.00036402497165841384,
+      "loss": 0.6965175151824952,
+      "mean_token_accuracy": 0.8016501650214195,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7868083715438843,
+      "eval_loss": 0.7481449842453003,
+      "eval_mean_token_accuracy": 0.799330582779445,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 89.8241,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 520
+    },
+    {
+      "entropy": 0.7535207405686378,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 1.213614821434021,
+      "learning_rate": 0.00036373126431453207,
+      "loss": 0.7078310489654541,
+      "mean_token_accuracy": 0.7998907208442688,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7829091950748743,
+      "eval_loss": 0.7416072487831116,
+      "eval_mean_token_accuracy": 0.7969146250339036,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 89.4607,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 540
+    },
+    {
+      "entropy": 0.7527363449335098,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 1.0929982662200928,
+      "learning_rate": 0.00036338623136761495,
+      "loss": 0.7058369636535644,
+      "mean_token_accuracy": 0.7997087433934211,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.728419938114252,
+      "eval_loss": 0.744873046875,
+      "eval_mean_token_accuracy": 0.7966634296299366,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 89.6597,
+      "eval_samples_per_second": 15.838,
+      "eval_steps_per_second": 1.985,
+      "step": 560
+    },
+    {
+      "entropy": 0.7406099259853363,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.409364104270935,
+      "learning_rate": 0.00036298997045790515,
+      "loss": 0.6902852058410645,
+      "mean_token_accuracy": 0.805386807769537,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.6991632758231645,
+      "eval_loss": 0.7480319738388062,
+      "eval_mean_token_accuracy": 0.7970068006033308,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 89.8329,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 580
+    },
+    {
+      "entropy": 0.7459449704736472,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.7053111791610718,
+      "learning_rate": 0.00036254259372255277,
+      "loss": 0.7044640064239502,
+      "mean_token_accuracy": 0.8005018755793571,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7454566660891758,
+      "eval_loss": 0.7551484704017639,
+      "eval_mean_token_accuracy": 0.7897093731365846,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 89.8313,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 600
+    },
+    {
+      "entropy": 0.7621391348540782,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 1.174443244934082,
+      "learning_rate": 0.000362044227763882,
+      "loss": 0.7111573696136475,
+      "mean_token_accuracy": 0.798752411454916,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7664387949397055,
+      "eval_loss": 0.7365804314613342,
+      "eval_mean_token_accuracy": 0.7963101200843126,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 89.9623,
+      "eval_samples_per_second": 15.784,
+      "eval_steps_per_second": 1.979,
+      "step": 620
+    },
+    {
+      "entropy": 0.7424944408237935,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.2898184061050415,
+      "learning_rate": 0.000361495013613564,
+      "loss": 0.6913110733032226,
+      "mean_token_accuracy": 0.8009015507996082,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7255479529332579,
+      "eval_loss": 0.7345991730690002,
+      "eval_mean_token_accuracy": 0.8013592945056015,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 89.817,
+      "eval_samples_per_second": 15.81,
+      "eval_steps_per_second": 1.982,
+      "step": 640
+    },
+    {
+      "entropy": 0.7303367637097835,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.1310476064682007,
+      "learning_rate": 0.00036089510669270666,
+      "loss": 0.6875617027282714,
+      "mean_token_accuracy": 0.8035397171974182,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7101308248016271,
+      "eval_loss": 0.7394917011260986,
+      "eval_mean_token_accuracy": 0.7990792942850777,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 89.7431,
+      "eval_samples_per_second": 15.823,
+      "eval_steps_per_second": 1.983,
+      "step": 660
+    },
+    {
+      "entropy": 0.7387678384780884,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.9645389914512634,
+      "learning_rate": 0.0003602446767678725,
+      "loss": 0.6911204338073731,
+      "mean_token_accuracy": 0.8056333385407924,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7281073737010527,
+      "eval_loss": 0.7376708984375,
+      "eval_mean_token_accuracy": 0.7997356415464637,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 90.0338,
+      "eval_samples_per_second": 15.772,
+      "eval_steps_per_second": 1.977,
+      "step": 680
+    },
+    {
+      "entropy": 0.7449554048478604,
+      "epoch": 1.687990343995172,
+      "grad_norm": 1.9069629907608032,
+      "learning_rate": 0.0003595439079030364,
+      "loss": 0.7042286396026611,
+      "mean_token_accuracy": 0.8018639378249646,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7022633817088738,
+      "eval_loss": 0.7320939898490906,
+      "eval_mean_token_accuracy": 0.8006051541044471,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 89.8948,
+      "eval_samples_per_second": 15.796,
+      "eval_steps_per_second": 1.98,
+      "step": 700
+    },
+    {
+      "entropy": 0.723249051719904,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.8333877325057983,
+      "learning_rate": 0.00035879299840749777,
+      "loss": 0.7043714046478271,
+      "mean_token_accuracy": 0.8028345607221127,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.672436946898364,
+      "eval_loss": 0.713233470916748,
+      "eval_mean_token_accuracy": 0.8047760577684038,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 89.6394,
+      "eval_samples_per_second": 15.841,
+      "eval_steps_per_second": 1.986,
+      "step": 720
+    },
+    {
+      "entropy": 0.730014206469059,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.4580037593841553,
+      "learning_rate": 0.00035799216077976137,
+      "loss": 0.6985176563262939,
+      "mean_token_accuracy": 0.8015547141432762,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.6908873795123582,
+      "eval_loss": 0.7198202013969421,
+      "eval_mean_token_accuracy": 0.8024131682481659,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 89.3771,
+      "eval_samples_per_second": 15.888,
+      "eval_steps_per_second": 1.992,
+      "step": 740
+    },
+    {
+      "entropy": 0.7219605796039105,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.51682710647583,
+      "learning_rate": 0.000357141621647403,
+      "loss": 0.6844424724578857,
+      "mean_token_accuracy": 0.8032813094556331,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7113401109582922,
+      "eval_loss": 0.7195309400558472,
+      "eval_mean_token_accuracy": 0.8019187035185568,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 89.4658,
+      "eval_samples_per_second": 15.872,
+      "eval_steps_per_second": 1.99,
+      "step": 760
+    },
+    {
+      "entropy": 0.729826743900776,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 1.0838000774383545,
+      "learning_rate": 0.0003562416217029361,
+      "loss": 0.685858678817749,
+      "mean_token_accuracy": 0.8050567395985126,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.649370267484965,
+      "eval_loss": 0.7177029252052307,
+      "eval_mean_token_accuracy": 0.8050174043419656,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 89.2287,
+      "eval_samples_per_second": 15.914,
+      "eval_steps_per_second": 1.995,
+      "step": 780
+    },
+    {
+      "entropy": 0.6799364153295755,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 1.1364924907684326,
+      "learning_rate": 0.00035529241563569903,
+      "loss": 0.6714428901672364,
+      "mean_token_accuracy": 0.8087341658771038,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6603799047095052,
+      "eval_loss": 0.7201809287071228,
+      "eval_mean_token_accuracy": 0.802863868769635,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 89.2718,
+      "eval_samples_per_second": 15.906,
+      "eval_steps_per_second": 1.994,
+      "step": 800
+    },
+    {
+      "entropy": 0.7078610375523567,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 1.2191942930221558,
+      "learning_rate": 0.0003542942720597808,
+      "loss": 0.6996590614318847,
+      "mean_token_accuracy": 0.8049512051045895,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.7240545264120852,
+      "eval_loss": 0.7301309108734131,
+      "eval_mean_token_accuracy": 0.8015973166133581,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 89.3438,
+      "eval_samples_per_second": 15.894,
+      "eval_steps_per_second": 1.992,
+      "step": 820
+    },
+    {
+      "entropy": 0.6557391556826505,
+      "epoch": 2.024140012070006,
+      "grad_norm": 1.36203134059906,
+      "learning_rate": 0.0003532474734380064,
+      "loss": 0.6327113628387451,
+      "mean_token_accuracy": 0.8142829847026181,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6573808648613062,
+      "eval_loss": 0.7334365844726562,
+      "eval_mean_token_accuracy": 0.8043633893634496,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 89.3359,
+      "eval_samples_per_second": 15.895,
+      "eval_steps_per_second": 1.992,
+      "step": 840
+    },
+    {
+      "entropy": 0.6409128502011299,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 1.5804955959320068,
+      "learning_rate": 0.0003521523160020035,
+      "loss": 0.6023559093475341,
+      "mean_token_accuracy": 0.8222251623868942,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6983234788594621,
+      "eval_loss": 0.7232135534286499,
+      "eval_mean_token_accuracy": 0.8010654268639811,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 89.2615,
+      "eval_samples_per_second": 15.908,
+      "eval_steps_per_second": 1.994,
+      "step": 860
+    },
+    {
+      "entropy": 0.6151149723678827,
+      "epoch": 2.12070006035003,
+      "grad_norm": 1.1975436210632324,
+      "learning_rate": 0.00035100910966837193,
+      "loss": 0.5803564548492431,
+      "mean_token_accuracy": 0.8256325736641884,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6636555573243773,
+      "eval_loss": 0.7205662727355957,
+      "eval_mean_token_accuracy": 0.8036768369460374,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 89.2957,
+      "eval_samples_per_second": 15.902,
+      "eval_steps_per_second": 1.993,
+      "step": 880
+    },
+    {
+      "entropy": 0.6201727617532015,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9907131195068359,
+      "learning_rate": 0.0003498181779509813,
+      "loss": 0.6101872444152832,
+      "mean_token_accuracy": 0.8196512959897518,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.6473788086617931,
+      "eval_loss": 0.7062397003173828,
+      "eval_mean_token_accuracy": 0.8078362184963869,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 88.6198,
+      "eval_samples_per_second": 16.024,
+      "eval_steps_per_second": 2.009,
+      "step": 900
+    },
+    {
+      "entropy": 0.6097332935780286,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.9055633544921875,
+      "learning_rate": 0.00034857985786942026,
+      "loss": 0.5956094264984131,
+      "mean_token_accuracy": 0.8240575887262821,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.612629868341296,
+      "eval_loss": 0.7106770277023315,
+      "eval_mean_token_accuracy": 0.805324277181304,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 89.4383,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 920
+    },
+    {
+      "entropy": 0.6182701248675585,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.9513736367225647,
+      "learning_rate": 0.00034729449985362404,
+      "loss": 0.6019269466400147,
+      "mean_token_accuracy": 0.8187106013298034,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6437647456533453,
+      "eval_loss": 0.7173502445220947,
+      "eval_mean_token_accuracy": 0.8056439573175451,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.9568,
+      "eval_samples_per_second": 15.963,
+      "eval_steps_per_second": 2.001,
+      "step": 940
+    },
+    {
+      "entropy": 0.6508570514619351,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.9606207013130188,
+      "learning_rate": 0.0003459624676447067,
+      "loss": 0.6296409606933594,
+      "mean_token_accuracy": 0.8167732007801533,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6288247135248077,
+      "eval_loss": 0.7160675525665283,
+      "eval_mean_token_accuracy": 0.8066010689467527,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 89.0044,
+      "eval_samples_per_second": 15.954,
+      "eval_steps_per_second": 2.0,
+      "step": 960
+    },
+    {
+      "entropy": 0.6380081418901682,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.9677181839942932,
+      "learning_rate": 0.000344584138192027,
+      "loss": 0.6210546493530273,
+      "mean_token_accuracy": 0.8186135642230511,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6407805031604981,
+      "eval_loss": 0.7055043578147888,
+      "eval_mean_token_accuracy": 0.8070103157772107,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 89.5606,
+      "eval_samples_per_second": 15.855,
+      "eval_steps_per_second": 1.987,
+      "step": 980
+    },
+    {
+      "entropy": 0.6310251638293266,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 2.1046407222747803,
+      "learning_rate": 0.000343159901546516,
+      "loss": 0.6165830135345459,
+      "mean_token_accuracy": 0.8207855716347694,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6461507470420237,
+      "eval_loss": 0.7292136549949646,
+      "eval_mean_token_accuracy": 0.8030418261383356,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 89.315,
+      "eval_samples_per_second": 15.899,
+      "eval_steps_per_second": 1.993,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6256617344915867,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 1.1768349409103394,
+      "learning_rate": 0.0003416901607502972,
+      "loss": 0.6142709255218506,
+      "mean_token_accuracy": 0.8220669947564602,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6150523993406403,
+      "eval_loss": 0.7019563317298889,
+      "eval_mean_token_accuracy": 0.8094323099998946,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 89.419,
+      "eval_samples_per_second": 15.88,
+      "eval_steps_per_second": 1.991,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6126735735684633,
+      "epoch": 2.506940253470127,
+      "grad_norm": 1.8065496683120728,
+      "learning_rate": 0.00034017533172263055,
+      "loss": 0.6056458473205566,
+      "mean_token_accuracy": 0.8203762136399746,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6544345456562685,
+      "eval_loss": 0.7086517214775085,
+      "eval_mean_token_accuracy": 0.806709805901131,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 88.981,
+      "eval_samples_per_second": 15.958,
+      "eval_steps_per_second": 2.0,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6311754353344441,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 1.0576369762420654,
+      "learning_rate": 0.00033861584314221236,
+      "loss": 0.6176392555236816,
+      "mean_token_accuracy": 0.8177410490810871,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6273799200406235,
+      "eval_loss": 0.708163321018219,
+      "eval_mean_token_accuracy": 0.8032017099053672,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 89.7668,
+      "eval_samples_per_second": 15.819,
+      "eval_steps_per_second": 1.983,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6200249589979648,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.997065007686615,
+      "learning_rate": 0.0003370121363258637,
+      "loss": 0.6203888893127442,
+      "mean_token_accuracy": 0.8196275025606156,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6457576811983344,
+      "eval_loss": 0.6972290277481079,
+      "eval_mean_token_accuracy": 0.8095980849158898,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 89.2027,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 1080
+    },
+    {
+      "entropy": 0.626810473203659,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.865475058555603,
+      "learning_rate": 0.0003353646651036438,
+      "loss": 0.6102585792541504,
+      "mean_token_accuracy": 0.819708751142025,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.6412022431914726,
+      "eval_loss": 0.7036139965057373,
+      "eval_mean_token_accuracy": 0.8070317135098275,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 89.4733,
+      "eval_samples_per_second": 15.871,
+      "eval_steps_per_second": 1.989,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6581619590520859,
+      "epoch": 2.700060350030175,
+      "grad_norm": 1.00307035446167,
+      "learning_rate": 0.00033367389569042064,
+      "loss": 0.6397720336914062,
+      "mean_token_accuracy": 0.81413309648633,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6288736402318719,
+      "eval_loss": 0.6966825723648071,
+      "eval_mean_token_accuracy": 0.8094039669867312,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 88.7341,
+      "eval_samples_per_second": 16.003,
+      "eval_steps_per_second": 2.006,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6399412982165813,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.8589329719543457,
+      "learning_rate": 0.0003319403065539384,
+      "loss": 0.6347338199615479,
+      "mean_token_accuracy": 0.8139534957706929,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.683491913455256,
+      "eval_loss": 0.6984149217605591,
+      "eval_mean_token_accuracy": 0.8083989469522841,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 89.3929,
+      "eval_samples_per_second": 15.885,
+      "eval_steps_per_second": 1.991,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6264835461974144,
+      "epoch": 2.796620398310199,
+      "grad_norm": 0.7807853817939758,
+      "learning_rate": 0.0003301643882794163,
+      "loss": 0.6162629127502441,
+      "mean_token_accuracy": 0.8216292977333068,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6578792159476977,
+      "eval_loss": 0.6921441555023193,
+      "eval_mean_token_accuracy": 0.8094299177775223,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 89.2356,
+      "eval_samples_per_second": 15.913,
+      "eval_steps_per_second": 1.995,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6269129011780024,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 1.1675593852996826,
+      "learning_rate": 0.0003283466434307189,
+      "loss": 0.6235532760620117,
+      "mean_token_accuracy": 0.8188897788524627,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6533248447970058,
+      "eval_loss": 0.696849524974823,
+      "eval_mean_token_accuracy": 0.8102703713968898,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2602,
+      "eval_samples_per_second": 16.089,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.6259873781353236,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.9596486687660217,
+      "learning_rate": 0.00032648758640813655,
+      "loss": 0.6118728160858155,
+      "mean_token_accuracy": 0.8171859130263328,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6212078754821521,
+      "eval_loss": 0.6942870020866394,
+      "eval_mean_token_accuracy": 0.810484343365337,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 89.4375,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 1200
+    },
+    {
+      "entropy": 0.6142537571489811,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 0.9625058770179749,
+      "learning_rate": 0.00032458774330281615,
+      "loss": 0.6060261249542236,
+      "mean_token_accuracy": 0.8223113484680653,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6553665667437436,
+      "eval_loss": 0.6835269331932068,
+      "eval_mean_token_accuracy": 0.8134723968720168,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 89.5092,
+      "eval_samples_per_second": 15.864,
+      "eval_steps_per_second": 1.989,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6301972426474094,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.9657333493232727,
+      "learning_rate": 0.0003226476517478835,
+      "loss": 0.6277759075164795,
+      "mean_token_accuracy": 0.8164266042411328,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6594253118788258,
+      "eval_loss": 0.6820935606956482,
+      "eval_mean_token_accuracy": 0.8132368198941263,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 89.6211,
+      "eval_samples_per_second": 15.844,
+      "eval_steps_per_second": 1.986,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5346100655469027,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8541185259819031,
+      "learning_rate": 0.0003206678607662996,
+      "loss": 0.5261692047119141,
+      "mean_token_accuracy": 0.8382289425119177,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5626645662476507,
+      "eval_loss": 0.7028542160987854,
+      "eval_mean_token_accuracy": 0.8133395572056931,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 89.1267,
+      "eval_samples_per_second": 15.932,
+      "eval_steps_per_second": 1.997,
+      "step": 1260
+    },
+    {
+      "entropy": 0.5001101832836866,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.8208815455436707,
+      "learning_rate": 0.0003186489306154935,
+      "loss": 0.4850132942199707,
+      "mean_token_accuracy": 0.8468983843922615,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5889462046743779,
+      "eval_loss": 0.691015899181366,
+      "eval_mean_token_accuracy": 0.8135226466012805,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.9677,
+      "eval_samples_per_second": 15.961,
+      "eval_steps_per_second": 2.001,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5242766339331866,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 0.7893072962760925,
+      "learning_rate": 0.0003165914326288163,
+      "loss": 0.5061577320098877,
+      "mean_token_accuracy": 0.8425608821213245,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5543691962957382,
+      "eval_loss": 0.6913734078407288,
+      "eval_mean_token_accuracy": 0.8154889680026622,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 89.6963,
+      "eval_samples_per_second": 15.831,
+      "eval_steps_per_second": 1.984,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5327721010893584,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.8992080092430115,
+      "learning_rate": 0.0003144959490538604,
+      "loss": 0.5068036556243897,
+      "mean_token_accuracy": 0.8420542575418949,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5734540803378887,
+      "eval_loss": 0.6948716044425964,
+      "eval_mean_token_accuracy": 0.8137217944257715,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 89.4597,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5247377116233111,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.9291845560073853,
+      "learning_rate": 0.0003123630728876902,
+      "loss": 0.5023272037506104,
+      "mean_token_accuracy": 0.8435182586312294,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5492362935891312,
+      "eval_loss": 0.6980633735656738,
+      "eval_mean_token_accuracy": 0.8155082919624415,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 89.2124,
+      "eval_samples_per_second": 15.917,
+      "eval_steps_per_second": 1.995,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5380499072372913,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 1.0477524995803833,
+      "learning_rate": 0.00031019340770903136,
+      "loss": 0.5264591693878173,
+      "mean_token_accuracy": 0.8391959741711617,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5896503307511297,
+      "eval_loss": 0.6997144222259521,
+      "eval_mean_token_accuracy": 0.8124373062942805,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 89.1147,
+      "eval_samples_per_second": 15.935,
+      "eval_steps_per_second": 1.997,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5460653610527515,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 1.2473069429397583,
+      "learning_rate": 0.00030798756750746476,
+      "loss": 0.5319677829742432,
+      "mean_token_accuracy": 0.8364320226013661,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5659052182114526,
+      "eval_loss": 0.6924049258232117,
+      "eval_mean_token_accuracy": 0.8153053585732921,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 89.6468,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5367407951503992,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 1.2679712772369385,
+      "learning_rate": 0.00030574617650967485,
+      "loss": 0.5190927028656006,
+      "mean_token_accuracy": 0.8399469174444676,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5581759144081159,
+      "eval_loss": 0.6933317184448242,
+      "eval_mean_token_accuracy": 0.8139757862251796,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 89.0749,
+      "eval_samples_per_second": 15.942,
+      "eval_steps_per_second": 1.998,
+      "step": 1400
+    },
+    {
+      "entropy": 0.530736118927598,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.9717612862586975,
+      "learning_rate": 0.0003034698690028009,
+      "loss": 0.512363052368164,
+      "mean_token_accuracy": 0.8419127956032753,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5233335836549823,
+      "eval_loss": 0.6958550214767456,
+      "eval_mean_token_accuracy": 0.8158778501360604,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 88.8664,
+      "eval_samples_per_second": 15.979,
+      "eval_steps_per_second": 2.003,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5327631575986743,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7649208903312683,
+      "learning_rate": 0.00030115928915494116,
+      "loss": 0.5202179431915284,
+      "mean_token_accuracy": 0.8384683802723885,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.5960972657364406,
+      "eval_loss": 0.6984783411026001,
+      "eval_mean_token_accuracy": 0.8095930597085631,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 89.5168,
+      "eval_samples_per_second": 15.863,
+      "eval_steps_per_second": 1.988,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5428155034780502,
+      "epoch": 3.51901025950513,
+      "grad_norm": 1.2790874242782593,
+      "learning_rate": 0.00029881509083286116,
+      "loss": 0.524717378616333,
+      "mean_token_accuracy": 0.8346644505858422,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.550318562917495,
+      "eval_loss": 0.6988471746444702,
+      "eval_mean_token_accuracy": 0.8132969897784544,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 89.2598,
+      "eval_samples_per_second": 15.909,
+      "eval_steps_per_second": 1.994,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5489993002265692,
+      "epoch": 3.567290283645142,
+      "grad_norm": 1.055821418762207,
+      "learning_rate": 0.00029643793741695687,
+      "loss": 0.5421150207519532,
+      "mean_token_accuracy": 0.8345710933208466,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5816821718818685,
+      "eval_loss": 0.6853668689727783,
+      "eval_mean_token_accuracy": 0.8142332360985574,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 89.0424,
+      "eval_samples_per_second": 15.947,
+      "eval_steps_per_second": 1.999,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5451365381479263,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.9027532935142517,
+      "learning_rate": 0.00029402850161352583,
+      "loss": 0.5463609218597412,
+      "mean_token_accuracy": 0.8369908526539802,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5344233209832331,
+      "eval_loss": 0.6794531345367432,
+      "eval_mean_token_accuracy": 0.8177759824843889,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 89.123,
+      "eval_samples_per_second": 15.933,
+      "eval_steps_per_second": 1.997,
+      "step": 1500
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.504547942909952e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1520/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1520/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1520/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1520/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1520/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..7ce4b072fcbe5b636867e13937c56fedee3b68b4
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1520/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.02728394274790723,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1520/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1520/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1520/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1520/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1520/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1520/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1520/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1520/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..85d87221ccc2fb5905a3a31bdc863756b734e02a
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1520/trainer_state.json
@@ -0,0 +1,1630 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 3.663850331925166,
+  "eval_steps": 20,
+  "global_step": 1520,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.024795526266098,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.945375919342041,
+      "learning_rate": 1.6698127428345936e-05,
+      "loss": 1.7825420379638672,
+      "mean_token_accuracy": 0.6315708436071873,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.29921497588747,
+      "eval_loss": 1.169131875038147,
+      "eval_mean_token_accuracy": 0.7185706342204233,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 91.7898,
+      "eval_samples_per_second": 15.47,
+      "eval_steps_per_second": 1.939,
+      "step": 20
+    },
+    {
+      "entropy": 1.0346344470977784,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.7916979789733887,
+      "learning_rate": 3.427510366871008e-05,
+      "loss": 0.9480395317077637,
+      "mean_token_accuracy": 0.7543429024517536,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 0.9673695430326997,
+      "eval_loss": 0.8588430285453796,
+      "eval_mean_token_accuracy": 0.7737233002534073,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 89.8219,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 40
+    },
+    {
+      "entropy": 0.8977566614747048,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7972187995910645,
+      "learning_rate": 5.185207990907422e-05,
+      "loss": 0.8029604911804199,
+      "mean_token_accuracy": 0.7818016350269318,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8436836982041263,
+      "eval_loss": 0.7893033623695374,
+      "eval_mean_token_accuracy": 0.7858914984076211,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 89.9349,
+      "eval_samples_per_second": 15.789,
+      "eval_steps_per_second": 1.979,
+      "step": 60
+    },
+    {
+      "entropy": 0.8650321319699288,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.2128729820251465,
+      "learning_rate": 6.942905614943836e-05,
+      "loss": 0.77417893409729,
+      "mean_token_accuracy": 0.7875479467213153,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8209857180547179,
+      "eval_loss": 0.7572280168533325,
+      "eval_mean_token_accuracy": 0.7925970383574453,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 89.9459,
+      "eval_samples_per_second": 15.787,
+      "eval_steps_per_second": 1.979,
+      "step": 80
+    },
+    {
+      "entropy": 0.8096660938113928,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.1254287958145142,
+      "learning_rate": 8.700603238980251e-05,
+      "loss": 0.7170331001281738,
+      "mean_token_accuracy": 0.7994555257260799,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.792780208788561,
+      "eval_loss": 0.7442984580993652,
+      "eval_mean_token_accuracy": 0.7954704573984896,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 90.294,
+      "eval_samples_per_second": 15.726,
+      "eval_steps_per_second": 1.971,
+      "step": 100
+    },
+    {
+      "entropy": 0.7979829199612141,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2140507698059082,
+      "learning_rate": 0.00010458300863016666,
+      "loss": 0.7141861915588379,
+      "mean_token_accuracy": 0.8000014744699001,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.7938692486018277,
+      "eval_loss": 0.7279797196388245,
+      "eval_mean_token_accuracy": 0.7989798734027348,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 89.7189,
+      "eval_samples_per_second": 15.827,
+      "eval_steps_per_second": 1.984,
+      "step": 120
+    },
+    {
+      "entropy": 0.7916347607970238,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.0506458282470703,
+      "learning_rate": 0.0001221599848705308,
+      "loss": 0.7117055416107178,
+      "mean_token_accuracy": 0.7970046654343605,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.8052781063519167,
+      "eval_loss": 0.722576916217804,
+      "eval_mean_token_accuracy": 0.7982530523551984,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 90.0376,
+      "eval_samples_per_second": 15.771,
+      "eval_steps_per_second": 1.977,
+      "step": 140
+    },
+    {
+      "entropy": 0.7880666613578796,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8409207463264465,
+      "learning_rate": 0.00013973696111089492,
+      "loss": 0.7052523136138916,
+      "mean_token_accuracy": 0.8000123649835587,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7787443134891853,
+      "eval_loss": 0.7137772440910339,
+      "eval_mean_token_accuracy": 0.8014209287220173,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 90.2562,
+      "eval_samples_per_second": 15.733,
+      "eval_steps_per_second": 1.972,
+      "step": 160
+    },
+    {
+      "entropy": 0.7696007996797561,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.0182477235794067,
+      "learning_rate": 0.00015731393735125907,
+      "loss": 0.7008797645568847,
+      "mean_token_accuracy": 0.8007259473204613,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.752259682068664,
+      "eval_loss": 0.7171286344528198,
+      "eval_mean_token_accuracy": 0.80052122574174,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 90.0797,
+      "eval_samples_per_second": 15.764,
+      "eval_steps_per_second": 1.976,
+      "step": 180
+    },
+    {
+      "entropy": 0.7828126326203346,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9553632736206055,
+      "learning_rate": 0.0001748909135916232,
+      "loss": 0.7046853542327881,
+      "mean_token_accuracy": 0.7988562889397144,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7389638169427936,
+      "eval_loss": 0.7187935709953308,
+      "eval_mean_token_accuracy": 0.7995899999409579,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 89.9223,
+      "eval_samples_per_second": 15.791,
+      "eval_steps_per_second": 1.979,
+      "step": 200
+    },
+    {
+      "entropy": 0.7909286297857762,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 1.134459376335144,
+      "learning_rate": 0.00019246788983198735,
+      "loss": 0.6989312171936035,
+      "mean_token_accuracy": 0.7991565138101577,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7496965199374082,
+      "eval_loss": 0.7106321454048157,
+      "eval_mean_token_accuracy": 0.8046183442131857,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 89.6899,
+      "eval_samples_per_second": 15.832,
+      "eval_steps_per_second": 1.985,
+      "step": 220
+    },
+    {
+      "entropy": 0.7651963323354721,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.9557585716247559,
+      "learning_rate": 0.0002100448660723515,
+      "loss": 0.6984559059143066,
+      "mean_token_accuracy": 0.8019823372364044,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7615563079212488,
+      "eval_loss": 0.7098539471626282,
+      "eval_mean_token_accuracy": 0.8047233508544022,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 90.3752,
+      "eval_samples_per_second": 15.712,
+      "eval_steps_per_second": 1.97,
+      "step": 240
+    },
+    {
+      "entropy": 0.787507726252079,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.9279311299324036,
+      "learning_rate": 0.00022762184231271568,
+      "loss": 0.708138370513916,
+      "mean_token_accuracy": 0.8013308703899383,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7856640896100676,
+      "eval_loss": 0.7191570401191711,
+      "eval_mean_token_accuracy": 0.8024677249153008,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 93.7336,
+      "eval_samples_per_second": 15.149,
+      "eval_steps_per_second": 1.899,
+      "step": 260
+    },
+    {
+      "entropy": 0.768489234894514,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.7992776036262512,
+      "learning_rate": 0.0002451988185530798,
+      "loss": 0.7035849571228028,
+      "mean_token_accuracy": 0.8031105332076549,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7682498740346244,
+      "eval_loss": 0.7149022221565247,
+      "eval_mean_token_accuracy": 0.7973497995499814,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 89.7309,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 280
+    },
+    {
+      "entropy": 0.7720955617725849,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.1931514739990234,
+      "learning_rate": 0.0002627757947934439,
+      "loss": 0.6965402603149414,
+      "mean_token_accuracy": 0.8040348663926125,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7214708304807042,
+      "eval_loss": 0.7350823879241943,
+      "eval_mean_token_accuracy": 0.7942911925610532,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 89.6451,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 300
+    },
+    {
+      "entropy": 0.7930095426738262,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8994467258453369,
+      "learning_rate": 0.0002803527710338081,
+      "loss": 0.7331492900848389,
+      "mean_token_accuracy": 0.796185302734375,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.8142535488926963,
+      "eval_loss": 0.7237834334373474,
+      "eval_mean_token_accuracy": 0.799495273761535,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 90.6759,
+      "eval_samples_per_second": 15.66,
+      "eval_steps_per_second": 1.963,
+      "step": 320
+    },
+    {
+      "entropy": 0.7825102441012859,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 1.8224239349365234,
+      "learning_rate": 0.00029792974727417223,
+      "loss": 0.7331175804138184,
+      "mean_token_accuracy": 0.7977667152881622,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7327135443017724,
+      "eval_loss": 0.7388784289360046,
+      "eval_mean_token_accuracy": 0.8007041252730938,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 89.7692,
+      "eval_samples_per_second": 15.818,
+      "eval_steps_per_second": 1.983,
+      "step": 340
+    },
+    {
+      "entropy": 0.819331557303667,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 1.0407779216766357,
+      "learning_rate": 0.0003155067235145364,
+      "loss": 0.7456695556640625,
+      "mean_token_accuracy": 0.7936980701982975,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.8065585420372781,
+      "eval_loss": 0.735519528388977,
+      "eval_mean_token_accuracy": 0.7974206296245704,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 89.4336,
+      "eval_samples_per_second": 15.878,
+      "eval_steps_per_second": 1.99,
+      "step": 360
+    },
+    {
+      "entropy": 0.8114653021097183,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.1209490299224854,
+      "learning_rate": 0.0003330836997549005,
+      "loss": 0.7340899467468261,
+      "mean_token_accuracy": 0.7914272703230381,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.8025527285056168,
+      "eval_loss": 0.7295576930046082,
+      "eval_mean_token_accuracy": 0.7990576628218876,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 89.9579,
+      "eval_samples_per_second": 15.785,
+      "eval_steps_per_second": 1.979,
+      "step": 380
+    },
+    {
+      "entropy": 0.8042497783899307,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 1.2437798976898193,
+      "learning_rate": 0.0003506606759952647,
+      "loss": 0.7381744384765625,
+      "mean_token_accuracy": 0.7967362694442273,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7717011336530193,
+      "eval_loss": 0.7436173558235168,
+      "eval_mean_token_accuracy": 0.7975849482450592,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 89.8411,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 400
+    },
+    {
+      "entropy": 0.7987750010056929,
+      "epoch": 1.012070006035003,
+      "grad_norm": 1.2390918731689453,
+      "learning_rate": 0.000364721224843344,
+      "loss": 0.7317790031433106,
+      "mean_token_accuracy": 0.7986087551364651,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7333170604170038,
+      "eval_loss": 0.7456948757171631,
+      "eval_mean_token_accuracy": 0.7964897577682238,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 89.7288,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 420
+    },
+    {
+      "entropy": 0.7372976846992969,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.1552740335464478,
+      "learning_rate": 0.00036468510102269575,
+      "loss": 0.6734249114990234,
+      "mean_token_accuracy": 0.805834824591875,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.7130355526891987,
+      "eval_loss": 0.7555333375930786,
+      "eval_mean_token_accuracy": 0.7961652781186479,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 89.8713,
+      "eval_samples_per_second": 15.8,
+      "eval_steps_per_second": 1.981,
+      "step": 440
+    },
+    {
+      "entropy": 0.7465668715536594,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 1.1991469860076904,
+      "learning_rate": 0.0003645973816745066,
+      "loss": 0.6871879577636719,
+      "mean_token_accuracy": 0.8042450234293937,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7772979542110743,
+      "eval_loss": 0.7561826705932617,
+      "eval_mean_token_accuracy": 0.7953876427720102,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 89.8395,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 460
+    },
+    {
+      "entropy": 0.7477855801582336,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 1.325486660003662,
+      "learning_rate": 0.00036445809162231435,
+      "loss": 0.6836830139160156,
+      "mean_token_accuracy": 0.8021619468927383,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.7098110676481483,
+      "eval_loss": 0.758007287979126,
+      "eval_mean_token_accuracy": 0.7909953189030122,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 89.6101,
+      "eval_samples_per_second": 15.846,
+      "eval_steps_per_second": 1.986,
+      "step": 480
+    },
+    {
+      "entropy": 0.7632691070437432,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 1.0122262239456177,
+      "learning_rate": 0.0003642672702835562,
+      "loss": 0.7071213245391845,
+      "mean_token_accuracy": 0.8000680930912495,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7645382720432924,
+      "eval_loss": 0.7492606043815613,
+      "eval_mean_token_accuracy": 0.7976858676149604,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 89.6092,
+      "eval_samples_per_second": 15.847,
+      "eval_steps_per_second": 1.986,
+      "step": 500
+    },
+    {
+      "entropy": 0.7623726457357407,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.367774486541748,
+      "learning_rate": 0.00036402497165841384,
+      "loss": 0.6965175151824952,
+      "mean_token_accuracy": 0.8016501650214195,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7868083715438843,
+      "eval_loss": 0.7481449842453003,
+      "eval_mean_token_accuracy": 0.799330582779445,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 89.8241,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 520
+    },
+    {
+      "entropy": 0.7535207405686378,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 1.213614821434021,
+      "learning_rate": 0.00036373126431453207,
+      "loss": 0.7078310489654541,
+      "mean_token_accuracy": 0.7998907208442688,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7829091950748743,
+      "eval_loss": 0.7416072487831116,
+      "eval_mean_token_accuracy": 0.7969146250339036,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 89.4607,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 540
+    },
+    {
+      "entropy": 0.7527363449335098,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 1.0929982662200928,
+      "learning_rate": 0.00036338623136761495,
+      "loss": 0.7058369636535644,
+      "mean_token_accuracy": 0.7997087433934211,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.728419938114252,
+      "eval_loss": 0.744873046875,
+      "eval_mean_token_accuracy": 0.7966634296299366,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 89.6597,
+      "eval_samples_per_second": 15.838,
+      "eval_steps_per_second": 1.985,
+      "step": 560
+    },
+    {
+      "entropy": 0.7406099259853363,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.409364104270935,
+      "learning_rate": 0.00036298997045790515,
+      "loss": 0.6902852058410645,
+      "mean_token_accuracy": 0.805386807769537,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.6991632758231645,
+      "eval_loss": 0.7480319738388062,
+      "eval_mean_token_accuracy": 0.7970068006033308,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 89.8329,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 580
+    },
+    {
+      "entropy": 0.7459449704736472,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.7053111791610718,
+      "learning_rate": 0.00036254259372255277,
+      "loss": 0.7044640064239502,
+      "mean_token_accuracy": 0.8005018755793571,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7454566660891758,
+      "eval_loss": 0.7551484704017639,
+      "eval_mean_token_accuracy": 0.7897093731365846,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 89.8313,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 600
+    },
+    {
+      "entropy": 0.7621391348540782,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 1.174443244934082,
+      "learning_rate": 0.000362044227763882,
+      "loss": 0.7111573696136475,
+      "mean_token_accuracy": 0.798752411454916,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7664387949397055,
+      "eval_loss": 0.7365804314613342,
+      "eval_mean_token_accuracy": 0.7963101200843126,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 89.9623,
+      "eval_samples_per_second": 15.784,
+      "eval_steps_per_second": 1.979,
+      "step": 620
+    },
+    {
+      "entropy": 0.7424944408237935,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.2898184061050415,
+      "learning_rate": 0.000361495013613564,
+      "loss": 0.6913110733032226,
+      "mean_token_accuracy": 0.8009015507996082,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7255479529332579,
+      "eval_loss": 0.7345991730690002,
+      "eval_mean_token_accuracy": 0.8013592945056015,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 89.817,
+      "eval_samples_per_second": 15.81,
+      "eval_steps_per_second": 1.982,
+      "step": 640
+    },
+    {
+      "entropy": 0.7303367637097835,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.1310476064682007,
+      "learning_rate": 0.00036089510669270666,
+      "loss": 0.6875617027282714,
+      "mean_token_accuracy": 0.8035397171974182,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7101308248016271,
+      "eval_loss": 0.7394917011260986,
+      "eval_mean_token_accuracy": 0.7990792942850777,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 89.7431,
+      "eval_samples_per_second": 15.823,
+      "eval_steps_per_second": 1.983,
+      "step": 660
+    },
+    {
+      "entropy": 0.7387678384780884,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.9645389914512634,
+      "learning_rate": 0.0003602446767678725,
+      "loss": 0.6911204338073731,
+      "mean_token_accuracy": 0.8056333385407924,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7281073737010527,
+      "eval_loss": 0.7376708984375,
+      "eval_mean_token_accuracy": 0.7997356415464637,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 90.0338,
+      "eval_samples_per_second": 15.772,
+      "eval_steps_per_second": 1.977,
+      "step": 680
+    },
+    {
+      "entropy": 0.7449554048478604,
+      "epoch": 1.687990343995172,
+      "grad_norm": 1.9069629907608032,
+      "learning_rate": 0.0003595439079030364,
+      "loss": 0.7042286396026611,
+      "mean_token_accuracy": 0.8018639378249646,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7022633817088738,
+      "eval_loss": 0.7320939898490906,
+      "eval_mean_token_accuracy": 0.8006051541044471,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 89.8948,
+      "eval_samples_per_second": 15.796,
+      "eval_steps_per_second": 1.98,
+      "step": 700
+    },
+    {
+      "entropy": 0.723249051719904,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.8333877325057983,
+      "learning_rate": 0.00035879299840749777,
+      "loss": 0.7043714046478271,
+      "mean_token_accuracy": 0.8028345607221127,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.672436946898364,
+      "eval_loss": 0.713233470916748,
+      "eval_mean_token_accuracy": 0.8047760577684038,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 89.6394,
+      "eval_samples_per_second": 15.841,
+      "eval_steps_per_second": 1.986,
+      "step": 720
+    },
+    {
+      "entropy": 0.730014206469059,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.4580037593841553,
+      "learning_rate": 0.00035799216077976137,
+      "loss": 0.6985176563262939,
+      "mean_token_accuracy": 0.8015547141432762,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.6908873795123582,
+      "eval_loss": 0.7198202013969421,
+      "eval_mean_token_accuracy": 0.8024131682481659,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 89.3771,
+      "eval_samples_per_second": 15.888,
+      "eval_steps_per_second": 1.992,
+      "step": 740
+    },
+    {
+      "entropy": 0.7219605796039105,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.51682710647583,
+      "learning_rate": 0.000357141621647403,
+      "loss": 0.6844424724578857,
+      "mean_token_accuracy": 0.8032813094556331,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7113401109582922,
+      "eval_loss": 0.7195309400558472,
+      "eval_mean_token_accuracy": 0.8019187035185568,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 89.4658,
+      "eval_samples_per_second": 15.872,
+      "eval_steps_per_second": 1.99,
+      "step": 760
+    },
+    {
+      "entropy": 0.729826743900776,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 1.0838000774383545,
+      "learning_rate": 0.0003562416217029361,
+      "loss": 0.685858678817749,
+      "mean_token_accuracy": 0.8050567395985126,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.649370267484965,
+      "eval_loss": 0.7177029252052307,
+      "eval_mean_token_accuracy": 0.8050174043419656,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 89.2287,
+      "eval_samples_per_second": 15.914,
+      "eval_steps_per_second": 1.995,
+      "step": 780
+    },
+    {
+      "entropy": 0.6799364153295755,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 1.1364924907684326,
+      "learning_rate": 0.00035529241563569903,
+      "loss": 0.6714428901672364,
+      "mean_token_accuracy": 0.8087341658771038,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6603799047095052,
+      "eval_loss": 0.7201809287071228,
+      "eval_mean_token_accuracy": 0.802863868769635,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 89.2718,
+      "eval_samples_per_second": 15.906,
+      "eval_steps_per_second": 1.994,
+      "step": 800
+    },
+    {
+      "entropy": 0.7078610375523567,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 1.2191942930221558,
+      "learning_rate": 0.0003542942720597808,
+      "loss": 0.6996590614318847,
+      "mean_token_accuracy": 0.8049512051045895,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.7240545264120852,
+      "eval_loss": 0.7301309108734131,
+      "eval_mean_token_accuracy": 0.8015973166133581,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 89.3438,
+      "eval_samples_per_second": 15.894,
+      "eval_steps_per_second": 1.992,
+      "step": 820
+    },
+    {
+      "entropy": 0.6557391556826505,
+      "epoch": 2.024140012070006,
+      "grad_norm": 1.36203134059906,
+      "learning_rate": 0.0003532474734380064,
+      "loss": 0.6327113628387451,
+      "mean_token_accuracy": 0.8142829847026181,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6573808648613062,
+      "eval_loss": 0.7334365844726562,
+      "eval_mean_token_accuracy": 0.8043633893634496,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 89.3359,
+      "eval_samples_per_second": 15.895,
+      "eval_steps_per_second": 1.992,
+      "step": 840
+    },
+    {
+      "entropy": 0.6409128502011299,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 1.5804955959320068,
+      "learning_rate": 0.0003521523160020035,
+      "loss": 0.6023559093475341,
+      "mean_token_accuracy": 0.8222251623868942,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6983234788594621,
+      "eval_loss": 0.7232135534286499,
+      "eval_mean_token_accuracy": 0.8010654268639811,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 89.2615,
+      "eval_samples_per_second": 15.908,
+      "eval_steps_per_second": 1.994,
+      "step": 860
+    },
+    {
+      "entropy": 0.6151149723678827,
+      "epoch": 2.12070006035003,
+      "grad_norm": 1.1975436210632324,
+      "learning_rate": 0.00035100910966837193,
+      "loss": 0.5803564548492431,
+      "mean_token_accuracy": 0.8256325736641884,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6636555573243773,
+      "eval_loss": 0.7205662727355957,
+      "eval_mean_token_accuracy": 0.8036768369460374,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 89.2957,
+      "eval_samples_per_second": 15.902,
+      "eval_steps_per_second": 1.993,
+      "step": 880
+    },
+    {
+      "entropy": 0.6201727617532015,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9907131195068359,
+      "learning_rate": 0.0003498181779509813,
+      "loss": 0.6101872444152832,
+      "mean_token_accuracy": 0.8196512959897518,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.6473788086617931,
+      "eval_loss": 0.7062397003173828,
+      "eval_mean_token_accuracy": 0.8078362184963869,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 88.6198,
+      "eval_samples_per_second": 16.024,
+      "eval_steps_per_second": 2.009,
+      "step": 900
+    },
+    {
+      "entropy": 0.6097332935780286,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.9055633544921875,
+      "learning_rate": 0.00034857985786942026,
+      "loss": 0.5956094264984131,
+      "mean_token_accuracy": 0.8240575887262821,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.612629868341296,
+      "eval_loss": 0.7106770277023315,
+      "eval_mean_token_accuracy": 0.805324277181304,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 89.4383,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 920
+    },
+    {
+      "entropy": 0.6182701248675585,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.9513736367225647,
+      "learning_rate": 0.00034729449985362404,
+      "loss": 0.6019269466400147,
+      "mean_token_accuracy": 0.8187106013298034,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6437647456533453,
+      "eval_loss": 0.7173502445220947,
+      "eval_mean_token_accuracy": 0.8056439573175451,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.9568,
+      "eval_samples_per_second": 15.963,
+      "eval_steps_per_second": 2.001,
+      "step": 940
+    },
+    {
+      "entropy": 0.6508570514619351,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.9606207013130188,
+      "learning_rate": 0.0003459624676447067,
+      "loss": 0.6296409606933594,
+      "mean_token_accuracy": 0.8167732007801533,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6288247135248077,
+      "eval_loss": 0.7160675525665283,
+      "eval_mean_token_accuracy": 0.8066010689467527,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 89.0044,
+      "eval_samples_per_second": 15.954,
+      "eval_steps_per_second": 2.0,
+      "step": 960
+    },
+    {
+      "entropy": 0.6380081418901682,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.9677181839942932,
+      "learning_rate": 0.000344584138192027,
+      "loss": 0.6210546493530273,
+      "mean_token_accuracy": 0.8186135642230511,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6407805031604981,
+      "eval_loss": 0.7055043578147888,
+      "eval_mean_token_accuracy": 0.8070103157772107,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 89.5606,
+      "eval_samples_per_second": 15.855,
+      "eval_steps_per_second": 1.987,
+      "step": 980
+    },
+    {
+      "entropy": 0.6310251638293266,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 2.1046407222747803,
+      "learning_rate": 0.000343159901546516,
+      "loss": 0.6165830135345459,
+      "mean_token_accuracy": 0.8207855716347694,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6461507470420237,
+      "eval_loss": 0.7292136549949646,
+      "eval_mean_token_accuracy": 0.8030418261383356,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 89.315,
+      "eval_samples_per_second": 15.899,
+      "eval_steps_per_second": 1.993,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6256617344915867,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 1.1768349409103394,
+      "learning_rate": 0.0003416901607502972,
+      "loss": 0.6142709255218506,
+      "mean_token_accuracy": 0.8220669947564602,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6150523993406403,
+      "eval_loss": 0.7019563317298889,
+      "eval_mean_token_accuracy": 0.8094323099998946,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 89.419,
+      "eval_samples_per_second": 15.88,
+      "eval_steps_per_second": 1.991,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6126735735684633,
+      "epoch": 2.506940253470127,
+      "grad_norm": 1.8065496683120728,
+      "learning_rate": 0.00034017533172263055,
+      "loss": 0.6056458473205566,
+      "mean_token_accuracy": 0.8203762136399746,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6544345456562685,
+      "eval_loss": 0.7086517214775085,
+      "eval_mean_token_accuracy": 0.806709805901131,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 88.981,
+      "eval_samples_per_second": 15.958,
+      "eval_steps_per_second": 2.0,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6311754353344441,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 1.0576369762420654,
+      "learning_rate": 0.00033861584314221236,
+      "loss": 0.6176392555236816,
+      "mean_token_accuracy": 0.8177410490810871,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6273799200406235,
+      "eval_loss": 0.708163321018219,
+      "eval_mean_token_accuracy": 0.8032017099053672,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 89.7668,
+      "eval_samples_per_second": 15.819,
+      "eval_steps_per_second": 1.983,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6200249589979648,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.997065007686615,
+      "learning_rate": 0.0003370121363258637,
+      "loss": 0.6203888893127442,
+      "mean_token_accuracy": 0.8196275025606156,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6457576811983344,
+      "eval_loss": 0.6972290277481079,
+      "eval_mean_token_accuracy": 0.8095980849158898,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 89.2027,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 1080
+    },
+    {
+      "entropy": 0.626810473203659,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.865475058555603,
+      "learning_rate": 0.0003353646651036438,
+      "loss": 0.6102585792541504,
+      "mean_token_accuracy": 0.819708751142025,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.6412022431914726,
+      "eval_loss": 0.7036139965057373,
+      "eval_mean_token_accuracy": 0.8070317135098275,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 89.4733,
+      "eval_samples_per_second": 15.871,
+      "eval_steps_per_second": 1.989,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6581619590520859,
+      "epoch": 2.700060350030175,
+      "grad_norm": 1.00307035446167,
+      "learning_rate": 0.00033367389569042064,
+      "loss": 0.6397720336914062,
+      "mean_token_accuracy": 0.81413309648633,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6288736402318719,
+      "eval_loss": 0.6966825723648071,
+      "eval_mean_token_accuracy": 0.8094039669867312,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 88.7341,
+      "eval_samples_per_second": 16.003,
+      "eval_steps_per_second": 2.006,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6399412982165813,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.8589329719543457,
+      "learning_rate": 0.0003319403065539384,
+      "loss": 0.6347338199615479,
+      "mean_token_accuracy": 0.8139534957706929,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.683491913455256,
+      "eval_loss": 0.6984149217605591,
+      "eval_mean_token_accuracy": 0.8083989469522841,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 89.3929,
+      "eval_samples_per_second": 15.885,
+      "eval_steps_per_second": 1.991,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6264835461974144,
+      "epoch": 2.796620398310199,
+      "grad_norm": 0.7807853817939758,
+      "learning_rate": 0.0003301643882794163,
+      "loss": 0.6162629127502441,
+      "mean_token_accuracy": 0.8216292977333068,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6578792159476977,
+      "eval_loss": 0.6921441555023193,
+      "eval_mean_token_accuracy": 0.8094299177775223,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 89.2356,
+      "eval_samples_per_second": 15.913,
+      "eval_steps_per_second": 1.995,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6269129011780024,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 1.1675593852996826,
+      "learning_rate": 0.0003283466434307189,
+      "loss": 0.6235532760620117,
+      "mean_token_accuracy": 0.8188897788524627,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6533248447970058,
+      "eval_loss": 0.696849524974823,
+      "eval_mean_token_accuracy": 0.8102703713968898,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2602,
+      "eval_samples_per_second": 16.089,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.6259873781353236,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.9596486687660217,
+      "learning_rate": 0.00032648758640813655,
+      "loss": 0.6118728160858155,
+      "mean_token_accuracy": 0.8171859130263328,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6212078754821521,
+      "eval_loss": 0.6942870020866394,
+      "eval_mean_token_accuracy": 0.810484343365337,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 89.4375,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 1200
+    },
+    {
+      "entropy": 0.6142537571489811,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 0.9625058770179749,
+      "learning_rate": 0.00032458774330281615,
+      "loss": 0.6060261249542236,
+      "mean_token_accuracy": 0.8223113484680653,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6553665667437436,
+      "eval_loss": 0.6835269331932068,
+      "eval_mean_token_accuracy": 0.8134723968720168,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 89.5092,
+      "eval_samples_per_second": 15.864,
+      "eval_steps_per_second": 1.989,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6301972426474094,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.9657333493232727,
+      "learning_rate": 0.0003226476517478835,
+      "loss": 0.6277759075164795,
+      "mean_token_accuracy": 0.8164266042411328,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6594253118788258,
+      "eval_loss": 0.6820935606956482,
+      "eval_mean_token_accuracy": 0.8132368198941263,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 89.6211,
+      "eval_samples_per_second": 15.844,
+      "eval_steps_per_second": 1.986,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5346100655469027,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8541185259819031,
+      "learning_rate": 0.0003206678607662996,
+      "loss": 0.5261692047119141,
+      "mean_token_accuracy": 0.8382289425119177,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5626645662476507,
+      "eval_loss": 0.7028542160987854,
+      "eval_mean_token_accuracy": 0.8133395572056931,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 89.1267,
+      "eval_samples_per_second": 15.932,
+      "eval_steps_per_second": 1.997,
+      "step": 1260
+    },
+    {
+      "entropy": 0.5001101832836866,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.8208815455436707,
+      "learning_rate": 0.0003186489306154935,
+      "loss": 0.4850132942199707,
+      "mean_token_accuracy": 0.8468983843922615,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5889462046743779,
+      "eval_loss": 0.691015899181366,
+      "eval_mean_token_accuracy": 0.8135226466012805,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.9677,
+      "eval_samples_per_second": 15.961,
+      "eval_steps_per_second": 2.001,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5242766339331866,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 0.7893072962760925,
+      "learning_rate": 0.0003165914326288163,
+      "loss": 0.5061577320098877,
+      "mean_token_accuracy": 0.8425608821213245,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5543691962957382,
+      "eval_loss": 0.6913734078407288,
+      "eval_mean_token_accuracy": 0.8154889680026622,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 89.6963,
+      "eval_samples_per_second": 15.831,
+      "eval_steps_per_second": 1.984,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5327721010893584,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.8992080092430115,
+      "learning_rate": 0.0003144959490538604,
+      "loss": 0.5068036556243897,
+      "mean_token_accuracy": 0.8420542575418949,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5734540803378887,
+      "eval_loss": 0.6948716044425964,
+      "eval_mean_token_accuracy": 0.8137217944257715,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 89.4597,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5247377116233111,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.9291845560073853,
+      "learning_rate": 0.0003123630728876902,
+      "loss": 0.5023272037506104,
+      "mean_token_accuracy": 0.8435182586312294,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5492362935891312,
+      "eval_loss": 0.6980633735656738,
+      "eval_mean_token_accuracy": 0.8155082919624415,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 89.2124,
+      "eval_samples_per_second": 15.917,
+      "eval_steps_per_second": 1.995,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5380499072372913,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 1.0477524995803833,
+      "learning_rate": 0.00031019340770903136,
+      "loss": 0.5264591693878173,
+      "mean_token_accuracy": 0.8391959741711617,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5896503307511297,
+      "eval_loss": 0.6997144222259521,
+      "eval_mean_token_accuracy": 0.8124373062942805,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 89.1147,
+      "eval_samples_per_second": 15.935,
+      "eval_steps_per_second": 1.997,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5460653610527515,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 1.2473069429397583,
+      "learning_rate": 0.00030798756750746476,
+      "loss": 0.5319677829742432,
+      "mean_token_accuracy": 0.8364320226013661,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5659052182114526,
+      "eval_loss": 0.6924049258232117,
+      "eval_mean_token_accuracy": 0.8153053585732921,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 89.6468,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5367407951503992,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 1.2679712772369385,
+      "learning_rate": 0.00030574617650967485,
+      "loss": 0.5190927028656006,
+      "mean_token_accuracy": 0.8399469174444676,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5581759144081159,
+      "eval_loss": 0.6933317184448242,
+      "eval_mean_token_accuracy": 0.8139757862251796,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 89.0749,
+      "eval_samples_per_second": 15.942,
+      "eval_steps_per_second": 1.998,
+      "step": 1400
+    },
+    {
+      "entropy": 0.530736118927598,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.9717612862586975,
+      "learning_rate": 0.0003034698690028009,
+      "loss": 0.512363052368164,
+      "mean_token_accuracy": 0.8419127956032753,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5233335836549823,
+      "eval_loss": 0.6958550214767456,
+      "eval_mean_token_accuracy": 0.8158778501360604,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 88.8664,
+      "eval_samples_per_second": 15.979,
+      "eval_steps_per_second": 2.003,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5327631575986743,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7649208903312683,
+      "learning_rate": 0.00030115928915494116,
+      "loss": 0.5202179431915284,
+      "mean_token_accuracy": 0.8384683802723885,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.5960972657364406,
+      "eval_loss": 0.6984783411026001,
+      "eval_mean_token_accuracy": 0.8095930597085631,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 89.5168,
+      "eval_samples_per_second": 15.863,
+      "eval_steps_per_second": 1.988,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5428155034780502,
+      "epoch": 3.51901025950513,
+      "grad_norm": 1.2790874242782593,
+      "learning_rate": 0.00029881509083286116,
+      "loss": 0.524717378616333,
+      "mean_token_accuracy": 0.8346644505858422,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.550318562917495,
+      "eval_loss": 0.6988471746444702,
+      "eval_mean_token_accuracy": 0.8132969897784544,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 89.2598,
+      "eval_samples_per_second": 15.909,
+      "eval_steps_per_second": 1.994,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5489993002265692,
+      "epoch": 3.567290283645142,
+      "grad_norm": 1.055821418762207,
+      "learning_rate": 0.00029643793741695687,
+      "loss": 0.5421150207519532,
+      "mean_token_accuracy": 0.8345710933208466,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5816821718818685,
+      "eval_loss": 0.6853668689727783,
+      "eval_mean_token_accuracy": 0.8142332360985574,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 89.0424,
+      "eval_samples_per_second": 15.947,
+      "eval_steps_per_second": 1.999,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5451365381479263,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.9027532935142517,
+      "learning_rate": 0.00029402850161352583,
+      "loss": 0.5463609218597412,
+      "mean_token_accuracy": 0.8369908526539802,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5344233209832331,
+      "eval_loss": 0.6794531345367432,
+      "eval_mean_token_accuracy": 0.8177759824843889,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 89.123,
+      "eval_samples_per_second": 15.933,
+      "eval_steps_per_second": 1.997,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5521467242389917,
+      "epoch": 3.663850331925166,
+      "grad_norm": 1.4606311321258545,
+      "learning_rate": 0.0002915874652643997,
+      "loss": 0.5421902179718018,
+      "mean_token_accuracy": 0.8339455373585224,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5807709094513668,
+      "eval_loss": 0.6949691772460938,
+      "eval_mean_token_accuracy": 0.8121141439743256,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.5437,
+      "eval_samples_per_second": 16.037,
+      "eval_steps_per_second": 2.01,
+      "step": 1520
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.5236305179021312e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1540/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1540/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1540/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1540/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1540/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..7ce4b072fcbe5b636867e13937c56fedee3b68b4
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1540/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.02728394274790723,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1540/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1540/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1540/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1540/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1540/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1540/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1540/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1540/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..b21ffc7cc9e7cf80527877e44e9aaf6ffa7f1d36
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1540/trainer_state.json
@@ -0,0 +1,1651 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 3.712130356065178,
+  "eval_steps": 20,
+  "global_step": 1540,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.024795526266098,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.945375919342041,
+      "learning_rate": 1.6698127428345936e-05,
+      "loss": 1.7825420379638672,
+      "mean_token_accuracy": 0.6315708436071873,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.29921497588747,
+      "eval_loss": 1.169131875038147,
+      "eval_mean_token_accuracy": 0.7185706342204233,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 91.7898,
+      "eval_samples_per_second": 15.47,
+      "eval_steps_per_second": 1.939,
+      "step": 20
+    },
+    {
+      "entropy": 1.0346344470977784,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.7916979789733887,
+      "learning_rate": 3.427510366871008e-05,
+      "loss": 0.9480395317077637,
+      "mean_token_accuracy": 0.7543429024517536,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 0.9673695430326997,
+      "eval_loss": 0.8588430285453796,
+      "eval_mean_token_accuracy": 0.7737233002534073,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 89.8219,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 40
+    },
+    {
+      "entropy": 0.8977566614747048,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7972187995910645,
+      "learning_rate": 5.185207990907422e-05,
+      "loss": 0.8029604911804199,
+      "mean_token_accuracy": 0.7818016350269318,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8436836982041263,
+      "eval_loss": 0.7893033623695374,
+      "eval_mean_token_accuracy": 0.7858914984076211,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 89.9349,
+      "eval_samples_per_second": 15.789,
+      "eval_steps_per_second": 1.979,
+      "step": 60
+    },
+    {
+      "entropy": 0.8650321319699288,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.2128729820251465,
+      "learning_rate": 6.942905614943836e-05,
+      "loss": 0.77417893409729,
+      "mean_token_accuracy": 0.7875479467213153,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8209857180547179,
+      "eval_loss": 0.7572280168533325,
+      "eval_mean_token_accuracy": 0.7925970383574453,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 89.9459,
+      "eval_samples_per_second": 15.787,
+      "eval_steps_per_second": 1.979,
+      "step": 80
+    },
+    {
+      "entropy": 0.8096660938113928,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.1254287958145142,
+      "learning_rate": 8.700603238980251e-05,
+      "loss": 0.7170331001281738,
+      "mean_token_accuracy": 0.7994555257260799,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.792780208788561,
+      "eval_loss": 0.7442984580993652,
+      "eval_mean_token_accuracy": 0.7954704573984896,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 90.294,
+      "eval_samples_per_second": 15.726,
+      "eval_steps_per_second": 1.971,
+      "step": 100
+    },
+    {
+      "entropy": 0.7979829199612141,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2140507698059082,
+      "learning_rate": 0.00010458300863016666,
+      "loss": 0.7141861915588379,
+      "mean_token_accuracy": 0.8000014744699001,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.7938692486018277,
+      "eval_loss": 0.7279797196388245,
+      "eval_mean_token_accuracy": 0.7989798734027348,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 89.7189,
+      "eval_samples_per_second": 15.827,
+      "eval_steps_per_second": 1.984,
+      "step": 120
+    },
+    {
+      "entropy": 0.7916347607970238,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.0506458282470703,
+      "learning_rate": 0.0001221599848705308,
+      "loss": 0.7117055416107178,
+      "mean_token_accuracy": 0.7970046654343605,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.8052781063519167,
+      "eval_loss": 0.722576916217804,
+      "eval_mean_token_accuracy": 0.7982530523551984,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 90.0376,
+      "eval_samples_per_second": 15.771,
+      "eval_steps_per_second": 1.977,
+      "step": 140
+    },
+    {
+      "entropy": 0.7880666613578796,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8409207463264465,
+      "learning_rate": 0.00013973696111089492,
+      "loss": 0.7052523136138916,
+      "mean_token_accuracy": 0.8000123649835587,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7787443134891853,
+      "eval_loss": 0.7137772440910339,
+      "eval_mean_token_accuracy": 0.8014209287220173,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 90.2562,
+      "eval_samples_per_second": 15.733,
+      "eval_steps_per_second": 1.972,
+      "step": 160
+    },
+    {
+      "entropy": 0.7696007996797561,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.0182477235794067,
+      "learning_rate": 0.00015731393735125907,
+      "loss": 0.7008797645568847,
+      "mean_token_accuracy": 0.8007259473204613,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.752259682068664,
+      "eval_loss": 0.7171286344528198,
+      "eval_mean_token_accuracy": 0.80052122574174,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 90.0797,
+      "eval_samples_per_second": 15.764,
+      "eval_steps_per_second": 1.976,
+      "step": 180
+    },
+    {
+      "entropy": 0.7828126326203346,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9553632736206055,
+      "learning_rate": 0.0001748909135916232,
+      "loss": 0.7046853542327881,
+      "mean_token_accuracy": 0.7988562889397144,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7389638169427936,
+      "eval_loss": 0.7187935709953308,
+      "eval_mean_token_accuracy": 0.7995899999409579,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 89.9223,
+      "eval_samples_per_second": 15.791,
+      "eval_steps_per_second": 1.979,
+      "step": 200
+    },
+    {
+      "entropy": 0.7909286297857762,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 1.134459376335144,
+      "learning_rate": 0.00019246788983198735,
+      "loss": 0.6989312171936035,
+      "mean_token_accuracy": 0.7991565138101577,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7496965199374082,
+      "eval_loss": 0.7106321454048157,
+      "eval_mean_token_accuracy": 0.8046183442131857,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 89.6899,
+      "eval_samples_per_second": 15.832,
+      "eval_steps_per_second": 1.985,
+      "step": 220
+    },
+    {
+      "entropy": 0.7651963323354721,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.9557585716247559,
+      "learning_rate": 0.0002100448660723515,
+      "loss": 0.6984559059143066,
+      "mean_token_accuracy": 0.8019823372364044,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7615563079212488,
+      "eval_loss": 0.7098539471626282,
+      "eval_mean_token_accuracy": 0.8047233508544022,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 90.3752,
+      "eval_samples_per_second": 15.712,
+      "eval_steps_per_second": 1.97,
+      "step": 240
+    },
+    {
+      "entropy": 0.787507726252079,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.9279311299324036,
+      "learning_rate": 0.00022762184231271568,
+      "loss": 0.708138370513916,
+      "mean_token_accuracy": 0.8013308703899383,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7856640896100676,
+      "eval_loss": 0.7191570401191711,
+      "eval_mean_token_accuracy": 0.8024677249153008,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 93.7336,
+      "eval_samples_per_second": 15.149,
+      "eval_steps_per_second": 1.899,
+      "step": 260
+    },
+    {
+      "entropy": 0.768489234894514,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.7992776036262512,
+      "learning_rate": 0.0002451988185530798,
+      "loss": 0.7035849571228028,
+      "mean_token_accuracy": 0.8031105332076549,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7682498740346244,
+      "eval_loss": 0.7149022221565247,
+      "eval_mean_token_accuracy": 0.7973497995499814,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 89.7309,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 280
+    },
+    {
+      "entropy": 0.7720955617725849,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.1931514739990234,
+      "learning_rate": 0.0002627757947934439,
+      "loss": 0.6965402603149414,
+      "mean_token_accuracy": 0.8040348663926125,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7214708304807042,
+      "eval_loss": 0.7350823879241943,
+      "eval_mean_token_accuracy": 0.7942911925610532,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 89.6451,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 300
+    },
+    {
+      "entropy": 0.7930095426738262,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8994467258453369,
+      "learning_rate": 0.0002803527710338081,
+      "loss": 0.7331492900848389,
+      "mean_token_accuracy": 0.796185302734375,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.8142535488926963,
+      "eval_loss": 0.7237834334373474,
+      "eval_mean_token_accuracy": 0.799495273761535,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 90.6759,
+      "eval_samples_per_second": 15.66,
+      "eval_steps_per_second": 1.963,
+      "step": 320
+    },
+    {
+      "entropy": 0.7825102441012859,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 1.8224239349365234,
+      "learning_rate": 0.00029792974727417223,
+      "loss": 0.7331175804138184,
+      "mean_token_accuracy": 0.7977667152881622,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7327135443017724,
+      "eval_loss": 0.7388784289360046,
+      "eval_mean_token_accuracy": 0.8007041252730938,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 89.7692,
+      "eval_samples_per_second": 15.818,
+      "eval_steps_per_second": 1.983,
+      "step": 340
+    },
+    {
+      "entropy": 0.819331557303667,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 1.0407779216766357,
+      "learning_rate": 0.0003155067235145364,
+      "loss": 0.7456695556640625,
+      "mean_token_accuracy": 0.7936980701982975,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.8065585420372781,
+      "eval_loss": 0.735519528388977,
+      "eval_mean_token_accuracy": 0.7974206296245704,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 89.4336,
+      "eval_samples_per_second": 15.878,
+      "eval_steps_per_second": 1.99,
+      "step": 360
+    },
+    {
+      "entropy": 0.8114653021097183,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.1209490299224854,
+      "learning_rate": 0.0003330836997549005,
+      "loss": 0.7340899467468261,
+      "mean_token_accuracy": 0.7914272703230381,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.8025527285056168,
+      "eval_loss": 0.7295576930046082,
+      "eval_mean_token_accuracy": 0.7990576628218876,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 89.9579,
+      "eval_samples_per_second": 15.785,
+      "eval_steps_per_second": 1.979,
+      "step": 380
+    },
+    {
+      "entropy": 0.8042497783899307,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 1.2437798976898193,
+      "learning_rate": 0.0003506606759952647,
+      "loss": 0.7381744384765625,
+      "mean_token_accuracy": 0.7967362694442273,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7717011336530193,
+      "eval_loss": 0.7436173558235168,
+      "eval_mean_token_accuracy": 0.7975849482450592,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 89.8411,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 400
+    },
+    {
+      "entropy": 0.7987750010056929,
+      "epoch": 1.012070006035003,
+      "grad_norm": 1.2390918731689453,
+      "learning_rate": 0.000364721224843344,
+      "loss": 0.7317790031433106,
+      "mean_token_accuracy": 0.7986087551364651,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7333170604170038,
+      "eval_loss": 0.7456948757171631,
+      "eval_mean_token_accuracy": 0.7964897577682238,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 89.7288,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 420
+    },
+    {
+      "entropy": 0.7372976846992969,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.1552740335464478,
+      "learning_rate": 0.00036468510102269575,
+      "loss": 0.6734249114990234,
+      "mean_token_accuracy": 0.805834824591875,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.7130355526891987,
+      "eval_loss": 0.7555333375930786,
+      "eval_mean_token_accuracy": 0.7961652781186479,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 89.8713,
+      "eval_samples_per_second": 15.8,
+      "eval_steps_per_second": 1.981,
+      "step": 440
+    },
+    {
+      "entropy": 0.7465668715536594,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 1.1991469860076904,
+      "learning_rate": 0.0003645973816745066,
+      "loss": 0.6871879577636719,
+      "mean_token_accuracy": 0.8042450234293937,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7772979542110743,
+      "eval_loss": 0.7561826705932617,
+      "eval_mean_token_accuracy": 0.7953876427720102,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 89.8395,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 460
+    },
+    {
+      "entropy": 0.7477855801582336,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 1.325486660003662,
+      "learning_rate": 0.00036445809162231435,
+      "loss": 0.6836830139160156,
+      "mean_token_accuracy": 0.8021619468927383,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.7098110676481483,
+      "eval_loss": 0.758007287979126,
+      "eval_mean_token_accuracy": 0.7909953189030122,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 89.6101,
+      "eval_samples_per_second": 15.846,
+      "eval_steps_per_second": 1.986,
+      "step": 480
+    },
+    {
+      "entropy": 0.7632691070437432,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 1.0122262239456177,
+      "learning_rate": 0.0003642672702835562,
+      "loss": 0.7071213245391845,
+      "mean_token_accuracy": 0.8000680930912495,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7645382720432924,
+      "eval_loss": 0.7492606043815613,
+      "eval_mean_token_accuracy": 0.7976858676149604,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 89.6092,
+      "eval_samples_per_second": 15.847,
+      "eval_steps_per_second": 1.986,
+      "step": 500
+    },
+    {
+      "entropy": 0.7623726457357407,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.367774486541748,
+      "learning_rate": 0.00036402497165841384,
+      "loss": 0.6965175151824952,
+      "mean_token_accuracy": 0.8016501650214195,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7868083715438843,
+      "eval_loss": 0.7481449842453003,
+      "eval_mean_token_accuracy": 0.799330582779445,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 89.8241,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 520
+    },
+    {
+      "entropy": 0.7535207405686378,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 1.213614821434021,
+      "learning_rate": 0.00036373126431453207,
+      "loss": 0.7078310489654541,
+      "mean_token_accuracy": 0.7998907208442688,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7829091950748743,
+      "eval_loss": 0.7416072487831116,
+      "eval_mean_token_accuracy": 0.7969146250339036,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 89.4607,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 540
+    },
+    {
+      "entropy": 0.7527363449335098,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 1.0929982662200928,
+      "learning_rate": 0.00036338623136761495,
+      "loss": 0.7058369636535644,
+      "mean_token_accuracy": 0.7997087433934211,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.728419938114252,
+      "eval_loss": 0.744873046875,
+      "eval_mean_token_accuracy": 0.7966634296299366,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 89.6597,
+      "eval_samples_per_second": 15.838,
+      "eval_steps_per_second": 1.985,
+      "step": 560
+    },
+    {
+      "entropy": 0.7406099259853363,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.409364104270935,
+      "learning_rate": 0.00036298997045790515,
+      "loss": 0.6902852058410645,
+      "mean_token_accuracy": 0.805386807769537,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.6991632758231645,
+      "eval_loss": 0.7480319738388062,
+      "eval_mean_token_accuracy": 0.7970068006033308,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 89.8329,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 580
+    },
+    {
+      "entropy": 0.7459449704736472,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.7053111791610718,
+      "learning_rate": 0.00036254259372255277,
+      "loss": 0.7044640064239502,
+      "mean_token_accuracy": 0.8005018755793571,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7454566660891758,
+      "eval_loss": 0.7551484704017639,
+      "eval_mean_token_accuracy": 0.7897093731365846,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 89.8313,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 600
+    },
+    {
+      "entropy": 0.7621391348540782,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 1.174443244934082,
+      "learning_rate": 0.000362044227763882,
+      "loss": 0.7111573696136475,
+      "mean_token_accuracy": 0.798752411454916,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7664387949397055,
+      "eval_loss": 0.7365804314613342,
+      "eval_mean_token_accuracy": 0.7963101200843126,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 89.9623,
+      "eval_samples_per_second": 15.784,
+      "eval_steps_per_second": 1.979,
+      "step": 620
+    },
+    {
+      "entropy": 0.7424944408237935,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.2898184061050415,
+      "learning_rate": 0.000361495013613564,
+      "loss": 0.6913110733032226,
+      "mean_token_accuracy": 0.8009015507996082,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7255479529332579,
+      "eval_loss": 0.7345991730690002,
+      "eval_mean_token_accuracy": 0.8013592945056015,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 89.817,
+      "eval_samples_per_second": 15.81,
+      "eval_steps_per_second": 1.982,
+      "step": 640
+    },
+    {
+      "entropy": 0.7303367637097835,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.1310476064682007,
+      "learning_rate": 0.00036089510669270666,
+      "loss": 0.6875617027282714,
+      "mean_token_accuracy": 0.8035397171974182,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7101308248016271,
+      "eval_loss": 0.7394917011260986,
+      "eval_mean_token_accuracy": 0.7990792942850777,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 89.7431,
+      "eval_samples_per_second": 15.823,
+      "eval_steps_per_second": 1.983,
+      "step": 660
+    },
+    {
+      "entropy": 0.7387678384780884,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.9645389914512634,
+      "learning_rate": 0.0003602446767678725,
+      "loss": 0.6911204338073731,
+      "mean_token_accuracy": 0.8056333385407924,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7281073737010527,
+      "eval_loss": 0.7376708984375,
+      "eval_mean_token_accuracy": 0.7997356415464637,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 90.0338,
+      "eval_samples_per_second": 15.772,
+      "eval_steps_per_second": 1.977,
+      "step": 680
+    },
+    {
+      "entropy": 0.7449554048478604,
+      "epoch": 1.687990343995172,
+      "grad_norm": 1.9069629907608032,
+      "learning_rate": 0.0003595439079030364,
+      "loss": 0.7042286396026611,
+      "mean_token_accuracy": 0.8018639378249646,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7022633817088738,
+      "eval_loss": 0.7320939898490906,
+      "eval_mean_token_accuracy": 0.8006051541044471,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 89.8948,
+      "eval_samples_per_second": 15.796,
+      "eval_steps_per_second": 1.98,
+      "step": 700
+    },
+    {
+      "entropy": 0.723249051719904,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.8333877325057983,
+      "learning_rate": 0.00035879299840749777,
+      "loss": 0.7043714046478271,
+      "mean_token_accuracy": 0.8028345607221127,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.672436946898364,
+      "eval_loss": 0.713233470916748,
+      "eval_mean_token_accuracy": 0.8047760577684038,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 89.6394,
+      "eval_samples_per_second": 15.841,
+      "eval_steps_per_second": 1.986,
+      "step": 720
+    },
+    {
+      "entropy": 0.730014206469059,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.4580037593841553,
+      "learning_rate": 0.00035799216077976137,
+      "loss": 0.6985176563262939,
+      "mean_token_accuracy": 0.8015547141432762,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.6908873795123582,
+      "eval_loss": 0.7198202013969421,
+      "eval_mean_token_accuracy": 0.8024131682481659,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 89.3771,
+      "eval_samples_per_second": 15.888,
+      "eval_steps_per_second": 1.992,
+      "step": 740
+    },
+    {
+      "entropy": 0.7219605796039105,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.51682710647583,
+      "learning_rate": 0.000357141621647403,
+      "loss": 0.6844424724578857,
+      "mean_token_accuracy": 0.8032813094556331,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7113401109582922,
+      "eval_loss": 0.7195309400558472,
+      "eval_mean_token_accuracy": 0.8019187035185568,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 89.4658,
+      "eval_samples_per_second": 15.872,
+      "eval_steps_per_second": 1.99,
+      "step": 760
+    },
+    {
+      "entropy": 0.729826743900776,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 1.0838000774383545,
+      "learning_rate": 0.0003562416217029361,
+      "loss": 0.685858678817749,
+      "mean_token_accuracy": 0.8050567395985126,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.649370267484965,
+      "eval_loss": 0.7177029252052307,
+      "eval_mean_token_accuracy": 0.8050174043419656,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 89.2287,
+      "eval_samples_per_second": 15.914,
+      "eval_steps_per_second": 1.995,
+      "step": 780
+    },
+    {
+      "entropy": 0.6799364153295755,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 1.1364924907684326,
+      "learning_rate": 0.00035529241563569903,
+      "loss": 0.6714428901672364,
+      "mean_token_accuracy": 0.8087341658771038,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6603799047095052,
+      "eval_loss": 0.7201809287071228,
+      "eval_mean_token_accuracy": 0.802863868769635,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 89.2718,
+      "eval_samples_per_second": 15.906,
+      "eval_steps_per_second": 1.994,
+      "step": 800
+    },
+    {
+      "entropy": 0.7078610375523567,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 1.2191942930221558,
+      "learning_rate": 0.0003542942720597808,
+      "loss": 0.6996590614318847,
+      "mean_token_accuracy": 0.8049512051045895,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.7240545264120852,
+      "eval_loss": 0.7301309108734131,
+      "eval_mean_token_accuracy": 0.8015973166133581,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 89.3438,
+      "eval_samples_per_second": 15.894,
+      "eval_steps_per_second": 1.992,
+      "step": 820
+    },
+    {
+      "entropy": 0.6557391556826505,
+      "epoch": 2.024140012070006,
+      "grad_norm": 1.36203134059906,
+      "learning_rate": 0.0003532474734380064,
+      "loss": 0.6327113628387451,
+      "mean_token_accuracy": 0.8142829847026181,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6573808648613062,
+      "eval_loss": 0.7334365844726562,
+      "eval_mean_token_accuracy": 0.8043633893634496,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 89.3359,
+      "eval_samples_per_second": 15.895,
+      "eval_steps_per_second": 1.992,
+      "step": 840
+    },
+    {
+      "entropy": 0.6409128502011299,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 1.5804955959320068,
+      "learning_rate": 0.0003521523160020035,
+      "loss": 0.6023559093475341,
+      "mean_token_accuracy": 0.8222251623868942,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6983234788594621,
+      "eval_loss": 0.7232135534286499,
+      "eval_mean_token_accuracy": 0.8010654268639811,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 89.2615,
+      "eval_samples_per_second": 15.908,
+      "eval_steps_per_second": 1.994,
+      "step": 860
+    },
+    {
+      "entropy": 0.6151149723678827,
+      "epoch": 2.12070006035003,
+      "grad_norm": 1.1975436210632324,
+      "learning_rate": 0.00035100910966837193,
+      "loss": 0.5803564548492431,
+      "mean_token_accuracy": 0.8256325736641884,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6636555573243773,
+      "eval_loss": 0.7205662727355957,
+      "eval_mean_token_accuracy": 0.8036768369460374,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 89.2957,
+      "eval_samples_per_second": 15.902,
+      "eval_steps_per_second": 1.993,
+      "step": 880
+    },
+    {
+      "entropy": 0.6201727617532015,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9907131195068359,
+      "learning_rate": 0.0003498181779509813,
+      "loss": 0.6101872444152832,
+      "mean_token_accuracy": 0.8196512959897518,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.6473788086617931,
+      "eval_loss": 0.7062397003173828,
+      "eval_mean_token_accuracy": 0.8078362184963869,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 88.6198,
+      "eval_samples_per_second": 16.024,
+      "eval_steps_per_second": 2.009,
+      "step": 900
+    },
+    {
+      "entropy": 0.6097332935780286,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.9055633544921875,
+      "learning_rate": 0.00034857985786942026,
+      "loss": 0.5956094264984131,
+      "mean_token_accuracy": 0.8240575887262821,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.612629868341296,
+      "eval_loss": 0.7106770277023315,
+      "eval_mean_token_accuracy": 0.805324277181304,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 89.4383,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 920
+    },
+    {
+      "entropy": 0.6182701248675585,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.9513736367225647,
+      "learning_rate": 0.00034729449985362404,
+      "loss": 0.6019269466400147,
+      "mean_token_accuracy": 0.8187106013298034,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6437647456533453,
+      "eval_loss": 0.7173502445220947,
+      "eval_mean_token_accuracy": 0.8056439573175451,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.9568,
+      "eval_samples_per_second": 15.963,
+      "eval_steps_per_second": 2.001,
+      "step": 940
+    },
+    {
+      "entropy": 0.6508570514619351,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.9606207013130188,
+      "learning_rate": 0.0003459624676447067,
+      "loss": 0.6296409606933594,
+      "mean_token_accuracy": 0.8167732007801533,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6288247135248077,
+      "eval_loss": 0.7160675525665283,
+      "eval_mean_token_accuracy": 0.8066010689467527,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 89.0044,
+      "eval_samples_per_second": 15.954,
+      "eval_steps_per_second": 2.0,
+      "step": 960
+    },
+    {
+      "entropy": 0.6380081418901682,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.9677181839942932,
+      "learning_rate": 0.000344584138192027,
+      "loss": 0.6210546493530273,
+      "mean_token_accuracy": 0.8186135642230511,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6407805031604981,
+      "eval_loss": 0.7055043578147888,
+      "eval_mean_token_accuracy": 0.8070103157772107,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 89.5606,
+      "eval_samples_per_second": 15.855,
+      "eval_steps_per_second": 1.987,
+      "step": 980
+    },
+    {
+      "entropy": 0.6310251638293266,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 2.1046407222747803,
+      "learning_rate": 0.000343159901546516,
+      "loss": 0.6165830135345459,
+      "mean_token_accuracy": 0.8207855716347694,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6461507470420237,
+      "eval_loss": 0.7292136549949646,
+      "eval_mean_token_accuracy": 0.8030418261383356,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 89.315,
+      "eval_samples_per_second": 15.899,
+      "eval_steps_per_second": 1.993,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6256617344915867,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 1.1768349409103394,
+      "learning_rate": 0.0003416901607502972,
+      "loss": 0.6142709255218506,
+      "mean_token_accuracy": 0.8220669947564602,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6150523993406403,
+      "eval_loss": 0.7019563317298889,
+      "eval_mean_token_accuracy": 0.8094323099998946,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 89.419,
+      "eval_samples_per_second": 15.88,
+      "eval_steps_per_second": 1.991,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6126735735684633,
+      "epoch": 2.506940253470127,
+      "grad_norm": 1.8065496683120728,
+      "learning_rate": 0.00034017533172263055,
+      "loss": 0.6056458473205566,
+      "mean_token_accuracy": 0.8203762136399746,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6544345456562685,
+      "eval_loss": 0.7086517214775085,
+      "eval_mean_token_accuracy": 0.806709805901131,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 88.981,
+      "eval_samples_per_second": 15.958,
+      "eval_steps_per_second": 2.0,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6311754353344441,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 1.0576369762420654,
+      "learning_rate": 0.00033861584314221236,
+      "loss": 0.6176392555236816,
+      "mean_token_accuracy": 0.8177410490810871,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6273799200406235,
+      "eval_loss": 0.708163321018219,
+      "eval_mean_token_accuracy": 0.8032017099053672,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 89.7668,
+      "eval_samples_per_second": 15.819,
+      "eval_steps_per_second": 1.983,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6200249589979648,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.997065007686615,
+      "learning_rate": 0.0003370121363258637,
+      "loss": 0.6203888893127442,
+      "mean_token_accuracy": 0.8196275025606156,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6457576811983344,
+      "eval_loss": 0.6972290277481079,
+      "eval_mean_token_accuracy": 0.8095980849158898,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 89.2027,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 1080
+    },
+    {
+      "entropy": 0.626810473203659,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.865475058555603,
+      "learning_rate": 0.0003353646651036438,
+      "loss": 0.6102585792541504,
+      "mean_token_accuracy": 0.819708751142025,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.6412022431914726,
+      "eval_loss": 0.7036139965057373,
+      "eval_mean_token_accuracy": 0.8070317135098275,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 89.4733,
+      "eval_samples_per_second": 15.871,
+      "eval_steps_per_second": 1.989,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6581619590520859,
+      "epoch": 2.700060350030175,
+      "grad_norm": 1.00307035446167,
+      "learning_rate": 0.00033367389569042064,
+      "loss": 0.6397720336914062,
+      "mean_token_accuracy": 0.81413309648633,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6288736402318719,
+      "eval_loss": 0.6966825723648071,
+      "eval_mean_token_accuracy": 0.8094039669867312,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 88.7341,
+      "eval_samples_per_second": 16.003,
+      "eval_steps_per_second": 2.006,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6399412982165813,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.8589329719543457,
+      "learning_rate": 0.0003319403065539384,
+      "loss": 0.6347338199615479,
+      "mean_token_accuracy": 0.8139534957706929,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.683491913455256,
+      "eval_loss": 0.6984149217605591,
+      "eval_mean_token_accuracy": 0.8083989469522841,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 89.3929,
+      "eval_samples_per_second": 15.885,
+      "eval_steps_per_second": 1.991,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6264835461974144,
+      "epoch": 2.796620398310199,
+      "grad_norm": 0.7807853817939758,
+      "learning_rate": 0.0003301643882794163,
+      "loss": 0.6162629127502441,
+      "mean_token_accuracy": 0.8216292977333068,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6578792159476977,
+      "eval_loss": 0.6921441555023193,
+      "eval_mean_token_accuracy": 0.8094299177775223,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 89.2356,
+      "eval_samples_per_second": 15.913,
+      "eval_steps_per_second": 1.995,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6269129011780024,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 1.1675593852996826,
+      "learning_rate": 0.0003283466434307189,
+      "loss": 0.6235532760620117,
+      "mean_token_accuracy": 0.8188897788524627,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6533248447970058,
+      "eval_loss": 0.696849524974823,
+      "eval_mean_token_accuracy": 0.8102703713968898,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2602,
+      "eval_samples_per_second": 16.089,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.6259873781353236,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.9596486687660217,
+      "learning_rate": 0.00032648758640813655,
+      "loss": 0.6118728160858155,
+      "mean_token_accuracy": 0.8171859130263328,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6212078754821521,
+      "eval_loss": 0.6942870020866394,
+      "eval_mean_token_accuracy": 0.810484343365337,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 89.4375,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 1200
+    },
+    {
+      "entropy": 0.6142537571489811,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 0.9625058770179749,
+      "learning_rate": 0.00032458774330281615,
+      "loss": 0.6060261249542236,
+      "mean_token_accuracy": 0.8223113484680653,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6553665667437436,
+      "eval_loss": 0.6835269331932068,
+      "eval_mean_token_accuracy": 0.8134723968720168,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 89.5092,
+      "eval_samples_per_second": 15.864,
+      "eval_steps_per_second": 1.989,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6301972426474094,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.9657333493232727,
+      "learning_rate": 0.0003226476517478835,
+      "loss": 0.6277759075164795,
+      "mean_token_accuracy": 0.8164266042411328,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6594253118788258,
+      "eval_loss": 0.6820935606956482,
+      "eval_mean_token_accuracy": 0.8132368198941263,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 89.6211,
+      "eval_samples_per_second": 15.844,
+      "eval_steps_per_second": 1.986,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5346100655469027,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8541185259819031,
+      "learning_rate": 0.0003206678607662996,
+      "loss": 0.5261692047119141,
+      "mean_token_accuracy": 0.8382289425119177,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5626645662476507,
+      "eval_loss": 0.7028542160987854,
+      "eval_mean_token_accuracy": 0.8133395572056931,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 89.1267,
+      "eval_samples_per_second": 15.932,
+      "eval_steps_per_second": 1.997,
+      "step": 1260
+    },
+    {
+      "entropy": 0.5001101832836866,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.8208815455436707,
+      "learning_rate": 0.0003186489306154935,
+      "loss": 0.4850132942199707,
+      "mean_token_accuracy": 0.8468983843922615,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5889462046743779,
+      "eval_loss": 0.691015899181366,
+      "eval_mean_token_accuracy": 0.8135226466012805,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.9677,
+      "eval_samples_per_second": 15.961,
+      "eval_steps_per_second": 2.001,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5242766339331866,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 0.7893072962760925,
+      "learning_rate": 0.0003165914326288163,
+      "loss": 0.5061577320098877,
+      "mean_token_accuracy": 0.8425608821213245,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5543691962957382,
+      "eval_loss": 0.6913734078407288,
+      "eval_mean_token_accuracy": 0.8154889680026622,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 89.6963,
+      "eval_samples_per_second": 15.831,
+      "eval_steps_per_second": 1.984,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5327721010893584,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.8992080092430115,
+      "learning_rate": 0.0003144959490538604,
+      "loss": 0.5068036556243897,
+      "mean_token_accuracy": 0.8420542575418949,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5734540803378887,
+      "eval_loss": 0.6948716044425964,
+      "eval_mean_token_accuracy": 0.8137217944257715,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 89.4597,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5247377116233111,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.9291845560073853,
+      "learning_rate": 0.0003123630728876902,
+      "loss": 0.5023272037506104,
+      "mean_token_accuracy": 0.8435182586312294,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5492362935891312,
+      "eval_loss": 0.6980633735656738,
+      "eval_mean_token_accuracy": 0.8155082919624415,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 89.2124,
+      "eval_samples_per_second": 15.917,
+      "eval_steps_per_second": 1.995,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5380499072372913,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 1.0477524995803833,
+      "learning_rate": 0.00031019340770903136,
+      "loss": 0.5264591693878173,
+      "mean_token_accuracy": 0.8391959741711617,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5896503307511297,
+      "eval_loss": 0.6997144222259521,
+      "eval_mean_token_accuracy": 0.8124373062942805,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 89.1147,
+      "eval_samples_per_second": 15.935,
+      "eval_steps_per_second": 1.997,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5460653610527515,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 1.2473069429397583,
+      "learning_rate": 0.00030798756750746476,
+      "loss": 0.5319677829742432,
+      "mean_token_accuracy": 0.8364320226013661,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5659052182114526,
+      "eval_loss": 0.6924049258232117,
+      "eval_mean_token_accuracy": 0.8153053585732921,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 89.6468,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5367407951503992,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 1.2679712772369385,
+      "learning_rate": 0.00030574617650967485,
+      "loss": 0.5190927028656006,
+      "mean_token_accuracy": 0.8399469174444676,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5581759144081159,
+      "eval_loss": 0.6933317184448242,
+      "eval_mean_token_accuracy": 0.8139757862251796,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 89.0749,
+      "eval_samples_per_second": 15.942,
+      "eval_steps_per_second": 1.998,
+      "step": 1400
+    },
+    {
+      "entropy": 0.530736118927598,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.9717612862586975,
+      "learning_rate": 0.0003034698690028009,
+      "loss": 0.512363052368164,
+      "mean_token_accuracy": 0.8419127956032753,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5233335836549823,
+      "eval_loss": 0.6958550214767456,
+      "eval_mean_token_accuracy": 0.8158778501360604,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 88.8664,
+      "eval_samples_per_second": 15.979,
+      "eval_steps_per_second": 2.003,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5327631575986743,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7649208903312683,
+      "learning_rate": 0.00030115928915494116,
+      "loss": 0.5202179431915284,
+      "mean_token_accuracy": 0.8384683802723885,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.5960972657364406,
+      "eval_loss": 0.6984783411026001,
+      "eval_mean_token_accuracy": 0.8095930597085631,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 89.5168,
+      "eval_samples_per_second": 15.863,
+      "eval_steps_per_second": 1.988,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5428155034780502,
+      "epoch": 3.51901025950513,
+      "grad_norm": 1.2790874242782593,
+      "learning_rate": 0.00029881509083286116,
+      "loss": 0.524717378616333,
+      "mean_token_accuracy": 0.8346644505858422,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.550318562917495,
+      "eval_loss": 0.6988471746444702,
+      "eval_mean_token_accuracy": 0.8132969897784544,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 89.2598,
+      "eval_samples_per_second": 15.909,
+      "eval_steps_per_second": 1.994,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5489993002265692,
+      "epoch": 3.567290283645142,
+      "grad_norm": 1.055821418762207,
+      "learning_rate": 0.00029643793741695687,
+      "loss": 0.5421150207519532,
+      "mean_token_accuracy": 0.8345710933208466,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5816821718818685,
+      "eval_loss": 0.6853668689727783,
+      "eval_mean_token_accuracy": 0.8142332360985574,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 89.0424,
+      "eval_samples_per_second": 15.947,
+      "eval_steps_per_second": 1.999,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5451365381479263,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.9027532935142517,
+      "learning_rate": 0.00029402850161352583,
+      "loss": 0.5463609218597412,
+      "mean_token_accuracy": 0.8369908526539802,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5344233209832331,
+      "eval_loss": 0.6794531345367432,
+      "eval_mean_token_accuracy": 0.8177759824843889,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 89.123,
+      "eval_samples_per_second": 15.933,
+      "eval_steps_per_second": 1.997,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5521467242389917,
+      "epoch": 3.663850331925166,
+      "grad_norm": 1.4606311321258545,
+      "learning_rate": 0.0002915874652643997,
+      "loss": 0.5421902179718018,
+      "mean_token_accuracy": 0.8339455373585224,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5807709094513668,
+      "eval_loss": 0.6949691772460938,
+      "eval_mean_token_accuracy": 0.8121141439743256,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.5437,
+      "eval_samples_per_second": 16.037,
+      "eval_steps_per_second": 2.01,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5455610822886229,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7269843220710754,
+      "learning_rate": 0.0002891155191539905,
+      "loss": 0.5407682895660401,
+      "mean_token_accuracy": 0.8393648222088814,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.6032205542151847,
+      "eval_loss": 0.6790974140167236,
+      "eval_mean_token_accuracy": 0.8165463112043531,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 89.517,
+      "eval_samples_per_second": 15.863,
+      "eval_steps_per_second": 1.988,
+      "step": 1540
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.544052283952087e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1560/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1560/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1560/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1560/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1560/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..7ce4b072fcbe5b636867e13937c56fedee3b68b4
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1560/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.02728394274790723,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1560/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1560/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1560/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1560/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1560/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1560/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1560/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1560/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..f23b17b0fc8777d6f3a042c826c3da2338e99c9e
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1560/trainer_state.json
@@ -0,0 +1,1672 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 3.7604103802051903,
+  "eval_steps": 20,
+  "global_step": 1560,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.024795526266098,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.945375919342041,
+      "learning_rate": 1.6698127428345936e-05,
+      "loss": 1.7825420379638672,
+      "mean_token_accuracy": 0.6315708436071873,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.29921497588747,
+      "eval_loss": 1.169131875038147,
+      "eval_mean_token_accuracy": 0.7185706342204233,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 91.7898,
+      "eval_samples_per_second": 15.47,
+      "eval_steps_per_second": 1.939,
+      "step": 20
+    },
+    {
+      "entropy": 1.0346344470977784,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.7916979789733887,
+      "learning_rate": 3.427510366871008e-05,
+      "loss": 0.9480395317077637,
+      "mean_token_accuracy": 0.7543429024517536,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 0.9673695430326997,
+      "eval_loss": 0.8588430285453796,
+      "eval_mean_token_accuracy": 0.7737233002534073,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 89.8219,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 40
+    },
+    {
+      "entropy": 0.8977566614747048,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7972187995910645,
+      "learning_rate": 5.185207990907422e-05,
+      "loss": 0.8029604911804199,
+      "mean_token_accuracy": 0.7818016350269318,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8436836982041263,
+      "eval_loss": 0.7893033623695374,
+      "eval_mean_token_accuracy": 0.7858914984076211,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 89.9349,
+      "eval_samples_per_second": 15.789,
+      "eval_steps_per_second": 1.979,
+      "step": 60
+    },
+    {
+      "entropy": 0.8650321319699288,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.2128729820251465,
+      "learning_rate": 6.942905614943836e-05,
+      "loss": 0.77417893409729,
+      "mean_token_accuracy": 0.7875479467213153,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8209857180547179,
+      "eval_loss": 0.7572280168533325,
+      "eval_mean_token_accuracy": 0.7925970383574453,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 89.9459,
+      "eval_samples_per_second": 15.787,
+      "eval_steps_per_second": 1.979,
+      "step": 80
+    },
+    {
+      "entropy": 0.8096660938113928,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.1254287958145142,
+      "learning_rate": 8.700603238980251e-05,
+      "loss": 0.7170331001281738,
+      "mean_token_accuracy": 0.7994555257260799,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.792780208788561,
+      "eval_loss": 0.7442984580993652,
+      "eval_mean_token_accuracy": 0.7954704573984896,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 90.294,
+      "eval_samples_per_second": 15.726,
+      "eval_steps_per_second": 1.971,
+      "step": 100
+    },
+    {
+      "entropy": 0.7979829199612141,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2140507698059082,
+      "learning_rate": 0.00010458300863016666,
+      "loss": 0.7141861915588379,
+      "mean_token_accuracy": 0.8000014744699001,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.7938692486018277,
+      "eval_loss": 0.7279797196388245,
+      "eval_mean_token_accuracy": 0.7989798734027348,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 89.7189,
+      "eval_samples_per_second": 15.827,
+      "eval_steps_per_second": 1.984,
+      "step": 120
+    },
+    {
+      "entropy": 0.7916347607970238,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.0506458282470703,
+      "learning_rate": 0.0001221599848705308,
+      "loss": 0.7117055416107178,
+      "mean_token_accuracy": 0.7970046654343605,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.8052781063519167,
+      "eval_loss": 0.722576916217804,
+      "eval_mean_token_accuracy": 0.7982530523551984,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 90.0376,
+      "eval_samples_per_second": 15.771,
+      "eval_steps_per_second": 1.977,
+      "step": 140
+    },
+    {
+      "entropy": 0.7880666613578796,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8409207463264465,
+      "learning_rate": 0.00013973696111089492,
+      "loss": 0.7052523136138916,
+      "mean_token_accuracy": 0.8000123649835587,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7787443134891853,
+      "eval_loss": 0.7137772440910339,
+      "eval_mean_token_accuracy": 0.8014209287220173,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 90.2562,
+      "eval_samples_per_second": 15.733,
+      "eval_steps_per_second": 1.972,
+      "step": 160
+    },
+    {
+      "entropy": 0.7696007996797561,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.0182477235794067,
+      "learning_rate": 0.00015731393735125907,
+      "loss": 0.7008797645568847,
+      "mean_token_accuracy": 0.8007259473204613,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.752259682068664,
+      "eval_loss": 0.7171286344528198,
+      "eval_mean_token_accuracy": 0.80052122574174,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 90.0797,
+      "eval_samples_per_second": 15.764,
+      "eval_steps_per_second": 1.976,
+      "step": 180
+    },
+    {
+      "entropy": 0.7828126326203346,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9553632736206055,
+      "learning_rate": 0.0001748909135916232,
+      "loss": 0.7046853542327881,
+      "mean_token_accuracy": 0.7988562889397144,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7389638169427936,
+      "eval_loss": 0.7187935709953308,
+      "eval_mean_token_accuracy": 0.7995899999409579,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 89.9223,
+      "eval_samples_per_second": 15.791,
+      "eval_steps_per_second": 1.979,
+      "step": 200
+    },
+    {
+      "entropy": 0.7909286297857762,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 1.134459376335144,
+      "learning_rate": 0.00019246788983198735,
+      "loss": 0.6989312171936035,
+      "mean_token_accuracy": 0.7991565138101577,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7496965199374082,
+      "eval_loss": 0.7106321454048157,
+      "eval_mean_token_accuracy": 0.8046183442131857,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 89.6899,
+      "eval_samples_per_second": 15.832,
+      "eval_steps_per_second": 1.985,
+      "step": 220
+    },
+    {
+      "entropy": 0.7651963323354721,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.9557585716247559,
+      "learning_rate": 0.0002100448660723515,
+      "loss": 0.6984559059143066,
+      "mean_token_accuracy": 0.8019823372364044,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7615563079212488,
+      "eval_loss": 0.7098539471626282,
+      "eval_mean_token_accuracy": 0.8047233508544022,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 90.3752,
+      "eval_samples_per_second": 15.712,
+      "eval_steps_per_second": 1.97,
+      "step": 240
+    },
+    {
+      "entropy": 0.787507726252079,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.9279311299324036,
+      "learning_rate": 0.00022762184231271568,
+      "loss": 0.708138370513916,
+      "mean_token_accuracy": 0.8013308703899383,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7856640896100676,
+      "eval_loss": 0.7191570401191711,
+      "eval_mean_token_accuracy": 0.8024677249153008,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 93.7336,
+      "eval_samples_per_second": 15.149,
+      "eval_steps_per_second": 1.899,
+      "step": 260
+    },
+    {
+      "entropy": 0.768489234894514,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.7992776036262512,
+      "learning_rate": 0.0002451988185530798,
+      "loss": 0.7035849571228028,
+      "mean_token_accuracy": 0.8031105332076549,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7682498740346244,
+      "eval_loss": 0.7149022221565247,
+      "eval_mean_token_accuracy": 0.7973497995499814,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 89.7309,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 280
+    },
+    {
+      "entropy": 0.7720955617725849,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.1931514739990234,
+      "learning_rate": 0.0002627757947934439,
+      "loss": 0.6965402603149414,
+      "mean_token_accuracy": 0.8040348663926125,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7214708304807042,
+      "eval_loss": 0.7350823879241943,
+      "eval_mean_token_accuracy": 0.7942911925610532,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 89.6451,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 300
+    },
+    {
+      "entropy": 0.7930095426738262,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8994467258453369,
+      "learning_rate": 0.0002803527710338081,
+      "loss": 0.7331492900848389,
+      "mean_token_accuracy": 0.796185302734375,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.8142535488926963,
+      "eval_loss": 0.7237834334373474,
+      "eval_mean_token_accuracy": 0.799495273761535,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 90.6759,
+      "eval_samples_per_second": 15.66,
+      "eval_steps_per_second": 1.963,
+      "step": 320
+    },
+    {
+      "entropy": 0.7825102441012859,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 1.8224239349365234,
+      "learning_rate": 0.00029792974727417223,
+      "loss": 0.7331175804138184,
+      "mean_token_accuracy": 0.7977667152881622,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7327135443017724,
+      "eval_loss": 0.7388784289360046,
+      "eval_mean_token_accuracy": 0.8007041252730938,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 89.7692,
+      "eval_samples_per_second": 15.818,
+      "eval_steps_per_second": 1.983,
+      "step": 340
+    },
+    {
+      "entropy": 0.819331557303667,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 1.0407779216766357,
+      "learning_rate": 0.0003155067235145364,
+      "loss": 0.7456695556640625,
+      "mean_token_accuracy": 0.7936980701982975,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.8065585420372781,
+      "eval_loss": 0.735519528388977,
+      "eval_mean_token_accuracy": 0.7974206296245704,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 89.4336,
+      "eval_samples_per_second": 15.878,
+      "eval_steps_per_second": 1.99,
+      "step": 360
+    },
+    {
+      "entropy": 0.8114653021097183,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.1209490299224854,
+      "learning_rate": 0.0003330836997549005,
+      "loss": 0.7340899467468261,
+      "mean_token_accuracy": 0.7914272703230381,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.8025527285056168,
+      "eval_loss": 0.7295576930046082,
+      "eval_mean_token_accuracy": 0.7990576628218876,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 89.9579,
+      "eval_samples_per_second": 15.785,
+      "eval_steps_per_second": 1.979,
+      "step": 380
+    },
+    {
+      "entropy": 0.8042497783899307,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 1.2437798976898193,
+      "learning_rate": 0.0003506606759952647,
+      "loss": 0.7381744384765625,
+      "mean_token_accuracy": 0.7967362694442273,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7717011336530193,
+      "eval_loss": 0.7436173558235168,
+      "eval_mean_token_accuracy": 0.7975849482450592,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 89.8411,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 400
+    },
+    {
+      "entropy": 0.7987750010056929,
+      "epoch": 1.012070006035003,
+      "grad_norm": 1.2390918731689453,
+      "learning_rate": 0.000364721224843344,
+      "loss": 0.7317790031433106,
+      "mean_token_accuracy": 0.7986087551364651,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7333170604170038,
+      "eval_loss": 0.7456948757171631,
+      "eval_mean_token_accuracy": 0.7964897577682238,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 89.7288,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 420
+    },
+    {
+      "entropy": 0.7372976846992969,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.1552740335464478,
+      "learning_rate": 0.00036468510102269575,
+      "loss": 0.6734249114990234,
+      "mean_token_accuracy": 0.805834824591875,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.7130355526891987,
+      "eval_loss": 0.7555333375930786,
+      "eval_mean_token_accuracy": 0.7961652781186479,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 89.8713,
+      "eval_samples_per_second": 15.8,
+      "eval_steps_per_second": 1.981,
+      "step": 440
+    },
+    {
+      "entropy": 0.7465668715536594,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 1.1991469860076904,
+      "learning_rate": 0.0003645973816745066,
+      "loss": 0.6871879577636719,
+      "mean_token_accuracy": 0.8042450234293937,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7772979542110743,
+      "eval_loss": 0.7561826705932617,
+      "eval_mean_token_accuracy": 0.7953876427720102,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 89.8395,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 460
+    },
+    {
+      "entropy": 0.7477855801582336,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 1.325486660003662,
+      "learning_rate": 0.00036445809162231435,
+      "loss": 0.6836830139160156,
+      "mean_token_accuracy": 0.8021619468927383,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.7098110676481483,
+      "eval_loss": 0.758007287979126,
+      "eval_mean_token_accuracy": 0.7909953189030122,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 89.6101,
+      "eval_samples_per_second": 15.846,
+      "eval_steps_per_second": 1.986,
+      "step": 480
+    },
+    {
+      "entropy": 0.7632691070437432,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 1.0122262239456177,
+      "learning_rate": 0.0003642672702835562,
+      "loss": 0.7071213245391845,
+      "mean_token_accuracy": 0.8000680930912495,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7645382720432924,
+      "eval_loss": 0.7492606043815613,
+      "eval_mean_token_accuracy": 0.7976858676149604,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 89.6092,
+      "eval_samples_per_second": 15.847,
+      "eval_steps_per_second": 1.986,
+      "step": 500
+    },
+    {
+      "entropy": 0.7623726457357407,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.367774486541748,
+      "learning_rate": 0.00036402497165841384,
+      "loss": 0.6965175151824952,
+      "mean_token_accuracy": 0.8016501650214195,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7868083715438843,
+      "eval_loss": 0.7481449842453003,
+      "eval_mean_token_accuracy": 0.799330582779445,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 89.8241,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 520
+    },
+    {
+      "entropy": 0.7535207405686378,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 1.213614821434021,
+      "learning_rate": 0.00036373126431453207,
+      "loss": 0.7078310489654541,
+      "mean_token_accuracy": 0.7998907208442688,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7829091950748743,
+      "eval_loss": 0.7416072487831116,
+      "eval_mean_token_accuracy": 0.7969146250339036,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 89.4607,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 540
+    },
+    {
+      "entropy": 0.7527363449335098,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 1.0929982662200928,
+      "learning_rate": 0.00036338623136761495,
+      "loss": 0.7058369636535644,
+      "mean_token_accuracy": 0.7997087433934211,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.728419938114252,
+      "eval_loss": 0.744873046875,
+      "eval_mean_token_accuracy": 0.7966634296299366,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 89.6597,
+      "eval_samples_per_second": 15.838,
+      "eval_steps_per_second": 1.985,
+      "step": 560
+    },
+    {
+      "entropy": 0.7406099259853363,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.409364104270935,
+      "learning_rate": 0.00036298997045790515,
+      "loss": 0.6902852058410645,
+      "mean_token_accuracy": 0.805386807769537,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.6991632758231645,
+      "eval_loss": 0.7480319738388062,
+      "eval_mean_token_accuracy": 0.7970068006033308,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 89.8329,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 580
+    },
+    {
+      "entropy": 0.7459449704736472,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.7053111791610718,
+      "learning_rate": 0.00036254259372255277,
+      "loss": 0.7044640064239502,
+      "mean_token_accuracy": 0.8005018755793571,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7454566660891758,
+      "eval_loss": 0.7551484704017639,
+      "eval_mean_token_accuracy": 0.7897093731365846,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 89.8313,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 600
+    },
+    {
+      "entropy": 0.7621391348540782,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 1.174443244934082,
+      "learning_rate": 0.000362044227763882,
+      "loss": 0.7111573696136475,
+      "mean_token_accuracy": 0.798752411454916,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7664387949397055,
+      "eval_loss": 0.7365804314613342,
+      "eval_mean_token_accuracy": 0.7963101200843126,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 89.9623,
+      "eval_samples_per_second": 15.784,
+      "eval_steps_per_second": 1.979,
+      "step": 620
+    },
+    {
+      "entropy": 0.7424944408237935,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.2898184061050415,
+      "learning_rate": 0.000361495013613564,
+      "loss": 0.6913110733032226,
+      "mean_token_accuracy": 0.8009015507996082,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7255479529332579,
+      "eval_loss": 0.7345991730690002,
+      "eval_mean_token_accuracy": 0.8013592945056015,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 89.817,
+      "eval_samples_per_second": 15.81,
+      "eval_steps_per_second": 1.982,
+      "step": 640
+    },
+    {
+      "entropy": 0.7303367637097835,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.1310476064682007,
+      "learning_rate": 0.00036089510669270666,
+      "loss": 0.6875617027282714,
+      "mean_token_accuracy": 0.8035397171974182,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7101308248016271,
+      "eval_loss": 0.7394917011260986,
+      "eval_mean_token_accuracy": 0.7990792942850777,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 89.7431,
+      "eval_samples_per_second": 15.823,
+      "eval_steps_per_second": 1.983,
+      "step": 660
+    },
+    {
+      "entropy": 0.7387678384780884,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.9645389914512634,
+      "learning_rate": 0.0003602446767678725,
+      "loss": 0.6911204338073731,
+      "mean_token_accuracy": 0.8056333385407924,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7281073737010527,
+      "eval_loss": 0.7376708984375,
+      "eval_mean_token_accuracy": 0.7997356415464637,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 90.0338,
+      "eval_samples_per_second": 15.772,
+      "eval_steps_per_second": 1.977,
+      "step": 680
+    },
+    {
+      "entropy": 0.7449554048478604,
+      "epoch": 1.687990343995172,
+      "grad_norm": 1.9069629907608032,
+      "learning_rate": 0.0003595439079030364,
+      "loss": 0.7042286396026611,
+      "mean_token_accuracy": 0.8018639378249646,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7022633817088738,
+      "eval_loss": 0.7320939898490906,
+      "eval_mean_token_accuracy": 0.8006051541044471,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 89.8948,
+      "eval_samples_per_second": 15.796,
+      "eval_steps_per_second": 1.98,
+      "step": 700
+    },
+    {
+      "entropy": 0.723249051719904,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.8333877325057983,
+      "learning_rate": 0.00035879299840749777,
+      "loss": 0.7043714046478271,
+      "mean_token_accuracy": 0.8028345607221127,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.672436946898364,
+      "eval_loss": 0.713233470916748,
+      "eval_mean_token_accuracy": 0.8047760577684038,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 89.6394,
+      "eval_samples_per_second": 15.841,
+      "eval_steps_per_second": 1.986,
+      "step": 720
+    },
+    {
+      "entropy": 0.730014206469059,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.4580037593841553,
+      "learning_rate": 0.00035799216077976137,
+      "loss": 0.6985176563262939,
+      "mean_token_accuracy": 0.8015547141432762,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.6908873795123582,
+      "eval_loss": 0.7198202013969421,
+      "eval_mean_token_accuracy": 0.8024131682481659,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 89.3771,
+      "eval_samples_per_second": 15.888,
+      "eval_steps_per_second": 1.992,
+      "step": 740
+    },
+    {
+      "entropy": 0.7219605796039105,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.51682710647583,
+      "learning_rate": 0.000357141621647403,
+      "loss": 0.6844424724578857,
+      "mean_token_accuracy": 0.8032813094556331,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7113401109582922,
+      "eval_loss": 0.7195309400558472,
+      "eval_mean_token_accuracy": 0.8019187035185568,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 89.4658,
+      "eval_samples_per_second": 15.872,
+      "eval_steps_per_second": 1.99,
+      "step": 760
+    },
+    {
+      "entropy": 0.729826743900776,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 1.0838000774383545,
+      "learning_rate": 0.0003562416217029361,
+      "loss": 0.685858678817749,
+      "mean_token_accuracy": 0.8050567395985126,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.649370267484965,
+      "eval_loss": 0.7177029252052307,
+      "eval_mean_token_accuracy": 0.8050174043419656,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 89.2287,
+      "eval_samples_per_second": 15.914,
+      "eval_steps_per_second": 1.995,
+      "step": 780
+    },
+    {
+      "entropy": 0.6799364153295755,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 1.1364924907684326,
+      "learning_rate": 0.00035529241563569903,
+      "loss": 0.6714428901672364,
+      "mean_token_accuracy": 0.8087341658771038,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6603799047095052,
+      "eval_loss": 0.7201809287071228,
+      "eval_mean_token_accuracy": 0.802863868769635,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 89.2718,
+      "eval_samples_per_second": 15.906,
+      "eval_steps_per_second": 1.994,
+      "step": 800
+    },
+    {
+      "entropy": 0.7078610375523567,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 1.2191942930221558,
+      "learning_rate": 0.0003542942720597808,
+      "loss": 0.6996590614318847,
+      "mean_token_accuracy": 0.8049512051045895,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.7240545264120852,
+      "eval_loss": 0.7301309108734131,
+      "eval_mean_token_accuracy": 0.8015973166133581,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 89.3438,
+      "eval_samples_per_second": 15.894,
+      "eval_steps_per_second": 1.992,
+      "step": 820
+    },
+    {
+      "entropy": 0.6557391556826505,
+      "epoch": 2.024140012070006,
+      "grad_norm": 1.36203134059906,
+      "learning_rate": 0.0003532474734380064,
+      "loss": 0.6327113628387451,
+      "mean_token_accuracy": 0.8142829847026181,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6573808648613062,
+      "eval_loss": 0.7334365844726562,
+      "eval_mean_token_accuracy": 0.8043633893634496,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 89.3359,
+      "eval_samples_per_second": 15.895,
+      "eval_steps_per_second": 1.992,
+      "step": 840
+    },
+    {
+      "entropy": 0.6409128502011299,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 1.5804955959320068,
+      "learning_rate": 0.0003521523160020035,
+      "loss": 0.6023559093475341,
+      "mean_token_accuracy": 0.8222251623868942,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6983234788594621,
+      "eval_loss": 0.7232135534286499,
+      "eval_mean_token_accuracy": 0.8010654268639811,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 89.2615,
+      "eval_samples_per_second": 15.908,
+      "eval_steps_per_second": 1.994,
+      "step": 860
+    },
+    {
+      "entropy": 0.6151149723678827,
+      "epoch": 2.12070006035003,
+      "grad_norm": 1.1975436210632324,
+      "learning_rate": 0.00035100910966837193,
+      "loss": 0.5803564548492431,
+      "mean_token_accuracy": 0.8256325736641884,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6636555573243773,
+      "eval_loss": 0.7205662727355957,
+      "eval_mean_token_accuracy": 0.8036768369460374,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 89.2957,
+      "eval_samples_per_second": 15.902,
+      "eval_steps_per_second": 1.993,
+      "step": 880
+    },
+    {
+      "entropy": 0.6201727617532015,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9907131195068359,
+      "learning_rate": 0.0003498181779509813,
+      "loss": 0.6101872444152832,
+      "mean_token_accuracy": 0.8196512959897518,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.6473788086617931,
+      "eval_loss": 0.7062397003173828,
+      "eval_mean_token_accuracy": 0.8078362184963869,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 88.6198,
+      "eval_samples_per_second": 16.024,
+      "eval_steps_per_second": 2.009,
+      "step": 900
+    },
+    {
+      "entropy": 0.6097332935780286,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.9055633544921875,
+      "learning_rate": 0.00034857985786942026,
+      "loss": 0.5956094264984131,
+      "mean_token_accuracy": 0.8240575887262821,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.612629868341296,
+      "eval_loss": 0.7106770277023315,
+      "eval_mean_token_accuracy": 0.805324277181304,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 89.4383,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 920
+    },
+    {
+      "entropy": 0.6182701248675585,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.9513736367225647,
+      "learning_rate": 0.00034729449985362404,
+      "loss": 0.6019269466400147,
+      "mean_token_accuracy": 0.8187106013298034,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6437647456533453,
+      "eval_loss": 0.7173502445220947,
+      "eval_mean_token_accuracy": 0.8056439573175451,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.9568,
+      "eval_samples_per_second": 15.963,
+      "eval_steps_per_second": 2.001,
+      "step": 940
+    },
+    {
+      "entropy": 0.6508570514619351,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.9606207013130188,
+      "learning_rate": 0.0003459624676447067,
+      "loss": 0.6296409606933594,
+      "mean_token_accuracy": 0.8167732007801533,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6288247135248077,
+      "eval_loss": 0.7160675525665283,
+      "eval_mean_token_accuracy": 0.8066010689467527,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 89.0044,
+      "eval_samples_per_second": 15.954,
+      "eval_steps_per_second": 2.0,
+      "step": 960
+    },
+    {
+      "entropy": 0.6380081418901682,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.9677181839942932,
+      "learning_rate": 0.000344584138192027,
+      "loss": 0.6210546493530273,
+      "mean_token_accuracy": 0.8186135642230511,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6407805031604981,
+      "eval_loss": 0.7055043578147888,
+      "eval_mean_token_accuracy": 0.8070103157772107,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 89.5606,
+      "eval_samples_per_second": 15.855,
+      "eval_steps_per_second": 1.987,
+      "step": 980
+    },
+    {
+      "entropy": 0.6310251638293266,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 2.1046407222747803,
+      "learning_rate": 0.000343159901546516,
+      "loss": 0.6165830135345459,
+      "mean_token_accuracy": 0.8207855716347694,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6461507470420237,
+      "eval_loss": 0.7292136549949646,
+      "eval_mean_token_accuracy": 0.8030418261383356,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 89.315,
+      "eval_samples_per_second": 15.899,
+      "eval_steps_per_second": 1.993,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6256617344915867,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 1.1768349409103394,
+      "learning_rate": 0.0003416901607502972,
+      "loss": 0.6142709255218506,
+      "mean_token_accuracy": 0.8220669947564602,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6150523993406403,
+      "eval_loss": 0.7019563317298889,
+      "eval_mean_token_accuracy": 0.8094323099998946,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 89.419,
+      "eval_samples_per_second": 15.88,
+      "eval_steps_per_second": 1.991,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6126735735684633,
+      "epoch": 2.506940253470127,
+      "grad_norm": 1.8065496683120728,
+      "learning_rate": 0.00034017533172263055,
+      "loss": 0.6056458473205566,
+      "mean_token_accuracy": 0.8203762136399746,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6544345456562685,
+      "eval_loss": 0.7086517214775085,
+      "eval_mean_token_accuracy": 0.806709805901131,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 88.981,
+      "eval_samples_per_second": 15.958,
+      "eval_steps_per_second": 2.0,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6311754353344441,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 1.0576369762420654,
+      "learning_rate": 0.00033861584314221236,
+      "loss": 0.6176392555236816,
+      "mean_token_accuracy": 0.8177410490810871,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6273799200406235,
+      "eval_loss": 0.708163321018219,
+      "eval_mean_token_accuracy": 0.8032017099053672,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 89.7668,
+      "eval_samples_per_second": 15.819,
+      "eval_steps_per_second": 1.983,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6200249589979648,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.997065007686615,
+      "learning_rate": 0.0003370121363258637,
+      "loss": 0.6203888893127442,
+      "mean_token_accuracy": 0.8196275025606156,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6457576811983344,
+      "eval_loss": 0.6972290277481079,
+      "eval_mean_token_accuracy": 0.8095980849158898,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 89.2027,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 1080
+    },
+    {
+      "entropy": 0.626810473203659,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.865475058555603,
+      "learning_rate": 0.0003353646651036438,
+      "loss": 0.6102585792541504,
+      "mean_token_accuracy": 0.819708751142025,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.6412022431914726,
+      "eval_loss": 0.7036139965057373,
+      "eval_mean_token_accuracy": 0.8070317135098275,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 89.4733,
+      "eval_samples_per_second": 15.871,
+      "eval_steps_per_second": 1.989,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6581619590520859,
+      "epoch": 2.700060350030175,
+      "grad_norm": 1.00307035446167,
+      "learning_rate": 0.00033367389569042064,
+      "loss": 0.6397720336914062,
+      "mean_token_accuracy": 0.81413309648633,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6288736402318719,
+      "eval_loss": 0.6966825723648071,
+      "eval_mean_token_accuracy": 0.8094039669867312,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 88.7341,
+      "eval_samples_per_second": 16.003,
+      "eval_steps_per_second": 2.006,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6399412982165813,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.8589329719543457,
+      "learning_rate": 0.0003319403065539384,
+      "loss": 0.6347338199615479,
+      "mean_token_accuracy": 0.8139534957706929,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.683491913455256,
+      "eval_loss": 0.6984149217605591,
+      "eval_mean_token_accuracy": 0.8083989469522841,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 89.3929,
+      "eval_samples_per_second": 15.885,
+      "eval_steps_per_second": 1.991,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6264835461974144,
+      "epoch": 2.796620398310199,
+      "grad_norm": 0.7807853817939758,
+      "learning_rate": 0.0003301643882794163,
+      "loss": 0.6162629127502441,
+      "mean_token_accuracy": 0.8216292977333068,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6578792159476977,
+      "eval_loss": 0.6921441555023193,
+      "eval_mean_token_accuracy": 0.8094299177775223,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 89.2356,
+      "eval_samples_per_second": 15.913,
+      "eval_steps_per_second": 1.995,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6269129011780024,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 1.1675593852996826,
+      "learning_rate": 0.0003283466434307189,
+      "loss": 0.6235532760620117,
+      "mean_token_accuracy": 0.8188897788524627,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6533248447970058,
+      "eval_loss": 0.696849524974823,
+      "eval_mean_token_accuracy": 0.8102703713968898,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2602,
+      "eval_samples_per_second": 16.089,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.6259873781353236,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.9596486687660217,
+      "learning_rate": 0.00032648758640813655,
+      "loss": 0.6118728160858155,
+      "mean_token_accuracy": 0.8171859130263328,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6212078754821521,
+      "eval_loss": 0.6942870020866394,
+      "eval_mean_token_accuracy": 0.810484343365337,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 89.4375,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 1200
+    },
+    {
+      "entropy": 0.6142537571489811,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 0.9625058770179749,
+      "learning_rate": 0.00032458774330281615,
+      "loss": 0.6060261249542236,
+      "mean_token_accuracy": 0.8223113484680653,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6553665667437436,
+      "eval_loss": 0.6835269331932068,
+      "eval_mean_token_accuracy": 0.8134723968720168,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 89.5092,
+      "eval_samples_per_second": 15.864,
+      "eval_steps_per_second": 1.989,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6301972426474094,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.9657333493232727,
+      "learning_rate": 0.0003226476517478835,
+      "loss": 0.6277759075164795,
+      "mean_token_accuracy": 0.8164266042411328,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6594253118788258,
+      "eval_loss": 0.6820935606956482,
+      "eval_mean_token_accuracy": 0.8132368198941263,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 89.6211,
+      "eval_samples_per_second": 15.844,
+      "eval_steps_per_second": 1.986,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5346100655469027,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8541185259819031,
+      "learning_rate": 0.0003206678607662996,
+      "loss": 0.5261692047119141,
+      "mean_token_accuracy": 0.8382289425119177,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5626645662476507,
+      "eval_loss": 0.7028542160987854,
+      "eval_mean_token_accuracy": 0.8133395572056931,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 89.1267,
+      "eval_samples_per_second": 15.932,
+      "eval_steps_per_second": 1.997,
+      "step": 1260
+    },
+    {
+      "entropy": 0.5001101832836866,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.8208815455436707,
+      "learning_rate": 0.0003186489306154935,
+      "loss": 0.4850132942199707,
+      "mean_token_accuracy": 0.8468983843922615,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5889462046743779,
+      "eval_loss": 0.691015899181366,
+      "eval_mean_token_accuracy": 0.8135226466012805,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.9677,
+      "eval_samples_per_second": 15.961,
+      "eval_steps_per_second": 2.001,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5242766339331866,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 0.7893072962760925,
+      "learning_rate": 0.0003165914326288163,
+      "loss": 0.5061577320098877,
+      "mean_token_accuracy": 0.8425608821213245,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5543691962957382,
+      "eval_loss": 0.6913734078407288,
+      "eval_mean_token_accuracy": 0.8154889680026622,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 89.6963,
+      "eval_samples_per_second": 15.831,
+      "eval_steps_per_second": 1.984,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5327721010893584,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.8992080092430115,
+      "learning_rate": 0.0003144959490538604,
+      "loss": 0.5068036556243897,
+      "mean_token_accuracy": 0.8420542575418949,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5734540803378887,
+      "eval_loss": 0.6948716044425964,
+      "eval_mean_token_accuracy": 0.8137217944257715,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 89.4597,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5247377116233111,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.9291845560073853,
+      "learning_rate": 0.0003123630728876902,
+      "loss": 0.5023272037506104,
+      "mean_token_accuracy": 0.8435182586312294,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5492362935891312,
+      "eval_loss": 0.6980633735656738,
+      "eval_mean_token_accuracy": 0.8155082919624415,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 89.2124,
+      "eval_samples_per_second": 15.917,
+      "eval_steps_per_second": 1.995,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5380499072372913,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 1.0477524995803833,
+      "learning_rate": 0.00031019340770903136,
+      "loss": 0.5264591693878173,
+      "mean_token_accuracy": 0.8391959741711617,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5896503307511297,
+      "eval_loss": 0.6997144222259521,
+      "eval_mean_token_accuracy": 0.8124373062942805,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 89.1147,
+      "eval_samples_per_second": 15.935,
+      "eval_steps_per_second": 1.997,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5460653610527515,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 1.2473069429397583,
+      "learning_rate": 0.00030798756750746476,
+      "loss": 0.5319677829742432,
+      "mean_token_accuracy": 0.8364320226013661,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5659052182114526,
+      "eval_loss": 0.6924049258232117,
+      "eval_mean_token_accuracy": 0.8153053585732921,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 89.6468,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5367407951503992,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 1.2679712772369385,
+      "learning_rate": 0.00030574617650967485,
+      "loss": 0.5190927028656006,
+      "mean_token_accuracy": 0.8399469174444676,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5581759144081159,
+      "eval_loss": 0.6933317184448242,
+      "eval_mean_token_accuracy": 0.8139757862251796,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 89.0749,
+      "eval_samples_per_second": 15.942,
+      "eval_steps_per_second": 1.998,
+      "step": 1400
+    },
+    {
+      "entropy": 0.530736118927598,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.9717612862586975,
+      "learning_rate": 0.0003034698690028009,
+      "loss": 0.512363052368164,
+      "mean_token_accuracy": 0.8419127956032753,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5233335836549823,
+      "eval_loss": 0.6958550214767456,
+      "eval_mean_token_accuracy": 0.8158778501360604,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 88.8664,
+      "eval_samples_per_second": 15.979,
+      "eval_steps_per_second": 2.003,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5327631575986743,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7649208903312683,
+      "learning_rate": 0.00030115928915494116,
+      "loss": 0.5202179431915284,
+      "mean_token_accuracy": 0.8384683802723885,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.5960972657364406,
+      "eval_loss": 0.6984783411026001,
+      "eval_mean_token_accuracy": 0.8095930597085631,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 89.5168,
+      "eval_samples_per_second": 15.863,
+      "eval_steps_per_second": 1.988,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5428155034780502,
+      "epoch": 3.51901025950513,
+      "grad_norm": 1.2790874242782593,
+      "learning_rate": 0.00029881509083286116,
+      "loss": 0.524717378616333,
+      "mean_token_accuracy": 0.8346644505858422,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.550318562917495,
+      "eval_loss": 0.6988471746444702,
+      "eval_mean_token_accuracy": 0.8132969897784544,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 89.2598,
+      "eval_samples_per_second": 15.909,
+      "eval_steps_per_second": 1.994,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5489993002265692,
+      "epoch": 3.567290283645142,
+      "grad_norm": 1.055821418762207,
+      "learning_rate": 0.00029643793741695687,
+      "loss": 0.5421150207519532,
+      "mean_token_accuracy": 0.8345710933208466,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5816821718818685,
+      "eval_loss": 0.6853668689727783,
+      "eval_mean_token_accuracy": 0.8142332360985574,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 89.0424,
+      "eval_samples_per_second": 15.947,
+      "eval_steps_per_second": 1.999,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5451365381479263,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.9027532935142517,
+      "learning_rate": 0.00029402850161352583,
+      "loss": 0.5463609218597412,
+      "mean_token_accuracy": 0.8369908526539802,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5344233209832331,
+      "eval_loss": 0.6794531345367432,
+      "eval_mean_token_accuracy": 0.8177759824843889,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 89.123,
+      "eval_samples_per_second": 15.933,
+      "eval_steps_per_second": 1.997,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5521467242389917,
+      "epoch": 3.663850331925166,
+      "grad_norm": 1.4606311321258545,
+      "learning_rate": 0.0002915874652643997,
+      "loss": 0.5421902179718018,
+      "mean_token_accuracy": 0.8339455373585224,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5807709094513668,
+      "eval_loss": 0.6949691772460938,
+      "eval_mean_token_accuracy": 0.8121141439743256,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.5437,
+      "eval_samples_per_second": 16.037,
+      "eval_steps_per_second": 2.01,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5455610822886229,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7269843220710754,
+      "learning_rate": 0.0002891155191539905,
+      "loss": 0.5407682895660401,
+      "mean_token_accuracy": 0.8393648222088814,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.6032205542151847,
+      "eval_loss": 0.6790974140167236,
+      "eval_mean_token_accuracy": 0.8165463112043531,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 89.517,
+      "eval_samples_per_second": 15.863,
+      "eval_steps_per_second": 1.988,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5474163968116045,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.8148866295814514,
+      "learning_rate": 0.00028661336281380717,
+      "loss": 0.5394441604614257,
+      "mean_token_accuracy": 0.8336028568446636,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5797032238392348,
+      "eval_loss": 0.670870304107666,
+      "eval_mean_token_accuracy": 0.8182691593518417,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 89.4468,
+      "eval_samples_per_second": 15.875,
+      "eval_steps_per_second": 1.99,
+      "step": 1560
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.5635917277434675e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1580/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1580/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1580/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1580/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1580/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..7ce4b072fcbe5b636867e13937c56fedee3b68b4
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1580/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.02728394274790723,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1580/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1580/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1580/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1580/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1580/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1580/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1580/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1580/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..c1375941d51c9a44bb7ae7cfb11a9ff8ab5ce42b
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1580/trainer_state.json
@@ -0,0 +1,1693 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 3.808690404345202,
+  "eval_steps": 20,
+  "global_step": 1580,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.024795526266098,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.945375919342041,
+      "learning_rate": 1.6698127428345936e-05,
+      "loss": 1.7825420379638672,
+      "mean_token_accuracy": 0.6315708436071873,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.29921497588747,
+      "eval_loss": 1.169131875038147,
+      "eval_mean_token_accuracy": 0.7185706342204233,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 91.7898,
+      "eval_samples_per_second": 15.47,
+      "eval_steps_per_second": 1.939,
+      "step": 20
+    },
+    {
+      "entropy": 1.0346344470977784,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.7916979789733887,
+      "learning_rate": 3.427510366871008e-05,
+      "loss": 0.9480395317077637,
+      "mean_token_accuracy": 0.7543429024517536,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 0.9673695430326997,
+      "eval_loss": 0.8588430285453796,
+      "eval_mean_token_accuracy": 0.7737233002534073,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 89.8219,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 40
+    },
+    {
+      "entropy": 0.8977566614747048,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7972187995910645,
+      "learning_rate": 5.185207990907422e-05,
+      "loss": 0.8029604911804199,
+      "mean_token_accuracy": 0.7818016350269318,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8436836982041263,
+      "eval_loss": 0.7893033623695374,
+      "eval_mean_token_accuracy": 0.7858914984076211,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 89.9349,
+      "eval_samples_per_second": 15.789,
+      "eval_steps_per_second": 1.979,
+      "step": 60
+    },
+    {
+      "entropy": 0.8650321319699288,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.2128729820251465,
+      "learning_rate": 6.942905614943836e-05,
+      "loss": 0.77417893409729,
+      "mean_token_accuracy": 0.7875479467213153,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8209857180547179,
+      "eval_loss": 0.7572280168533325,
+      "eval_mean_token_accuracy": 0.7925970383574453,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 89.9459,
+      "eval_samples_per_second": 15.787,
+      "eval_steps_per_second": 1.979,
+      "step": 80
+    },
+    {
+      "entropy": 0.8096660938113928,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.1254287958145142,
+      "learning_rate": 8.700603238980251e-05,
+      "loss": 0.7170331001281738,
+      "mean_token_accuracy": 0.7994555257260799,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.792780208788561,
+      "eval_loss": 0.7442984580993652,
+      "eval_mean_token_accuracy": 0.7954704573984896,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 90.294,
+      "eval_samples_per_second": 15.726,
+      "eval_steps_per_second": 1.971,
+      "step": 100
+    },
+    {
+      "entropy": 0.7979829199612141,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2140507698059082,
+      "learning_rate": 0.00010458300863016666,
+      "loss": 0.7141861915588379,
+      "mean_token_accuracy": 0.8000014744699001,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.7938692486018277,
+      "eval_loss": 0.7279797196388245,
+      "eval_mean_token_accuracy": 0.7989798734027348,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 89.7189,
+      "eval_samples_per_second": 15.827,
+      "eval_steps_per_second": 1.984,
+      "step": 120
+    },
+    {
+      "entropy": 0.7916347607970238,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.0506458282470703,
+      "learning_rate": 0.0001221599848705308,
+      "loss": 0.7117055416107178,
+      "mean_token_accuracy": 0.7970046654343605,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.8052781063519167,
+      "eval_loss": 0.722576916217804,
+      "eval_mean_token_accuracy": 0.7982530523551984,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 90.0376,
+      "eval_samples_per_second": 15.771,
+      "eval_steps_per_second": 1.977,
+      "step": 140
+    },
+    {
+      "entropy": 0.7880666613578796,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8409207463264465,
+      "learning_rate": 0.00013973696111089492,
+      "loss": 0.7052523136138916,
+      "mean_token_accuracy": 0.8000123649835587,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7787443134891853,
+      "eval_loss": 0.7137772440910339,
+      "eval_mean_token_accuracy": 0.8014209287220173,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 90.2562,
+      "eval_samples_per_second": 15.733,
+      "eval_steps_per_second": 1.972,
+      "step": 160
+    },
+    {
+      "entropy": 0.7696007996797561,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.0182477235794067,
+      "learning_rate": 0.00015731393735125907,
+      "loss": 0.7008797645568847,
+      "mean_token_accuracy": 0.8007259473204613,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.752259682068664,
+      "eval_loss": 0.7171286344528198,
+      "eval_mean_token_accuracy": 0.80052122574174,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 90.0797,
+      "eval_samples_per_second": 15.764,
+      "eval_steps_per_second": 1.976,
+      "step": 180
+    },
+    {
+      "entropy": 0.7828126326203346,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9553632736206055,
+      "learning_rate": 0.0001748909135916232,
+      "loss": 0.7046853542327881,
+      "mean_token_accuracy": 0.7988562889397144,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7389638169427936,
+      "eval_loss": 0.7187935709953308,
+      "eval_mean_token_accuracy": 0.7995899999409579,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 89.9223,
+      "eval_samples_per_second": 15.791,
+      "eval_steps_per_second": 1.979,
+      "step": 200
+    },
+    {
+      "entropy": 0.7909286297857762,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 1.134459376335144,
+      "learning_rate": 0.00019246788983198735,
+      "loss": 0.6989312171936035,
+      "mean_token_accuracy": 0.7991565138101577,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7496965199374082,
+      "eval_loss": 0.7106321454048157,
+      "eval_mean_token_accuracy": 0.8046183442131857,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 89.6899,
+      "eval_samples_per_second": 15.832,
+      "eval_steps_per_second": 1.985,
+      "step": 220
+    },
+    {
+      "entropy": 0.7651963323354721,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.9557585716247559,
+      "learning_rate": 0.0002100448660723515,
+      "loss": 0.6984559059143066,
+      "mean_token_accuracy": 0.8019823372364044,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7615563079212488,
+      "eval_loss": 0.7098539471626282,
+      "eval_mean_token_accuracy": 0.8047233508544022,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 90.3752,
+      "eval_samples_per_second": 15.712,
+      "eval_steps_per_second": 1.97,
+      "step": 240
+    },
+    {
+      "entropy": 0.787507726252079,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.9279311299324036,
+      "learning_rate": 0.00022762184231271568,
+      "loss": 0.708138370513916,
+      "mean_token_accuracy": 0.8013308703899383,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7856640896100676,
+      "eval_loss": 0.7191570401191711,
+      "eval_mean_token_accuracy": 0.8024677249153008,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 93.7336,
+      "eval_samples_per_second": 15.149,
+      "eval_steps_per_second": 1.899,
+      "step": 260
+    },
+    {
+      "entropy": 0.768489234894514,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.7992776036262512,
+      "learning_rate": 0.0002451988185530798,
+      "loss": 0.7035849571228028,
+      "mean_token_accuracy": 0.8031105332076549,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7682498740346244,
+      "eval_loss": 0.7149022221565247,
+      "eval_mean_token_accuracy": 0.7973497995499814,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 89.7309,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 280
+    },
+    {
+      "entropy": 0.7720955617725849,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.1931514739990234,
+      "learning_rate": 0.0002627757947934439,
+      "loss": 0.6965402603149414,
+      "mean_token_accuracy": 0.8040348663926125,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7214708304807042,
+      "eval_loss": 0.7350823879241943,
+      "eval_mean_token_accuracy": 0.7942911925610532,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 89.6451,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 300
+    },
+    {
+      "entropy": 0.7930095426738262,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8994467258453369,
+      "learning_rate": 0.0002803527710338081,
+      "loss": 0.7331492900848389,
+      "mean_token_accuracy": 0.796185302734375,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.8142535488926963,
+      "eval_loss": 0.7237834334373474,
+      "eval_mean_token_accuracy": 0.799495273761535,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 90.6759,
+      "eval_samples_per_second": 15.66,
+      "eval_steps_per_second": 1.963,
+      "step": 320
+    },
+    {
+      "entropy": 0.7825102441012859,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 1.8224239349365234,
+      "learning_rate": 0.00029792974727417223,
+      "loss": 0.7331175804138184,
+      "mean_token_accuracy": 0.7977667152881622,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7327135443017724,
+      "eval_loss": 0.7388784289360046,
+      "eval_mean_token_accuracy": 0.8007041252730938,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 89.7692,
+      "eval_samples_per_second": 15.818,
+      "eval_steps_per_second": 1.983,
+      "step": 340
+    },
+    {
+      "entropy": 0.819331557303667,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 1.0407779216766357,
+      "learning_rate": 0.0003155067235145364,
+      "loss": 0.7456695556640625,
+      "mean_token_accuracy": 0.7936980701982975,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.8065585420372781,
+      "eval_loss": 0.735519528388977,
+      "eval_mean_token_accuracy": 0.7974206296245704,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 89.4336,
+      "eval_samples_per_second": 15.878,
+      "eval_steps_per_second": 1.99,
+      "step": 360
+    },
+    {
+      "entropy": 0.8114653021097183,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.1209490299224854,
+      "learning_rate": 0.0003330836997549005,
+      "loss": 0.7340899467468261,
+      "mean_token_accuracy": 0.7914272703230381,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.8025527285056168,
+      "eval_loss": 0.7295576930046082,
+      "eval_mean_token_accuracy": 0.7990576628218876,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 89.9579,
+      "eval_samples_per_second": 15.785,
+      "eval_steps_per_second": 1.979,
+      "step": 380
+    },
+    {
+      "entropy": 0.8042497783899307,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 1.2437798976898193,
+      "learning_rate": 0.0003506606759952647,
+      "loss": 0.7381744384765625,
+      "mean_token_accuracy": 0.7967362694442273,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7717011336530193,
+      "eval_loss": 0.7436173558235168,
+      "eval_mean_token_accuracy": 0.7975849482450592,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 89.8411,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 400
+    },
+    {
+      "entropy": 0.7987750010056929,
+      "epoch": 1.012070006035003,
+      "grad_norm": 1.2390918731689453,
+      "learning_rate": 0.000364721224843344,
+      "loss": 0.7317790031433106,
+      "mean_token_accuracy": 0.7986087551364651,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7333170604170038,
+      "eval_loss": 0.7456948757171631,
+      "eval_mean_token_accuracy": 0.7964897577682238,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 89.7288,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 420
+    },
+    {
+      "entropy": 0.7372976846992969,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.1552740335464478,
+      "learning_rate": 0.00036468510102269575,
+      "loss": 0.6734249114990234,
+      "mean_token_accuracy": 0.805834824591875,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.7130355526891987,
+      "eval_loss": 0.7555333375930786,
+      "eval_mean_token_accuracy": 0.7961652781186479,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 89.8713,
+      "eval_samples_per_second": 15.8,
+      "eval_steps_per_second": 1.981,
+      "step": 440
+    },
+    {
+      "entropy": 0.7465668715536594,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 1.1991469860076904,
+      "learning_rate": 0.0003645973816745066,
+      "loss": 0.6871879577636719,
+      "mean_token_accuracy": 0.8042450234293937,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7772979542110743,
+      "eval_loss": 0.7561826705932617,
+      "eval_mean_token_accuracy": 0.7953876427720102,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 89.8395,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 460
+    },
+    {
+      "entropy": 0.7477855801582336,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 1.325486660003662,
+      "learning_rate": 0.00036445809162231435,
+      "loss": 0.6836830139160156,
+      "mean_token_accuracy": 0.8021619468927383,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.7098110676481483,
+      "eval_loss": 0.758007287979126,
+      "eval_mean_token_accuracy": 0.7909953189030122,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 89.6101,
+      "eval_samples_per_second": 15.846,
+      "eval_steps_per_second": 1.986,
+      "step": 480
+    },
+    {
+      "entropy": 0.7632691070437432,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 1.0122262239456177,
+      "learning_rate": 0.0003642672702835562,
+      "loss": 0.7071213245391845,
+      "mean_token_accuracy": 0.8000680930912495,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7645382720432924,
+      "eval_loss": 0.7492606043815613,
+      "eval_mean_token_accuracy": 0.7976858676149604,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 89.6092,
+      "eval_samples_per_second": 15.847,
+      "eval_steps_per_second": 1.986,
+      "step": 500
+    },
+    {
+      "entropy": 0.7623726457357407,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.367774486541748,
+      "learning_rate": 0.00036402497165841384,
+      "loss": 0.6965175151824952,
+      "mean_token_accuracy": 0.8016501650214195,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7868083715438843,
+      "eval_loss": 0.7481449842453003,
+      "eval_mean_token_accuracy": 0.799330582779445,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 89.8241,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 520
+    },
+    {
+      "entropy": 0.7535207405686378,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 1.213614821434021,
+      "learning_rate": 0.00036373126431453207,
+      "loss": 0.7078310489654541,
+      "mean_token_accuracy": 0.7998907208442688,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7829091950748743,
+      "eval_loss": 0.7416072487831116,
+      "eval_mean_token_accuracy": 0.7969146250339036,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 89.4607,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 540
+    },
+    {
+      "entropy": 0.7527363449335098,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 1.0929982662200928,
+      "learning_rate": 0.00036338623136761495,
+      "loss": 0.7058369636535644,
+      "mean_token_accuracy": 0.7997087433934211,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.728419938114252,
+      "eval_loss": 0.744873046875,
+      "eval_mean_token_accuracy": 0.7966634296299366,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 89.6597,
+      "eval_samples_per_second": 15.838,
+      "eval_steps_per_second": 1.985,
+      "step": 560
+    },
+    {
+      "entropy": 0.7406099259853363,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.409364104270935,
+      "learning_rate": 0.00036298997045790515,
+      "loss": 0.6902852058410645,
+      "mean_token_accuracy": 0.805386807769537,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.6991632758231645,
+      "eval_loss": 0.7480319738388062,
+      "eval_mean_token_accuracy": 0.7970068006033308,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 89.8329,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 580
+    },
+    {
+      "entropy": 0.7459449704736472,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.7053111791610718,
+      "learning_rate": 0.00036254259372255277,
+      "loss": 0.7044640064239502,
+      "mean_token_accuracy": 0.8005018755793571,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7454566660891758,
+      "eval_loss": 0.7551484704017639,
+      "eval_mean_token_accuracy": 0.7897093731365846,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 89.8313,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 600
+    },
+    {
+      "entropy": 0.7621391348540782,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 1.174443244934082,
+      "learning_rate": 0.000362044227763882,
+      "loss": 0.7111573696136475,
+      "mean_token_accuracy": 0.798752411454916,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7664387949397055,
+      "eval_loss": 0.7365804314613342,
+      "eval_mean_token_accuracy": 0.7963101200843126,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 89.9623,
+      "eval_samples_per_second": 15.784,
+      "eval_steps_per_second": 1.979,
+      "step": 620
+    },
+    {
+      "entropy": 0.7424944408237935,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.2898184061050415,
+      "learning_rate": 0.000361495013613564,
+      "loss": 0.6913110733032226,
+      "mean_token_accuracy": 0.8009015507996082,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7255479529332579,
+      "eval_loss": 0.7345991730690002,
+      "eval_mean_token_accuracy": 0.8013592945056015,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 89.817,
+      "eval_samples_per_second": 15.81,
+      "eval_steps_per_second": 1.982,
+      "step": 640
+    },
+    {
+      "entropy": 0.7303367637097835,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.1310476064682007,
+      "learning_rate": 0.00036089510669270666,
+      "loss": 0.6875617027282714,
+      "mean_token_accuracy": 0.8035397171974182,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7101308248016271,
+      "eval_loss": 0.7394917011260986,
+      "eval_mean_token_accuracy": 0.7990792942850777,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 89.7431,
+      "eval_samples_per_second": 15.823,
+      "eval_steps_per_second": 1.983,
+      "step": 660
+    },
+    {
+      "entropy": 0.7387678384780884,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.9645389914512634,
+      "learning_rate": 0.0003602446767678725,
+      "loss": 0.6911204338073731,
+      "mean_token_accuracy": 0.8056333385407924,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7281073737010527,
+      "eval_loss": 0.7376708984375,
+      "eval_mean_token_accuracy": 0.7997356415464637,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 90.0338,
+      "eval_samples_per_second": 15.772,
+      "eval_steps_per_second": 1.977,
+      "step": 680
+    },
+    {
+      "entropy": 0.7449554048478604,
+      "epoch": 1.687990343995172,
+      "grad_norm": 1.9069629907608032,
+      "learning_rate": 0.0003595439079030364,
+      "loss": 0.7042286396026611,
+      "mean_token_accuracy": 0.8018639378249646,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7022633817088738,
+      "eval_loss": 0.7320939898490906,
+      "eval_mean_token_accuracy": 0.8006051541044471,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 89.8948,
+      "eval_samples_per_second": 15.796,
+      "eval_steps_per_second": 1.98,
+      "step": 700
+    },
+    {
+      "entropy": 0.723249051719904,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.8333877325057983,
+      "learning_rate": 0.00035879299840749777,
+      "loss": 0.7043714046478271,
+      "mean_token_accuracy": 0.8028345607221127,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.672436946898364,
+      "eval_loss": 0.713233470916748,
+      "eval_mean_token_accuracy": 0.8047760577684038,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 89.6394,
+      "eval_samples_per_second": 15.841,
+      "eval_steps_per_second": 1.986,
+      "step": 720
+    },
+    {
+      "entropy": 0.730014206469059,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.4580037593841553,
+      "learning_rate": 0.00035799216077976137,
+      "loss": 0.6985176563262939,
+      "mean_token_accuracy": 0.8015547141432762,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.6908873795123582,
+      "eval_loss": 0.7198202013969421,
+      "eval_mean_token_accuracy": 0.8024131682481659,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 89.3771,
+      "eval_samples_per_second": 15.888,
+      "eval_steps_per_second": 1.992,
+      "step": 740
+    },
+    {
+      "entropy": 0.7219605796039105,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.51682710647583,
+      "learning_rate": 0.000357141621647403,
+      "loss": 0.6844424724578857,
+      "mean_token_accuracy": 0.8032813094556331,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7113401109582922,
+      "eval_loss": 0.7195309400558472,
+      "eval_mean_token_accuracy": 0.8019187035185568,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 89.4658,
+      "eval_samples_per_second": 15.872,
+      "eval_steps_per_second": 1.99,
+      "step": 760
+    },
+    {
+      "entropy": 0.729826743900776,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 1.0838000774383545,
+      "learning_rate": 0.0003562416217029361,
+      "loss": 0.685858678817749,
+      "mean_token_accuracy": 0.8050567395985126,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.649370267484965,
+      "eval_loss": 0.7177029252052307,
+      "eval_mean_token_accuracy": 0.8050174043419656,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 89.2287,
+      "eval_samples_per_second": 15.914,
+      "eval_steps_per_second": 1.995,
+      "step": 780
+    },
+    {
+      "entropy": 0.6799364153295755,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 1.1364924907684326,
+      "learning_rate": 0.00035529241563569903,
+      "loss": 0.6714428901672364,
+      "mean_token_accuracy": 0.8087341658771038,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6603799047095052,
+      "eval_loss": 0.7201809287071228,
+      "eval_mean_token_accuracy": 0.802863868769635,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 89.2718,
+      "eval_samples_per_second": 15.906,
+      "eval_steps_per_second": 1.994,
+      "step": 800
+    },
+    {
+      "entropy": 0.7078610375523567,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 1.2191942930221558,
+      "learning_rate": 0.0003542942720597808,
+      "loss": 0.6996590614318847,
+      "mean_token_accuracy": 0.8049512051045895,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.7240545264120852,
+      "eval_loss": 0.7301309108734131,
+      "eval_mean_token_accuracy": 0.8015973166133581,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 89.3438,
+      "eval_samples_per_second": 15.894,
+      "eval_steps_per_second": 1.992,
+      "step": 820
+    },
+    {
+      "entropy": 0.6557391556826505,
+      "epoch": 2.024140012070006,
+      "grad_norm": 1.36203134059906,
+      "learning_rate": 0.0003532474734380064,
+      "loss": 0.6327113628387451,
+      "mean_token_accuracy": 0.8142829847026181,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6573808648613062,
+      "eval_loss": 0.7334365844726562,
+      "eval_mean_token_accuracy": 0.8043633893634496,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 89.3359,
+      "eval_samples_per_second": 15.895,
+      "eval_steps_per_second": 1.992,
+      "step": 840
+    },
+    {
+      "entropy": 0.6409128502011299,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 1.5804955959320068,
+      "learning_rate": 0.0003521523160020035,
+      "loss": 0.6023559093475341,
+      "mean_token_accuracy": 0.8222251623868942,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6983234788594621,
+      "eval_loss": 0.7232135534286499,
+      "eval_mean_token_accuracy": 0.8010654268639811,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 89.2615,
+      "eval_samples_per_second": 15.908,
+      "eval_steps_per_second": 1.994,
+      "step": 860
+    },
+    {
+      "entropy": 0.6151149723678827,
+      "epoch": 2.12070006035003,
+      "grad_norm": 1.1975436210632324,
+      "learning_rate": 0.00035100910966837193,
+      "loss": 0.5803564548492431,
+      "mean_token_accuracy": 0.8256325736641884,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6636555573243773,
+      "eval_loss": 0.7205662727355957,
+      "eval_mean_token_accuracy": 0.8036768369460374,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 89.2957,
+      "eval_samples_per_second": 15.902,
+      "eval_steps_per_second": 1.993,
+      "step": 880
+    },
+    {
+      "entropy": 0.6201727617532015,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9907131195068359,
+      "learning_rate": 0.0003498181779509813,
+      "loss": 0.6101872444152832,
+      "mean_token_accuracy": 0.8196512959897518,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.6473788086617931,
+      "eval_loss": 0.7062397003173828,
+      "eval_mean_token_accuracy": 0.8078362184963869,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 88.6198,
+      "eval_samples_per_second": 16.024,
+      "eval_steps_per_second": 2.009,
+      "step": 900
+    },
+    {
+      "entropy": 0.6097332935780286,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.9055633544921875,
+      "learning_rate": 0.00034857985786942026,
+      "loss": 0.5956094264984131,
+      "mean_token_accuracy": 0.8240575887262821,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.612629868341296,
+      "eval_loss": 0.7106770277023315,
+      "eval_mean_token_accuracy": 0.805324277181304,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 89.4383,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 920
+    },
+    {
+      "entropy": 0.6182701248675585,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.9513736367225647,
+      "learning_rate": 0.00034729449985362404,
+      "loss": 0.6019269466400147,
+      "mean_token_accuracy": 0.8187106013298034,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6437647456533453,
+      "eval_loss": 0.7173502445220947,
+      "eval_mean_token_accuracy": 0.8056439573175451,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.9568,
+      "eval_samples_per_second": 15.963,
+      "eval_steps_per_second": 2.001,
+      "step": 940
+    },
+    {
+      "entropy": 0.6508570514619351,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.9606207013130188,
+      "learning_rate": 0.0003459624676447067,
+      "loss": 0.6296409606933594,
+      "mean_token_accuracy": 0.8167732007801533,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6288247135248077,
+      "eval_loss": 0.7160675525665283,
+      "eval_mean_token_accuracy": 0.8066010689467527,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 89.0044,
+      "eval_samples_per_second": 15.954,
+      "eval_steps_per_second": 2.0,
+      "step": 960
+    },
+    {
+      "entropy": 0.6380081418901682,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.9677181839942932,
+      "learning_rate": 0.000344584138192027,
+      "loss": 0.6210546493530273,
+      "mean_token_accuracy": 0.8186135642230511,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6407805031604981,
+      "eval_loss": 0.7055043578147888,
+      "eval_mean_token_accuracy": 0.8070103157772107,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 89.5606,
+      "eval_samples_per_second": 15.855,
+      "eval_steps_per_second": 1.987,
+      "step": 980
+    },
+    {
+      "entropy": 0.6310251638293266,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 2.1046407222747803,
+      "learning_rate": 0.000343159901546516,
+      "loss": 0.6165830135345459,
+      "mean_token_accuracy": 0.8207855716347694,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6461507470420237,
+      "eval_loss": 0.7292136549949646,
+      "eval_mean_token_accuracy": 0.8030418261383356,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 89.315,
+      "eval_samples_per_second": 15.899,
+      "eval_steps_per_second": 1.993,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6256617344915867,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 1.1768349409103394,
+      "learning_rate": 0.0003416901607502972,
+      "loss": 0.6142709255218506,
+      "mean_token_accuracy": 0.8220669947564602,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6150523993406403,
+      "eval_loss": 0.7019563317298889,
+      "eval_mean_token_accuracy": 0.8094323099998946,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 89.419,
+      "eval_samples_per_second": 15.88,
+      "eval_steps_per_second": 1.991,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6126735735684633,
+      "epoch": 2.506940253470127,
+      "grad_norm": 1.8065496683120728,
+      "learning_rate": 0.00034017533172263055,
+      "loss": 0.6056458473205566,
+      "mean_token_accuracy": 0.8203762136399746,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6544345456562685,
+      "eval_loss": 0.7086517214775085,
+      "eval_mean_token_accuracy": 0.806709805901131,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 88.981,
+      "eval_samples_per_second": 15.958,
+      "eval_steps_per_second": 2.0,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6311754353344441,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 1.0576369762420654,
+      "learning_rate": 0.00033861584314221236,
+      "loss": 0.6176392555236816,
+      "mean_token_accuracy": 0.8177410490810871,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6273799200406235,
+      "eval_loss": 0.708163321018219,
+      "eval_mean_token_accuracy": 0.8032017099053672,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 89.7668,
+      "eval_samples_per_second": 15.819,
+      "eval_steps_per_second": 1.983,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6200249589979648,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.997065007686615,
+      "learning_rate": 0.0003370121363258637,
+      "loss": 0.6203888893127442,
+      "mean_token_accuracy": 0.8196275025606156,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6457576811983344,
+      "eval_loss": 0.6972290277481079,
+      "eval_mean_token_accuracy": 0.8095980849158898,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 89.2027,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 1080
+    },
+    {
+      "entropy": 0.626810473203659,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.865475058555603,
+      "learning_rate": 0.0003353646651036438,
+      "loss": 0.6102585792541504,
+      "mean_token_accuracy": 0.819708751142025,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.6412022431914726,
+      "eval_loss": 0.7036139965057373,
+      "eval_mean_token_accuracy": 0.8070317135098275,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 89.4733,
+      "eval_samples_per_second": 15.871,
+      "eval_steps_per_second": 1.989,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6581619590520859,
+      "epoch": 2.700060350030175,
+      "grad_norm": 1.00307035446167,
+      "learning_rate": 0.00033367389569042064,
+      "loss": 0.6397720336914062,
+      "mean_token_accuracy": 0.81413309648633,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6288736402318719,
+      "eval_loss": 0.6966825723648071,
+      "eval_mean_token_accuracy": 0.8094039669867312,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 88.7341,
+      "eval_samples_per_second": 16.003,
+      "eval_steps_per_second": 2.006,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6399412982165813,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.8589329719543457,
+      "learning_rate": 0.0003319403065539384,
+      "loss": 0.6347338199615479,
+      "mean_token_accuracy": 0.8139534957706929,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.683491913455256,
+      "eval_loss": 0.6984149217605591,
+      "eval_mean_token_accuracy": 0.8083989469522841,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 89.3929,
+      "eval_samples_per_second": 15.885,
+      "eval_steps_per_second": 1.991,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6264835461974144,
+      "epoch": 2.796620398310199,
+      "grad_norm": 0.7807853817939758,
+      "learning_rate": 0.0003301643882794163,
+      "loss": 0.6162629127502441,
+      "mean_token_accuracy": 0.8216292977333068,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6578792159476977,
+      "eval_loss": 0.6921441555023193,
+      "eval_mean_token_accuracy": 0.8094299177775223,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 89.2356,
+      "eval_samples_per_second": 15.913,
+      "eval_steps_per_second": 1.995,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6269129011780024,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 1.1675593852996826,
+      "learning_rate": 0.0003283466434307189,
+      "loss": 0.6235532760620117,
+      "mean_token_accuracy": 0.8188897788524627,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6533248447970058,
+      "eval_loss": 0.696849524974823,
+      "eval_mean_token_accuracy": 0.8102703713968898,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2602,
+      "eval_samples_per_second": 16.089,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.6259873781353236,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.9596486687660217,
+      "learning_rate": 0.00032648758640813655,
+      "loss": 0.6118728160858155,
+      "mean_token_accuracy": 0.8171859130263328,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6212078754821521,
+      "eval_loss": 0.6942870020866394,
+      "eval_mean_token_accuracy": 0.810484343365337,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 89.4375,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 1200
+    },
+    {
+      "entropy": 0.6142537571489811,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 0.9625058770179749,
+      "learning_rate": 0.00032458774330281615,
+      "loss": 0.6060261249542236,
+      "mean_token_accuracy": 0.8223113484680653,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6553665667437436,
+      "eval_loss": 0.6835269331932068,
+      "eval_mean_token_accuracy": 0.8134723968720168,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 89.5092,
+      "eval_samples_per_second": 15.864,
+      "eval_steps_per_second": 1.989,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6301972426474094,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.9657333493232727,
+      "learning_rate": 0.0003226476517478835,
+      "loss": 0.6277759075164795,
+      "mean_token_accuracy": 0.8164266042411328,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6594253118788258,
+      "eval_loss": 0.6820935606956482,
+      "eval_mean_token_accuracy": 0.8132368198941263,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 89.6211,
+      "eval_samples_per_second": 15.844,
+      "eval_steps_per_second": 1.986,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5346100655469027,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8541185259819031,
+      "learning_rate": 0.0003206678607662996,
+      "loss": 0.5261692047119141,
+      "mean_token_accuracy": 0.8382289425119177,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5626645662476507,
+      "eval_loss": 0.7028542160987854,
+      "eval_mean_token_accuracy": 0.8133395572056931,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 89.1267,
+      "eval_samples_per_second": 15.932,
+      "eval_steps_per_second": 1.997,
+      "step": 1260
+    },
+    {
+      "entropy": 0.5001101832836866,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.8208815455436707,
+      "learning_rate": 0.0003186489306154935,
+      "loss": 0.4850132942199707,
+      "mean_token_accuracy": 0.8468983843922615,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5889462046743779,
+      "eval_loss": 0.691015899181366,
+      "eval_mean_token_accuracy": 0.8135226466012805,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.9677,
+      "eval_samples_per_second": 15.961,
+      "eval_steps_per_second": 2.001,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5242766339331866,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 0.7893072962760925,
+      "learning_rate": 0.0003165914326288163,
+      "loss": 0.5061577320098877,
+      "mean_token_accuracy": 0.8425608821213245,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5543691962957382,
+      "eval_loss": 0.6913734078407288,
+      "eval_mean_token_accuracy": 0.8154889680026622,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 89.6963,
+      "eval_samples_per_second": 15.831,
+      "eval_steps_per_second": 1.984,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5327721010893584,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.8992080092430115,
+      "learning_rate": 0.0003144959490538604,
+      "loss": 0.5068036556243897,
+      "mean_token_accuracy": 0.8420542575418949,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5734540803378887,
+      "eval_loss": 0.6948716044425964,
+      "eval_mean_token_accuracy": 0.8137217944257715,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 89.4597,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5247377116233111,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.9291845560073853,
+      "learning_rate": 0.0003123630728876902,
+      "loss": 0.5023272037506104,
+      "mean_token_accuracy": 0.8435182586312294,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5492362935891312,
+      "eval_loss": 0.6980633735656738,
+      "eval_mean_token_accuracy": 0.8155082919624415,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 89.2124,
+      "eval_samples_per_second": 15.917,
+      "eval_steps_per_second": 1.995,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5380499072372913,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 1.0477524995803833,
+      "learning_rate": 0.00031019340770903136,
+      "loss": 0.5264591693878173,
+      "mean_token_accuracy": 0.8391959741711617,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5896503307511297,
+      "eval_loss": 0.6997144222259521,
+      "eval_mean_token_accuracy": 0.8124373062942805,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 89.1147,
+      "eval_samples_per_second": 15.935,
+      "eval_steps_per_second": 1.997,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5460653610527515,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 1.2473069429397583,
+      "learning_rate": 0.00030798756750746476,
+      "loss": 0.5319677829742432,
+      "mean_token_accuracy": 0.8364320226013661,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5659052182114526,
+      "eval_loss": 0.6924049258232117,
+      "eval_mean_token_accuracy": 0.8153053585732921,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 89.6468,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5367407951503992,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 1.2679712772369385,
+      "learning_rate": 0.00030574617650967485,
+      "loss": 0.5190927028656006,
+      "mean_token_accuracy": 0.8399469174444676,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5581759144081159,
+      "eval_loss": 0.6933317184448242,
+      "eval_mean_token_accuracy": 0.8139757862251796,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 89.0749,
+      "eval_samples_per_second": 15.942,
+      "eval_steps_per_second": 1.998,
+      "step": 1400
+    },
+    {
+      "entropy": 0.530736118927598,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.9717612862586975,
+      "learning_rate": 0.0003034698690028009,
+      "loss": 0.512363052368164,
+      "mean_token_accuracy": 0.8419127956032753,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5233335836549823,
+      "eval_loss": 0.6958550214767456,
+      "eval_mean_token_accuracy": 0.8158778501360604,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 88.8664,
+      "eval_samples_per_second": 15.979,
+      "eval_steps_per_second": 2.003,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5327631575986743,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7649208903312683,
+      "learning_rate": 0.00030115928915494116,
+      "loss": 0.5202179431915284,
+      "mean_token_accuracy": 0.8384683802723885,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.5960972657364406,
+      "eval_loss": 0.6984783411026001,
+      "eval_mean_token_accuracy": 0.8095930597085631,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 89.5168,
+      "eval_samples_per_second": 15.863,
+      "eval_steps_per_second": 1.988,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5428155034780502,
+      "epoch": 3.51901025950513,
+      "grad_norm": 1.2790874242782593,
+      "learning_rate": 0.00029881509083286116,
+      "loss": 0.524717378616333,
+      "mean_token_accuracy": 0.8346644505858422,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.550318562917495,
+      "eval_loss": 0.6988471746444702,
+      "eval_mean_token_accuracy": 0.8132969897784544,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 89.2598,
+      "eval_samples_per_second": 15.909,
+      "eval_steps_per_second": 1.994,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5489993002265692,
+      "epoch": 3.567290283645142,
+      "grad_norm": 1.055821418762207,
+      "learning_rate": 0.00029643793741695687,
+      "loss": 0.5421150207519532,
+      "mean_token_accuracy": 0.8345710933208466,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5816821718818685,
+      "eval_loss": 0.6853668689727783,
+      "eval_mean_token_accuracy": 0.8142332360985574,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 89.0424,
+      "eval_samples_per_second": 15.947,
+      "eval_steps_per_second": 1.999,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5451365381479263,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.9027532935142517,
+      "learning_rate": 0.00029402850161352583,
+      "loss": 0.5463609218597412,
+      "mean_token_accuracy": 0.8369908526539802,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5344233209832331,
+      "eval_loss": 0.6794531345367432,
+      "eval_mean_token_accuracy": 0.8177759824843889,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 89.123,
+      "eval_samples_per_second": 15.933,
+      "eval_steps_per_second": 1.997,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5521467242389917,
+      "epoch": 3.663850331925166,
+      "grad_norm": 1.4606311321258545,
+      "learning_rate": 0.0002915874652643997,
+      "loss": 0.5421902179718018,
+      "mean_token_accuracy": 0.8339455373585224,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5807709094513668,
+      "eval_loss": 0.6949691772460938,
+      "eval_mean_token_accuracy": 0.8121141439743256,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.5437,
+      "eval_samples_per_second": 16.037,
+      "eval_steps_per_second": 2.01,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5455610822886229,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7269843220710754,
+      "learning_rate": 0.0002891155191539905,
+      "loss": 0.5407682895660401,
+      "mean_token_accuracy": 0.8393648222088814,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.6032205542151847,
+      "eval_loss": 0.6790974140167236,
+      "eval_mean_token_accuracy": 0.8165463112043531,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 89.517,
+      "eval_samples_per_second": 15.863,
+      "eval_steps_per_second": 1.988,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5474163968116045,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.8148866295814514,
+      "learning_rate": 0.00028661336281380717,
+      "loss": 0.5394441604614257,
+      "mean_token_accuracy": 0.8336028568446636,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5797032238392348,
+      "eval_loss": 0.670870304107666,
+      "eval_mean_token_accuracy": 0.8182691593518417,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 89.4468,
+      "eval_samples_per_second": 15.875,
+      "eval_steps_per_second": 1.99,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5511750865727663,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.3441656827926636,
+      "learning_rate": 0.0002840817043244964,
+      "loss": 0.5378933906555176,
+      "mean_token_accuracy": 0.8349151819944381,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5297858750217417,
+      "eval_loss": 0.680833637714386,
+      "eval_mean_token_accuracy": 0.8180924275617921,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 89.5417,
+      "eval_samples_per_second": 15.859,
+      "eval_steps_per_second": 1.988,
+      "step": 1580
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.58275059713237e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-160/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-160/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-160/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-160/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-160/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..7ce4b072fcbe5b636867e13937c56fedee3b68b4
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-160/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.02728394274790723,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-160/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-160/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-160/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-160/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-160/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-160/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-160/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-160/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..d04df3a93d23b659196202f272ca014cff2246a2
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-160/trainer_state.json
@@ -0,0 +1,202 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.38624019312009655,
+  "eval_steps": 20,
+  "global_step": 160,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.024795526266098,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.945375919342041,
+      "learning_rate": 1.6698127428345936e-05,
+      "loss": 1.7825420379638672,
+      "mean_token_accuracy": 0.6315708436071873,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.29921497588747,
+      "eval_loss": 1.169131875038147,
+      "eval_mean_token_accuracy": 0.7185706342204233,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 91.7898,
+      "eval_samples_per_second": 15.47,
+      "eval_steps_per_second": 1.939,
+      "step": 20
+    },
+    {
+      "entropy": 1.0346344470977784,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.7916979789733887,
+      "learning_rate": 3.427510366871008e-05,
+      "loss": 0.9480395317077637,
+      "mean_token_accuracy": 0.7543429024517536,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 0.9673695430326997,
+      "eval_loss": 0.8588430285453796,
+      "eval_mean_token_accuracy": 0.7737233002534073,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 89.8219,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 40
+    },
+    {
+      "entropy": 0.8977566614747048,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7972187995910645,
+      "learning_rate": 5.185207990907422e-05,
+      "loss": 0.8029604911804199,
+      "mean_token_accuracy": 0.7818016350269318,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8436836982041263,
+      "eval_loss": 0.7893033623695374,
+      "eval_mean_token_accuracy": 0.7858914984076211,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 89.9349,
+      "eval_samples_per_second": 15.789,
+      "eval_steps_per_second": 1.979,
+      "step": 60
+    },
+    {
+      "entropy": 0.8650321319699288,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.2128729820251465,
+      "learning_rate": 6.942905614943836e-05,
+      "loss": 0.77417893409729,
+      "mean_token_accuracy": 0.7875479467213153,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8209857180547179,
+      "eval_loss": 0.7572280168533325,
+      "eval_mean_token_accuracy": 0.7925970383574453,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 89.9459,
+      "eval_samples_per_second": 15.787,
+      "eval_steps_per_second": 1.979,
+      "step": 80
+    },
+    {
+      "entropy": 0.8096660938113928,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.1254287958145142,
+      "learning_rate": 8.700603238980251e-05,
+      "loss": 0.7170331001281738,
+      "mean_token_accuracy": 0.7994555257260799,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.792780208788561,
+      "eval_loss": 0.7442984580993652,
+      "eval_mean_token_accuracy": 0.7954704573984896,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 90.294,
+      "eval_samples_per_second": 15.726,
+      "eval_steps_per_second": 1.971,
+      "step": 100
+    },
+    {
+      "entropy": 0.7979829199612141,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2140507698059082,
+      "learning_rate": 0.00010458300863016666,
+      "loss": 0.7141861915588379,
+      "mean_token_accuracy": 0.8000014744699001,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.7938692486018277,
+      "eval_loss": 0.7279797196388245,
+      "eval_mean_token_accuracy": 0.7989798734027348,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 89.7189,
+      "eval_samples_per_second": 15.827,
+      "eval_steps_per_second": 1.984,
+      "step": 120
+    },
+    {
+      "entropy": 0.7916347607970238,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.0506458282470703,
+      "learning_rate": 0.0001221599848705308,
+      "loss": 0.7117055416107178,
+      "mean_token_accuracy": 0.7970046654343605,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.8052781063519167,
+      "eval_loss": 0.722576916217804,
+      "eval_mean_token_accuracy": 0.7982530523551984,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 90.0376,
+      "eval_samples_per_second": 15.771,
+      "eval_steps_per_second": 1.977,
+      "step": 140
+    },
+    {
+      "entropy": 0.7880666613578796,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8409207463264465,
+      "learning_rate": 0.00013973696111089492,
+      "loss": 0.7052523136138916,
+      "mean_token_accuracy": 0.8000123649835587,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7787443134891853,
+      "eval_loss": 0.7137772440910339,
+      "eval_mean_token_accuracy": 0.8014209287220173,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 90.2562,
+      "eval_samples_per_second": 15.733,
+      "eval_steps_per_second": 1.972,
+      "step": 160
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.6420618611449856e+16,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1600/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1600/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1600/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1600/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1600/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..7ce4b072fcbe5b636867e13937c56fedee3b68b4
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1600/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.02728394274790723,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1600/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1600/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1600/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1600/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1600/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1600/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1600/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1600/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..ed6fb502590ff22047fec334cfb36728afd7a77f
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1600/trainer_state.json
@@ -0,0 +1,1714 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 3.856970428485214,
+  "eval_steps": 20,
+  "global_step": 1600,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.024795526266098,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.945375919342041,
+      "learning_rate": 1.6698127428345936e-05,
+      "loss": 1.7825420379638672,
+      "mean_token_accuracy": 0.6315708436071873,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.29921497588747,
+      "eval_loss": 1.169131875038147,
+      "eval_mean_token_accuracy": 0.7185706342204233,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 91.7898,
+      "eval_samples_per_second": 15.47,
+      "eval_steps_per_second": 1.939,
+      "step": 20
+    },
+    {
+      "entropy": 1.0346344470977784,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.7916979789733887,
+      "learning_rate": 3.427510366871008e-05,
+      "loss": 0.9480395317077637,
+      "mean_token_accuracy": 0.7543429024517536,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 0.9673695430326997,
+      "eval_loss": 0.8588430285453796,
+      "eval_mean_token_accuracy": 0.7737233002534073,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 89.8219,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 40
+    },
+    {
+      "entropy": 0.8977566614747048,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7972187995910645,
+      "learning_rate": 5.185207990907422e-05,
+      "loss": 0.8029604911804199,
+      "mean_token_accuracy": 0.7818016350269318,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8436836982041263,
+      "eval_loss": 0.7893033623695374,
+      "eval_mean_token_accuracy": 0.7858914984076211,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 89.9349,
+      "eval_samples_per_second": 15.789,
+      "eval_steps_per_second": 1.979,
+      "step": 60
+    },
+    {
+      "entropy": 0.8650321319699288,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.2128729820251465,
+      "learning_rate": 6.942905614943836e-05,
+      "loss": 0.77417893409729,
+      "mean_token_accuracy": 0.7875479467213153,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8209857180547179,
+      "eval_loss": 0.7572280168533325,
+      "eval_mean_token_accuracy": 0.7925970383574453,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 89.9459,
+      "eval_samples_per_second": 15.787,
+      "eval_steps_per_second": 1.979,
+      "step": 80
+    },
+    {
+      "entropy": 0.8096660938113928,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.1254287958145142,
+      "learning_rate": 8.700603238980251e-05,
+      "loss": 0.7170331001281738,
+      "mean_token_accuracy": 0.7994555257260799,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.792780208788561,
+      "eval_loss": 0.7442984580993652,
+      "eval_mean_token_accuracy": 0.7954704573984896,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 90.294,
+      "eval_samples_per_second": 15.726,
+      "eval_steps_per_second": 1.971,
+      "step": 100
+    },
+    {
+      "entropy": 0.7979829199612141,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2140507698059082,
+      "learning_rate": 0.00010458300863016666,
+      "loss": 0.7141861915588379,
+      "mean_token_accuracy": 0.8000014744699001,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.7938692486018277,
+      "eval_loss": 0.7279797196388245,
+      "eval_mean_token_accuracy": 0.7989798734027348,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 89.7189,
+      "eval_samples_per_second": 15.827,
+      "eval_steps_per_second": 1.984,
+      "step": 120
+    },
+    {
+      "entropy": 0.7916347607970238,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.0506458282470703,
+      "learning_rate": 0.0001221599848705308,
+      "loss": 0.7117055416107178,
+      "mean_token_accuracy": 0.7970046654343605,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.8052781063519167,
+      "eval_loss": 0.722576916217804,
+      "eval_mean_token_accuracy": 0.7982530523551984,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 90.0376,
+      "eval_samples_per_second": 15.771,
+      "eval_steps_per_second": 1.977,
+      "step": 140
+    },
+    {
+      "entropy": 0.7880666613578796,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8409207463264465,
+      "learning_rate": 0.00013973696111089492,
+      "loss": 0.7052523136138916,
+      "mean_token_accuracy": 0.8000123649835587,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7787443134891853,
+      "eval_loss": 0.7137772440910339,
+      "eval_mean_token_accuracy": 0.8014209287220173,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 90.2562,
+      "eval_samples_per_second": 15.733,
+      "eval_steps_per_second": 1.972,
+      "step": 160
+    },
+    {
+      "entropy": 0.7696007996797561,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.0182477235794067,
+      "learning_rate": 0.00015731393735125907,
+      "loss": 0.7008797645568847,
+      "mean_token_accuracy": 0.8007259473204613,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.752259682068664,
+      "eval_loss": 0.7171286344528198,
+      "eval_mean_token_accuracy": 0.80052122574174,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 90.0797,
+      "eval_samples_per_second": 15.764,
+      "eval_steps_per_second": 1.976,
+      "step": 180
+    },
+    {
+      "entropy": 0.7828126326203346,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9553632736206055,
+      "learning_rate": 0.0001748909135916232,
+      "loss": 0.7046853542327881,
+      "mean_token_accuracy": 0.7988562889397144,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7389638169427936,
+      "eval_loss": 0.7187935709953308,
+      "eval_mean_token_accuracy": 0.7995899999409579,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 89.9223,
+      "eval_samples_per_second": 15.791,
+      "eval_steps_per_second": 1.979,
+      "step": 200
+    },
+    {
+      "entropy": 0.7909286297857762,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 1.134459376335144,
+      "learning_rate": 0.00019246788983198735,
+      "loss": 0.6989312171936035,
+      "mean_token_accuracy": 0.7991565138101577,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7496965199374082,
+      "eval_loss": 0.7106321454048157,
+      "eval_mean_token_accuracy": 0.8046183442131857,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 89.6899,
+      "eval_samples_per_second": 15.832,
+      "eval_steps_per_second": 1.985,
+      "step": 220
+    },
+    {
+      "entropy": 0.7651963323354721,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.9557585716247559,
+      "learning_rate": 0.0002100448660723515,
+      "loss": 0.6984559059143066,
+      "mean_token_accuracy": 0.8019823372364044,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7615563079212488,
+      "eval_loss": 0.7098539471626282,
+      "eval_mean_token_accuracy": 0.8047233508544022,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 90.3752,
+      "eval_samples_per_second": 15.712,
+      "eval_steps_per_second": 1.97,
+      "step": 240
+    },
+    {
+      "entropy": 0.787507726252079,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.9279311299324036,
+      "learning_rate": 0.00022762184231271568,
+      "loss": 0.708138370513916,
+      "mean_token_accuracy": 0.8013308703899383,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7856640896100676,
+      "eval_loss": 0.7191570401191711,
+      "eval_mean_token_accuracy": 0.8024677249153008,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 93.7336,
+      "eval_samples_per_second": 15.149,
+      "eval_steps_per_second": 1.899,
+      "step": 260
+    },
+    {
+      "entropy": 0.768489234894514,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.7992776036262512,
+      "learning_rate": 0.0002451988185530798,
+      "loss": 0.7035849571228028,
+      "mean_token_accuracy": 0.8031105332076549,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7682498740346244,
+      "eval_loss": 0.7149022221565247,
+      "eval_mean_token_accuracy": 0.7973497995499814,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 89.7309,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 280
+    },
+    {
+      "entropy": 0.7720955617725849,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.1931514739990234,
+      "learning_rate": 0.0002627757947934439,
+      "loss": 0.6965402603149414,
+      "mean_token_accuracy": 0.8040348663926125,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7214708304807042,
+      "eval_loss": 0.7350823879241943,
+      "eval_mean_token_accuracy": 0.7942911925610532,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 89.6451,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 300
+    },
+    {
+      "entropy": 0.7930095426738262,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8994467258453369,
+      "learning_rate": 0.0002803527710338081,
+      "loss": 0.7331492900848389,
+      "mean_token_accuracy": 0.796185302734375,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.8142535488926963,
+      "eval_loss": 0.7237834334373474,
+      "eval_mean_token_accuracy": 0.799495273761535,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 90.6759,
+      "eval_samples_per_second": 15.66,
+      "eval_steps_per_second": 1.963,
+      "step": 320
+    },
+    {
+      "entropy": 0.7825102441012859,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 1.8224239349365234,
+      "learning_rate": 0.00029792974727417223,
+      "loss": 0.7331175804138184,
+      "mean_token_accuracy": 0.7977667152881622,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7327135443017724,
+      "eval_loss": 0.7388784289360046,
+      "eval_mean_token_accuracy": 0.8007041252730938,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 89.7692,
+      "eval_samples_per_second": 15.818,
+      "eval_steps_per_second": 1.983,
+      "step": 340
+    },
+    {
+      "entropy": 0.819331557303667,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 1.0407779216766357,
+      "learning_rate": 0.0003155067235145364,
+      "loss": 0.7456695556640625,
+      "mean_token_accuracy": 0.7936980701982975,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.8065585420372781,
+      "eval_loss": 0.735519528388977,
+      "eval_mean_token_accuracy": 0.7974206296245704,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 89.4336,
+      "eval_samples_per_second": 15.878,
+      "eval_steps_per_second": 1.99,
+      "step": 360
+    },
+    {
+      "entropy": 0.8114653021097183,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.1209490299224854,
+      "learning_rate": 0.0003330836997549005,
+      "loss": 0.7340899467468261,
+      "mean_token_accuracy": 0.7914272703230381,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.8025527285056168,
+      "eval_loss": 0.7295576930046082,
+      "eval_mean_token_accuracy": 0.7990576628218876,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 89.9579,
+      "eval_samples_per_second": 15.785,
+      "eval_steps_per_second": 1.979,
+      "step": 380
+    },
+    {
+      "entropy": 0.8042497783899307,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 1.2437798976898193,
+      "learning_rate": 0.0003506606759952647,
+      "loss": 0.7381744384765625,
+      "mean_token_accuracy": 0.7967362694442273,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7717011336530193,
+      "eval_loss": 0.7436173558235168,
+      "eval_mean_token_accuracy": 0.7975849482450592,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 89.8411,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 400
+    },
+    {
+      "entropy": 0.7987750010056929,
+      "epoch": 1.012070006035003,
+      "grad_norm": 1.2390918731689453,
+      "learning_rate": 0.000364721224843344,
+      "loss": 0.7317790031433106,
+      "mean_token_accuracy": 0.7986087551364651,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7333170604170038,
+      "eval_loss": 0.7456948757171631,
+      "eval_mean_token_accuracy": 0.7964897577682238,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 89.7288,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 420
+    },
+    {
+      "entropy": 0.7372976846992969,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.1552740335464478,
+      "learning_rate": 0.00036468510102269575,
+      "loss": 0.6734249114990234,
+      "mean_token_accuracy": 0.805834824591875,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.7130355526891987,
+      "eval_loss": 0.7555333375930786,
+      "eval_mean_token_accuracy": 0.7961652781186479,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 89.8713,
+      "eval_samples_per_second": 15.8,
+      "eval_steps_per_second": 1.981,
+      "step": 440
+    },
+    {
+      "entropy": 0.7465668715536594,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 1.1991469860076904,
+      "learning_rate": 0.0003645973816745066,
+      "loss": 0.6871879577636719,
+      "mean_token_accuracy": 0.8042450234293937,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7772979542110743,
+      "eval_loss": 0.7561826705932617,
+      "eval_mean_token_accuracy": 0.7953876427720102,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 89.8395,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 460
+    },
+    {
+      "entropy": 0.7477855801582336,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 1.325486660003662,
+      "learning_rate": 0.00036445809162231435,
+      "loss": 0.6836830139160156,
+      "mean_token_accuracy": 0.8021619468927383,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.7098110676481483,
+      "eval_loss": 0.758007287979126,
+      "eval_mean_token_accuracy": 0.7909953189030122,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 89.6101,
+      "eval_samples_per_second": 15.846,
+      "eval_steps_per_second": 1.986,
+      "step": 480
+    },
+    {
+      "entropy": 0.7632691070437432,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 1.0122262239456177,
+      "learning_rate": 0.0003642672702835562,
+      "loss": 0.7071213245391845,
+      "mean_token_accuracy": 0.8000680930912495,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7645382720432924,
+      "eval_loss": 0.7492606043815613,
+      "eval_mean_token_accuracy": 0.7976858676149604,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 89.6092,
+      "eval_samples_per_second": 15.847,
+      "eval_steps_per_second": 1.986,
+      "step": 500
+    },
+    {
+      "entropy": 0.7623726457357407,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.367774486541748,
+      "learning_rate": 0.00036402497165841384,
+      "loss": 0.6965175151824952,
+      "mean_token_accuracy": 0.8016501650214195,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7868083715438843,
+      "eval_loss": 0.7481449842453003,
+      "eval_mean_token_accuracy": 0.799330582779445,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 89.8241,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 520
+    },
+    {
+      "entropy": 0.7535207405686378,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 1.213614821434021,
+      "learning_rate": 0.00036373126431453207,
+      "loss": 0.7078310489654541,
+      "mean_token_accuracy": 0.7998907208442688,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7829091950748743,
+      "eval_loss": 0.7416072487831116,
+      "eval_mean_token_accuracy": 0.7969146250339036,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 89.4607,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 540
+    },
+    {
+      "entropy": 0.7527363449335098,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 1.0929982662200928,
+      "learning_rate": 0.00036338623136761495,
+      "loss": 0.7058369636535644,
+      "mean_token_accuracy": 0.7997087433934211,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.728419938114252,
+      "eval_loss": 0.744873046875,
+      "eval_mean_token_accuracy": 0.7966634296299366,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 89.6597,
+      "eval_samples_per_second": 15.838,
+      "eval_steps_per_second": 1.985,
+      "step": 560
+    },
+    {
+      "entropy": 0.7406099259853363,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.409364104270935,
+      "learning_rate": 0.00036298997045790515,
+      "loss": 0.6902852058410645,
+      "mean_token_accuracy": 0.805386807769537,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.6991632758231645,
+      "eval_loss": 0.7480319738388062,
+      "eval_mean_token_accuracy": 0.7970068006033308,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 89.8329,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 580
+    },
+    {
+      "entropy": 0.7459449704736472,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.7053111791610718,
+      "learning_rate": 0.00036254259372255277,
+      "loss": 0.7044640064239502,
+      "mean_token_accuracy": 0.8005018755793571,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7454566660891758,
+      "eval_loss": 0.7551484704017639,
+      "eval_mean_token_accuracy": 0.7897093731365846,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 89.8313,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 600
+    },
+    {
+      "entropy": 0.7621391348540782,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 1.174443244934082,
+      "learning_rate": 0.000362044227763882,
+      "loss": 0.7111573696136475,
+      "mean_token_accuracy": 0.798752411454916,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7664387949397055,
+      "eval_loss": 0.7365804314613342,
+      "eval_mean_token_accuracy": 0.7963101200843126,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 89.9623,
+      "eval_samples_per_second": 15.784,
+      "eval_steps_per_second": 1.979,
+      "step": 620
+    },
+    {
+      "entropy": 0.7424944408237935,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.2898184061050415,
+      "learning_rate": 0.000361495013613564,
+      "loss": 0.6913110733032226,
+      "mean_token_accuracy": 0.8009015507996082,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7255479529332579,
+      "eval_loss": 0.7345991730690002,
+      "eval_mean_token_accuracy": 0.8013592945056015,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 89.817,
+      "eval_samples_per_second": 15.81,
+      "eval_steps_per_second": 1.982,
+      "step": 640
+    },
+    {
+      "entropy": 0.7303367637097835,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.1310476064682007,
+      "learning_rate": 0.00036089510669270666,
+      "loss": 0.6875617027282714,
+      "mean_token_accuracy": 0.8035397171974182,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7101308248016271,
+      "eval_loss": 0.7394917011260986,
+      "eval_mean_token_accuracy": 0.7990792942850777,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 89.7431,
+      "eval_samples_per_second": 15.823,
+      "eval_steps_per_second": 1.983,
+      "step": 660
+    },
+    {
+      "entropy": 0.7387678384780884,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.9645389914512634,
+      "learning_rate": 0.0003602446767678725,
+      "loss": 0.6911204338073731,
+      "mean_token_accuracy": 0.8056333385407924,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7281073737010527,
+      "eval_loss": 0.7376708984375,
+      "eval_mean_token_accuracy": 0.7997356415464637,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 90.0338,
+      "eval_samples_per_second": 15.772,
+      "eval_steps_per_second": 1.977,
+      "step": 680
+    },
+    {
+      "entropy": 0.7449554048478604,
+      "epoch": 1.687990343995172,
+      "grad_norm": 1.9069629907608032,
+      "learning_rate": 0.0003595439079030364,
+      "loss": 0.7042286396026611,
+      "mean_token_accuracy": 0.8018639378249646,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7022633817088738,
+      "eval_loss": 0.7320939898490906,
+      "eval_mean_token_accuracy": 0.8006051541044471,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 89.8948,
+      "eval_samples_per_second": 15.796,
+      "eval_steps_per_second": 1.98,
+      "step": 700
+    },
+    {
+      "entropy": 0.723249051719904,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.8333877325057983,
+      "learning_rate": 0.00035879299840749777,
+      "loss": 0.7043714046478271,
+      "mean_token_accuracy": 0.8028345607221127,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.672436946898364,
+      "eval_loss": 0.713233470916748,
+      "eval_mean_token_accuracy": 0.8047760577684038,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 89.6394,
+      "eval_samples_per_second": 15.841,
+      "eval_steps_per_second": 1.986,
+      "step": 720
+    },
+    {
+      "entropy": 0.730014206469059,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.4580037593841553,
+      "learning_rate": 0.00035799216077976137,
+      "loss": 0.6985176563262939,
+      "mean_token_accuracy": 0.8015547141432762,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.6908873795123582,
+      "eval_loss": 0.7198202013969421,
+      "eval_mean_token_accuracy": 0.8024131682481659,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 89.3771,
+      "eval_samples_per_second": 15.888,
+      "eval_steps_per_second": 1.992,
+      "step": 740
+    },
+    {
+      "entropy": 0.7219605796039105,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.51682710647583,
+      "learning_rate": 0.000357141621647403,
+      "loss": 0.6844424724578857,
+      "mean_token_accuracy": 0.8032813094556331,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7113401109582922,
+      "eval_loss": 0.7195309400558472,
+      "eval_mean_token_accuracy": 0.8019187035185568,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 89.4658,
+      "eval_samples_per_second": 15.872,
+      "eval_steps_per_second": 1.99,
+      "step": 760
+    },
+    {
+      "entropy": 0.729826743900776,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 1.0838000774383545,
+      "learning_rate": 0.0003562416217029361,
+      "loss": 0.685858678817749,
+      "mean_token_accuracy": 0.8050567395985126,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.649370267484965,
+      "eval_loss": 0.7177029252052307,
+      "eval_mean_token_accuracy": 0.8050174043419656,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 89.2287,
+      "eval_samples_per_second": 15.914,
+      "eval_steps_per_second": 1.995,
+      "step": 780
+    },
+    {
+      "entropy": 0.6799364153295755,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 1.1364924907684326,
+      "learning_rate": 0.00035529241563569903,
+      "loss": 0.6714428901672364,
+      "mean_token_accuracy": 0.8087341658771038,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6603799047095052,
+      "eval_loss": 0.7201809287071228,
+      "eval_mean_token_accuracy": 0.802863868769635,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 89.2718,
+      "eval_samples_per_second": 15.906,
+      "eval_steps_per_second": 1.994,
+      "step": 800
+    },
+    {
+      "entropy": 0.7078610375523567,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 1.2191942930221558,
+      "learning_rate": 0.0003542942720597808,
+      "loss": 0.6996590614318847,
+      "mean_token_accuracy": 0.8049512051045895,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.7240545264120852,
+      "eval_loss": 0.7301309108734131,
+      "eval_mean_token_accuracy": 0.8015973166133581,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 89.3438,
+      "eval_samples_per_second": 15.894,
+      "eval_steps_per_second": 1.992,
+      "step": 820
+    },
+    {
+      "entropy": 0.6557391556826505,
+      "epoch": 2.024140012070006,
+      "grad_norm": 1.36203134059906,
+      "learning_rate": 0.0003532474734380064,
+      "loss": 0.6327113628387451,
+      "mean_token_accuracy": 0.8142829847026181,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6573808648613062,
+      "eval_loss": 0.7334365844726562,
+      "eval_mean_token_accuracy": 0.8043633893634496,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 89.3359,
+      "eval_samples_per_second": 15.895,
+      "eval_steps_per_second": 1.992,
+      "step": 840
+    },
+    {
+      "entropy": 0.6409128502011299,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 1.5804955959320068,
+      "learning_rate": 0.0003521523160020035,
+      "loss": 0.6023559093475341,
+      "mean_token_accuracy": 0.8222251623868942,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6983234788594621,
+      "eval_loss": 0.7232135534286499,
+      "eval_mean_token_accuracy": 0.8010654268639811,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 89.2615,
+      "eval_samples_per_second": 15.908,
+      "eval_steps_per_second": 1.994,
+      "step": 860
+    },
+    {
+      "entropy": 0.6151149723678827,
+      "epoch": 2.12070006035003,
+      "grad_norm": 1.1975436210632324,
+      "learning_rate": 0.00035100910966837193,
+      "loss": 0.5803564548492431,
+      "mean_token_accuracy": 0.8256325736641884,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6636555573243773,
+      "eval_loss": 0.7205662727355957,
+      "eval_mean_token_accuracy": 0.8036768369460374,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 89.2957,
+      "eval_samples_per_second": 15.902,
+      "eval_steps_per_second": 1.993,
+      "step": 880
+    },
+    {
+      "entropy": 0.6201727617532015,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9907131195068359,
+      "learning_rate": 0.0003498181779509813,
+      "loss": 0.6101872444152832,
+      "mean_token_accuracy": 0.8196512959897518,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.6473788086617931,
+      "eval_loss": 0.7062397003173828,
+      "eval_mean_token_accuracy": 0.8078362184963869,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 88.6198,
+      "eval_samples_per_second": 16.024,
+      "eval_steps_per_second": 2.009,
+      "step": 900
+    },
+    {
+      "entropy": 0.6097332935780286,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.9055633544921875,
+      "learning_rate": 0.00034857985786942026,
+      "loss": 0.5956094264984131,
+      "mean_token_accuracy": 0.8240575887262821,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.612629868341296,
+      "eval_loss": 0.7106770277023315,
+      "eval_mean_token_accuracy": 0.805324277181304,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 89.4383,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 920
+    },
+    {
+      "entropy": 0.6182701248675585,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.9513736367225647,
+      "learning_rate": 0.00034729449985362404,
+      "loss": 0.6019269466400147,
+      "mean_token_accuracy": 0.8187106013298034,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6437647456533453,
+      "eval_loss": 0.7173502445220947,
+      "eval_mean_token_accuracy": 0.8056439573175451,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.9568,
+      "eval_samples_per_second": 15.963,
+      "eval_steps_per_second": 2.001,
+      "step": 940
+    },
+    {
+      "entropy": 0.6508570514619351,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.9606207013130188,
+      "learning_rate": 0.0003459624676447067,
+      "loss": 0.6296409606933594,
+      "mean_token_accuracy": 0.8167732007801533,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6288247135248077,
+      "eval_loss": 0.7160675525665283,
+      "eval_mean_token_accuracy": 0.8066010689467527,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 89.0044,
+      "eval_samples_per_second": 15.954,
+      "eval_steps_per_second": 2.0,
+      "step": 960
+    },
+    {
+      "entropy": 0.6380081418901682,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.9677181839942932,
+      "learning_rate": 0.000344584138192027,
+      "loss": 0.6210546493530273,
+      "mean_token_accuracy": 0.8186135642230511,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6407805031604981,
+      "eval_loss": 0.7055043578147888,
+      "eval_mean_token_accuracy": 0.8070103157772107,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 89.5606,
+      "eval_samples_per_second": 15.855,
+      "eval_steps_per_second": 1.987,
+      "step": 980
+    },
+    {
+      "entropy": 0.6310251638293266,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 2.1046407222747803,
+      "learning_rate": 0.000343159901546516,
+      "loss": 0.6165830135345459,
+      "mean_token_accuracy": 0.8207855716347694,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6461507470420237,
+      "eval_loss": 0.7292136549949646,
+      "eval_mean_token_accuracy": 0.8030418261383356,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 89.315,
+      "eval_samples_per_second": 15.899,
+      "eval_steps_per_second": 1.993,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6256617344915867,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 1.1768349409103394,
+      "learning_rate": 0.0003416901607502972,
+      "loss": 0.6142709255218506,
+      "mean_token_accuracy": 0.8220669947564602,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6150523993406403,
+      "eval_loss": 0.7019563317298889,
+      "eval_mean_token_accuracy": 0.8094323099998946,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 89.419,
+      "eval_samples_per_second": 15.88,
+      "eval_steps_per_second": 1.991,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6126735735684633,
+      "epoch": 2.506940253470127,
+      "grad_norm": 1.8065496683120728,
+      "learning_rate": 0.00034017533172263055,
+      "loss": 0.6056458473205566,
+      "mean_token_accuracy": 0.8203762136399746,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6544345456562685,
+      "eval_loss": 0.7086517214775085,
+      "eval_mean_token_accuracy": 0.806709805901131,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 88.981,
+      "eval_samples_per_second": 15.958,
+      "eval_steps_per_second": 2.0,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6311754353344441,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 1.0576369762420654,
+      "learning_rate": 0.00033861584314221236,
+      "loss": 0.6176392555236816,
+      "mean_token_accuracy": 0.8177410490810871,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6273799200406235,
+      "eval_loss": 0.708163321018219,
+      "eval_mean_token_accuracy": 0.8032017099053672,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 89.7668,
+      "eval_samples_per_second": 15.819,
+      "eval_steps_per_second": 1.983,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6200249589979648,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.997065007686615,
+      "learning_rate": 0.0003370121363258637,
+      "loss": 0.6203888893127442,
+      "mean_token_accuracy": 0.8196275025606156,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6457576811983344,
+      "eval_loss": 0.6972290277481079,
+      "eval_mean_token_accuracy": 0.8095980849158898,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 89.2027,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 1080
+    },
+    {
+      "entropy": 0.626810473203659,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.865475058555603,
+      "learning_rate": 0.0003353646651036438,
+      "loss": 0.6102585792541504,
+      "mean_token_accuracy": 0.819708751142025,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.6412022431914726,
+      "eval_loss": 0.7036139965057373,
+      "eval_mean_token_accuracy": 0.8070317135098275,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 89.4733,
+      "eval_samples_per_second": 15.871,
+      "eval_steps_per_second": 1.989,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6581619590520859,
+      "epoch": 2.700060350030175,
+      "grad_norm": 1.00307035446167,
+      "learning_rate": 0.00033367389569042064,
+      "loss": 0.6397720336914062,
+      "mean_token_accuracy": 0.81413309648633,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6288736402318719,
+      "eval_loss": 0.6966825723648071,
+      "eval_mean_token_accuracy": 0.8094039669867312,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 88.7341,
+      "eval_samples_per_second": 16.003,
+      "eval_steps_per_second": 2.006,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6399412982165813,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.8589329719543457,
+      "learning_rate": 0.0003319403065539384,
+      "loss": 0.6347338199615479,
+      "mean_token_accuracy": 0.8139534957706929,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.683491913455256,
+      "eval_loss": 0.6984149217605591,
+      "eval_mean_token_accuracy": 0.8083989469522841,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 89.3929,
+      "eval_samples_per_second": 15.885,
+      "eval_steps_per_second": 1.991,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6264835461974144,
+      "epoch": 2.796620398310199,
+      "grad_norm": 0.7807853817939758,
+      "learning_rate": 0.0003301643882794163,
+      "loss": 0.6162629127502441,
+      "mean_token_accuracy": 0.8216292977333068,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6578792159476977,
+      "eval_loss": 0.6921441555023193,
+      "eval_mean_token_accuracy": 0.8094299177775223,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 89.2356,
+      "eval_samples_per_second": 15.913,
+      "eval_steps_per_second": 1.995,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6269129011780024,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 1.1675593852996826,
+      "learning_rate": 0.0003283466434307189,
+      "loss": 0.6235532760620117,
+      "mean_token_accuracy": 0.8188897788524627,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6533248447970058,
+      "eval_loss": 0.696849524974823,
+      "eval_mean_token_accuracy": 0.8102703713968898,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2602,
+      "eval_samples_per_second": 16.089,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.6259873781353236,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.9596486687660217,
+      "learning_rate": 0.00032648758640813655,
+      "loss": 0.6118728160858155,
+      "mean_token_accuracy": 0.8171859130263328,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6212078754821521,
+      "eval_loss": 0.6942870020866394,
+      "eval_mean_token_accuracy": 0.810484343365337,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 89.4375,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 1200
+    },
+    {
+      "entropy": 0.6142537571489811,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 0.9625058770179749,
+      "learning_rate": 0.00032458774330281615,
+      "loss": 0.6060261249542236,
+      "mean_token_accuracy": 0.8223113484680653,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6553665667437436,
+      "eval_loss": 0.6835269331932068,
+      "eval_mean_token_accuracy": 0.8134723968720168,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 89.5092,
+      "eval_samples_per_second": 15.864,
+      "eval_steps_per_second": 1.989,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6301972426474094,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.9657333493232727,
+      "learning_rate": 0.0003226476517478835,
+      "loss": 0.6277759075164795,
+      "mean_token_accuracy": 0.8164266042411328,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6594253118788258,
+      "eval_loss": 0.6820935606956482,
+      "eval_mean_token_accuracy": 0.8132368198941263,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 89.6211,
+      "eval_samples_per_second": 15.844,
+      "eval_steps_per_second": 1.986,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5346100655469027,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8541185259819031,
+      "learning_rate": 0.0003206678607662996,
+      "loss": 0.5261692047119141,
+      "mean_token_accuracy": 0.8382289425119177,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5626645662476507,
+      "eval_loss": 0.7028542160987854,
+      "eval_mean_token_accuracy": 0.8133395572056931,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 89.1267,
+      "eval_samples_per_second": 15.932,
+      "eval_steps_per_second": 1.997,
+      "step": 1260
+    },
+    {
+      "entropy": 0.5001101832836866,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.8208815455436707,
+      "learning_rate": 0.0003186489306154935,
+      "loss": 0.4850132942199707,
+      "mean_token_accuracy": 0.8468983843922615,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5889462046743779,
+      "eval_loss": 0.691015899181366,
+      "eval_mean_token_accuracy": 0.8135226466012805,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.9677,
+      "eval_samples_per_second": 15.961,
+      "eval_steps_per_second": 2.001,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5242766339331866,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 0.7893072962760925,
+      "learning_rate": 0.0003165914326288163,
+      "loss": 0.5061577320098877,
+      "mean_token_accuracy": 0.8425608821213245,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5543691962957382,
+      "eval_loss": 0.6913734078407288,
+      "eval_mean_token_accuracy": 0.8154889680026622,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 89.6963,
+      "eval_samples_per_second": 15.831,
+      "eval_steps_per_second": 1.984,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5327721010893584,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.8992080092430115,
+      "learning_rate": 0.0003144959490538604,
+      "loss": 0.5068036556243897,
+      "mean_token_accuracy": 0.8420542575418949,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5734540803378887,
+      "eval_loss": 0.6948716044425964,
+      "eval_mean_token_accuracy": 0.8137217944257715,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 89.4597,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5247377116233111,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.9291845560073853,
+      "learning_rate": 0.0003123630728876902,
+      "loss": 0.5023272037506104,
+      "mean_token_accuracy": 0.8435182586312294,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5492362935891312,
+      "eval_loss": 0.6980633735656738,
+      "eval_mean_token_accuracy": 0.8155082919624415,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 89.2124,
+      "eval_samples_per_second": 15.917,
+      "eval_steps_per_second": 1.995,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5380499072372913,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 1.0477524995803833,
+      "learning_rate": 0.00031019340770903136,
+      "loss": 0.5264591693878173,
+      "mean_token_accuracy": 0.8391959741711617,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5896503307511297,
+      "eval_loss": 0.6997144222259521,
+      "eval_mean_token_accuracy": 0.8124373062942805,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 89.1147,
+      "eval_samples_per_second": 15.935,
+      "eval_steps_per_second": 1.997,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5460653610527515,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 1.2473069429397583,
+      "learning_rate": 0.00030798756750746476,
+      "loss": 0.5319677829742432,
+      "mean_token_accuracy": 0.8364320226013661,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5659052182114526,
+      "eval_loss": 0.6924049258232117,
+      "eval_mean_token_accuracy": 0.8153053585732921,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 89.6468,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5367407951503992,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 1.2679712772369385,
+      "learning_rate": 0.00030574617650967485,
+      "loss": 0.5190927028656006,
+      "mean_token_accuracy": 0.8399469174444676,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5581759144081159,
+      "eval_loss": 0.6933317184448242,
+      "eval_mean_token_accuracy": 0.8139757862251796,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 89.0749,
+      "eval_samples_per_second": 15.942,
+      "eval_steps_per_second": 1.998,
+      "step": 1400
+    },
+    {
+      "entropy": 0.530736118927598,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.9717612862586975,
+      "learning_rate": 0.0003034698690028009,
+      "loss": 0.512363052368164,
+      "mean_token_accuracy": 0.8419127956032753,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5233335836549823,
+      "eval_loss": 0.6958550214767456,
+      "eval_mean_token_accuracy": 0.8158778501360604,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 88.8664,
+      "eval_samples_per_second": 15.979,
+      "eval_steps_per_second": 2.003,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5327631575986743,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7649208903312683,
+      "learning_rate": 0.00030115928915494116,
+      "loss": 0.5202179431915284,
+      "mean_token_accuracy": 0.8384683802723885,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.5960972657364406,
+      "eval_loss": 0.6984783411026001,
+      "eval_mean_token_accuracy": 0.8095930597085631,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 89.5168,
+      "eval_samples_per_second": 15.863,
+      "eval_steps_per_second": 1.988,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5428155034780502,
+      "epoch": 3.51901025950513,
+      "grad_norm": 1.2790874242782593,
+      "learning_rate": 0.00029881509083286116,
+      "loss": 0.524717378616333,
+      "mean_token_accuracy": 0.8346644505858422,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.550318562917495,
+      "eval_loss": 0.6988471746444702,
+      "eval_mean_token_accuracy": 0.8132969897784544,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 89.2598,
+      "eval_samples_per_second": 15.909,
+      "eval_steps_per_second": 1.994,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5489993002265692,
+      "epoch": 3.567290283645142,
+      "grad_norm": 1.055821418762207,
+      "learning_rate": 0.00029643793741695687,
+      "loss": 0.5421150207519532,
+      "mean_token_accuracy": 0.8345710933208466,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5816821718818685,
+      "eval_loss": 0.6853668689727783,
+      "eval_mean_token_accuracy": 0.8142332360985574,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 89.0424,
+      "eval_samples_per_second": 15.947,
+      "eval_steps_per_second": 1.999,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5451365381479263,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.9027532935142517,
+      "learning_rate": 0.00029402850161352583,
+      "loss": 0.5463609218597412,
+      "mean_token_accuracy": 0.8369908526539802,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5344233209832331,
+      "eval_loss": 0.6794531345367432,
+      "eval_mean_token_accuracy": 0.8177759824843889,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 89.123,
+      "eval_samples_per_second": 15.933,
+      "eval_steps_per_second": 1.997,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5521467242389917,
+      "epoch": 3.663850331925166,
+      "grad_norm": 1.4606311321258545,
+      "learning_rate": 0.0002915874652643997,
+      "loss": 0.5421902179718018,
+      "mean_token_accuracy": 0.8339455373585224,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5807709094513668,
+      "eval_loss": 0.6949691772460938,
+      "eval_mean_token_accuracy": 0.8121141439743256,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.5437,
+      "eval_samples_per_second": 16.037,
+      "eval_steps_per_second": 2.01,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5455610822886229,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7269843220710754,
+      "learning_rate": 0.0002891155191539905,
+      "loss": 0.5407682895660401,
+      "mean_token_accuracy": 0.8393648222088814,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.6032205542151847,
+      "eval_loss": 0.6790974140167236,
+      "eval_mean_token_accuracy": 0.8165463112043531,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 89.517,
+      "eval_samples_per_second": 15.863,
+      "eval_steps_per_second": 1.988,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5474163968116045,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.8148866295814514,
+      "learning_rate": 0.00028661336281380717,
+      "loss": 0.5394441604614257,
+      "mean_token_accuracy": 0.8336028568446636,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5797032238392348,
+      "eval_loss": 0.670870304107666,
+      "eval_mean_token_accuracy": 0.8182691593518417,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 89.4468,
+      "eval_samples_per_second": 15.875,
+      "eval_steps_per_second": 1.99,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5511750865727663,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.3441656827926636,
+      "learning_rate": 0.0002840817043244964,
+      "loss": 0.5378933906555176,
+      "mean_token_accuracy": 0.8349151819944381,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5297858750217417,
+      "eval_loss": 0.680833637714386,
+      "eval_mean_token_accuracy": 0.8180924275617921,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 89.5417,
+      "eval_samples_per_second": 15.859,
+      "eval_steps_per_second": 1.988,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5379613988101483,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9813932776451111,
+      "learning_rate": 0.00028152126011546396,
+      "loss": 0.5340342044830322,
+      "mean_token_accuracy": 0.8346010789275169,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5811898510777549,
+      "eval_loss": 0.6764267683029175,
+      "eval_mean_token_accuracy": 0.8180528236239144,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 88.7831,
+      "eval_samples_per_second": 15.994,
+      "eval_steps_per_second": 2.005,
+      "step": 1600
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.6031894172239462e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1620/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1620/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1620/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1620/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1620/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..7ce4b072fcbe5b636867e13937c56fedee3b68b4
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1620/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.02728394274790723,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1620/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1620/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1620/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1620/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1620/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1620/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1620/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1620/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..b52b40f924bd4c06c11b91332f1fabc8db1b37a0
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1620/trainer_state.json
@@ -0,0 +1,1735 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 3.9052504526252263,
+  "eval_steps": 20,
+  "global_step": 1620,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.024795526266098,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.945375919342041,
+      "learning_rate": 1.6698127428345936e-05,
+      "loss": 1.7825420379638672,
+      "mean_token_accuracy": 0.6315708436071873,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.29921497588747,
+      "eval_loss": 1.169131875038147,
+      "eval_mean_token_accuracy": 0.7185706342204233,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 91.7898,
+      "eval_samples_per_second": 15.47,
+      "eval_steps_per_second": 1.939,
+      "step": 20
+    },
+    {
+      "entropy": 1.0346344470977784,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.7916979789733887,
+      "learning_rate": 3.427510366871008e-05,
+      "loss": 0.9480395317077637,
+      "mean_token_accuracy": 0.7543429024517536,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 0.9673695430326997,
+      "eval_loss": 0.8588430285453796,
+      "eval_mean_token_accuracy": 0.7737233002534073,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 89.8219,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 40
+    },
+    {
+      "entropy": 0.8977566614747048,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7972187995910645,
+      "learning_rate": 5.185207990907422e-05,
+      "loss": 0.8029604911804199,
+      "mean_token_accuracy": 0.7818016350269318,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8436836982041263,
+      "eval_loss": 0.7893033623695374,
+      "eval_mean_token_accuracy": 0.7858914984076211,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 89.9349,
+      "eval_samples_per_second": 15.789,
+      "eval_steps_per_second": 1.979,
+      "step": 60
+    },
+    {
+      "entropy": 0.8650321319699288,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.2128729820251465,
+      "learning_rate": 6.942905614943836e-05,
+      "loss": 0.77417893409729,
+      "mean_token_accuracy": 0.7875479467213153,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8209857180547179,
+      "eval_loss": 0.7572280168533325,
+      "eval_mean_token_accuracy": 0.7925970383574453,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 89.9459,
+      "eval_samples_per_second": 15.787,
+      "eval_steps_per_second": 1.979,
+      "step": 80
+    },
+    {
+      "entropy": 0.8096660938113928,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.1254287958145142,
+      "learning_rate": 8.700603238980251e-05,
+      "loss": 0.7170331001281738,
+      "mean_token_accuracy": 0.7994555257260799,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.792780208788561,
+      "eval_loss": 0.7442984580993652,
+      "eval_mean_token_accuracy": 0.7954704573984896,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 90.294,
+      "eval_samples_per_second": 15.726,
+      "eval_steps_per_second": 1.971,
+      "step": 100
+    },
+    {
+      "entropy": 0.7979829199612141,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2140507698059082,
+      "learning_rate": 0.00010458300863016666,
+      "loss": 0.7141861915588379,
+      "mean_token_accuracy": 0.8000014744699001,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.7938692486018277,
+      "eval_loss": 0.7279797196388245,
+      "eval_mean_token_accuracy": 0.7989798734027348,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 89.7189,
+      "eval_samples_per_second": 15.827,
+      "eval_steps_per_second": 1.984,
+      "step": 120
+    },
+    {
+      "entropy": 0.7916347607970238,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.0506458282470703,
+      "learning_rate": 0.0001221599848705308,
+      "loss": 0.7117055416107178,
+      "mean_token_accuracy": 0.7970046654343605,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.8052781063519167,
+      "eval_loss": 0.722576916217804,
+      "eval_mean_token_accuracy": 0.7982530523551984,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 90.0376,
+      "eval_samples_per_second": 15.771,
+      "eval_steps_per_second": 1.977,
+      "step": 140
+    },
+    {
+      "entropy": 0.7880666613578796,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8409207463264465,
+      "learning_rate": 0.00013973696111089492,
+      "loss": 0.7052523136138916,
+      "mean_token_accuracy": 0.8000123649835587,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7787443134891853,
+      "eval_loss": 0.7137772440910339,
+      "eval_mean_token_accuracy": 0.8014209287220173,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 90.2562,
+      "eval_samples_per_second": 15.733,
+      "eval_steps_per_second": 1.972,
+      "step": 160
+    },
+    {
+      "entropy": 0.7696007996797561,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.0182477235794067,
+      "learning_rate": 0.00015731393735125907,
+      "loss": 0.7008797645568847,
+      "mean_token_accuracy": 0.8007259473204613,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.752259682068664,
+      "eval_loss": 0.7171286344528198,
+      "eval_mean_token_accuracy": 0.80052122574174,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 90.0797,
+      "eval_samples_per_second": 15.764,
+      "eval_steps_per_second": 1.976,
+      "step": 180
+    },
+    {
+      "entropy": 0.7828126326203346,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9553632736206055,
+      "learning_rate": 0.0001748909135916232,
+      "loss": 0.7046853542327881,
+      "mean_token_accuracy": 0.7988562889397144,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7389638169427936,
+      "eval_loss": 0.7187935709953308,
+      "eval_mean_token_accuracy": 0.7995899999409579,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 89.9223,
+      "eval_samples_per_second": 15.791,
+      "eval_steps_per_second": 1.979,
+      "step": 200
+    },
+    {
+      "entropy": 0.7909286297857762,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 1.134459376335144,
+      "learning_rate": 0.00019246788983198735,
+      "loss": 0.6989312171936035,
+      "mean_token_accuracy": 0.7991565138101577,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7496965199374082,
+      "eval_loss": 0.7106321454048157,
+      "eval_mean_token_accuracy": 0.8046183442131857,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 89.6899,
+      "eval_samples_per_second": 15.832,
+      "eval_steps_per_second": 1.985,
+      "step": 220
+    },
+    {
+      "entropy": 0.7651963323354721,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.9557585716247559,
+      "learning_rate": 0.0002100448660723515,
+      "loss": 0.6984559059143066,
+      "mean_token_accuracy": 0.8019823372364044,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7615563079212488,
+      "eval_loss": 0.7098539471626282,
+      "eval_mean_token_accuracy": 0.8047233508544022,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 90.3752,
+      "eval_samples_per_second": 15.712,
+      "eval_steps_per_second": 1.97,
+      "step": 240
+    },
+    {
+      "entropy": 0.787507726252079,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.9279311299324036,
+      "learning_rate": 0.00022762184231271568,
+      "loss": 0.708138370513916,
+      "mean_token_accuracy": 0.8013308703899383,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7856640896100676,
+      "eval_loss": 0.7191570401191711,
+      "eval_mean_token_accuracy": 0.8024677249153008,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 93.7336,
+      "eval_samples_per_second": 15.149,
+      "eval_steps_per_second": 1.899,
+      "step": 260
+    },
+    {
+      "entropy": 0.768489234894514,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.7992776036262512,
+      "learning_rate": 0.0002451988185530798,
+      "loss": 0.7035849571228028,
+      "mean_token_accuracy": 0.8031105332076549,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7682498740346244,
+      "eval_loss": 0.7149022221565247,
+      "eval_mean_token_accuracy": 0.7973497995499814,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 89.7309,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 280
+    },
+    {
+      "entropy": 0.7720955617725849,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.1931514739990234,
+      "learning_rate": 0.0002627757947934439,
+      "loss": 0.6965402603149414,
+      "mean_token_accuracy": 0.8040348663926125,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7214708304807042,
+      "eval_loss": 0.7350823879241943,
+      "eval_mean_token_accuracy": 0.7942911925610532,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 89.6451,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 300
+    },
+    {
+      "entropy": 0.7930095426738262,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8994467258453369,
+      "learning_rate": 0.0002803527710338081,
+      "loss": 0.7331492900848389,
+      "mean_token_accuracy": 0.796185302734375,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.8142535488926963,
+      "eval_loss": 0.7237834334373474,
+      "eval_mean_token_accuracy": 0.799495273761535,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 90.6759,
+      "eval_samples_per_second": 15.66,
+      "eval_steps_per_second": 1.963,
+      "step": 320
+    },
+    {
+      "entropy": 0.7825102441012859,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 1.8224239349365234,
+      "learning_rate": 0.00029792974727417223,
+      "loss": 0.7331175804138184,
+      "mean_token_accuracy": 0.7977667152881622,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7327135443017724,
+      "eval_loss": 0.7388784289360046,
+      "eval_mean_token_accuracy": 0.8007041252730938,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 89.7692,
+      "eval_samples_per_second": 15.818,
+      "eval_steps_per_second": 1.983,
+      "step": 340
+    },
+    {
+      "entropy": 0.819331557303667,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 1.0407779216766357,
+      "learning_rate": 0.0003155067235145364,
+      "loss": 0.7456695556640625,
+      "mean_token_accuracy": 0.7936980701982975,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.8065585420372781,
+      "eval_loss": 0.735519528388977,
+      "eval_mean_token_accuracy": 0.7974206296245704,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 89.4336,
+      "eval_samples_per_second": 15.878,
+      "eval_steps_per_second": 1.99,
+      "step": 360
+    },
+    {
+      "entropy": 0.8114653021097183,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.1209490299224854,
+      "learning_rate": 0.0003330836997549005,
+      "loss": 0.7340899467468261,
+      "mean_token_accuracy": 0.7914272703230381,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.8025527285056168,
+      "eval_loss": 0.7295576930046082,
+      "eval_mean_token_accuracy": 0.7990576628218876,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 89.9579,
+      "eval_samples_per_second": 15.785,
+      "eval_steps_per_second": 1.979,
+      "step": 380
+    },
+    {
+      "entropy": 0.8042497783899307,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 1.2437798976898193,
+      "learning_rate": 0.0003506606759952647,
+      "loss": 0.7381744384765625,
+      "mean_token_accuracy": 0.7967362694442273,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7717011336530193,
+      "eval_loss": 0.7436173558235168,
+      "eval_mean_token_accuracy": 0.7975849482450592,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 89.8411,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 400
+    },
+    {
+      "entropy": 0.7987750010056929,
+      "epoch": 1.012070006035003,
+      "grad_norm": 1.2390918731689453,
+      "learning_rate": 0.000364721224843344,
+      "loss": 0.7317790031433106,
+      "mean_token_accuracy": 0.7986087551364651,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7333170604170038,
+      "eval_loss": 0.7456948757171631,
+      "eval_mean_token_accuracy": 0.7964897577682238,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 89.7288,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 420
+    },
+    {
+      "entropy": 0.7372976846992969,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.1552740335464478,
+      "learning_rate": 0.00036468510102269575,
+      "loss": 0.6734249114990234,
+      "mean_token_accuracy": 0.805834824591875,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.7130355526891987,
+      "eval_loss": 0.7555333375930786,
+      "eval_mean_token_accuracy": 0.7961652781186479,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 89.8713,
+      "eval_samples_per_second": 15.8,
+      "eval_steps_per_second": 1.981,
+      "step": 440
+    },
+    {
+      "entropy": 0.7465668715536594,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 1.1991469860076904,
+      "learning_rate": 0.0003645973816745066,
+      "loss": 0.6871879577636719,
+      "mean_token_accuracy": 0.8042450234293937,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7772979542110743,
+      "eval_loss": 0.7561826705932617,
+      "eval_mean_token_accuracy": 0.7953876427720102,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 89.8395,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 460
+    },
+    {
+      "entropy": 0.7477855801582336,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 1.325486660003662,
+      "learning_rate": 0.00036445809162231435,
+      "loss": 0.6836830139160156,
+      "mean_token_accuracy": 0.8021619468927383,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.7098110676481483,
+      "eval_loss": 0.758007287979126,
+      "eval_mean_token_accuracy": 0.7909953189030122,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 89.6101,
+      "eval_samples_per_second": 15.846,
+      "eval_steps_per_second": 1.986,
+      "step": 480
+    },
+    {
+      "entropy": 0.7632691070437432,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 1.0122262239456177,
+      "learning_rate": 0.0003642672702835562,
+      "loss": 0.7071213245391845,
+      "mean_token_accuracy": 0.8000680930912495,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7645382720432924,
+      "eval_loss": 0.7492606043815613,
+      "eval_mean_token_accuracy": 0.7976858676149604,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 89.6092,
+      "eval_samples_per_second": 15.847,
+      "eval_steps_per_second": 1.986,
+      "step": 500
+    },
+    {
+      "entropy": 0.7623726457357407,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.367774486541748,
+      "learning_rate": 0.00036402497165841384,
+      "loss": 0.6965175151824952,
+      "mean_token_accuracy": 0.8016501650214195,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7868083715438843,
+      "eval_loss": 0.7481449842453003,
+      "eval_mean_token_accuracy": 0.799330582779445,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 89.8241,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 520
+    },
+    {
+      "entropy": 0.7535207405686378,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 1.213614821434021,
+      "learning_rate": 0.00036373126431453207,
+      "loss": 0.7078310489654541,
+      "mean_token_accuracy": 0.7998907208442688,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7829091950748743,
+      "eval_loss": 0.7416072487831116,
+      "eval_mean_token_accuracy": 0.7969146250339036,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 89.4607,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 540
+    },
+    {
+      "entropy": 0.7527363449335098,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 1.0929982662200928,
+      "learning_rate": 0.00036338623136761495,
+      "loss": 0.7058369636535644,
+      "mean_token_accuracy": 0.7997087433934211,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.728419938114252,
+      "eval_loss": 0.744873046875,
+      "eval_mean_token_accuracy": 0.7966634296299366,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 89.6597,
+      "eval_samples_per_second": 15.838,
+      "eval_steps_per_second": 1.985,
+      "step": 560
+    },
+    {
+      "entropy": 0.7406099259853363,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.409364104270935,
+      "learning_rate": 0.00036298997045790515,
+      "loss": 0.6902852058410645,
+      "mean_token_accuracy": 0.805386807769537,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.6991632758231645,
+      "eval_loss": 0.7480319738388062,
+      "eval_mean_token_accuracy": 0.7970068006033308,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 89.8329,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 580
+    },
+    {
+      "entropy": 0.7459449704736472,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.7053111791610718,
+      "learning_rate": 0.00036254259372255277,
+      "loss": 0.7044640064239502,
+      "mean_token_accuracy": 0.8005018755793571,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7454566660891758,
+      "eval_loss": 0.7551484704017639,
+      "eval_mean_token_accuracy": 0.7897093731365846,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 89.8313,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 600
+    },
+    {
+      "entropy": 0.7621391348540782,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 1.174443244934082,
+      "learning_rate": 0.000362044227763882,
+      "loss": 0.7111573696136475,
+      "mean_token_accuracy": 0.798752411454916,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7664387949397055,
+      "eval_loss": 0.7365804314613342,
+      "eval_mean_token_accuracy": 0.7963101200843126,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 89.9623,
+      "eval_samples_per_second": 15.784,
+      "eval_steps_per_second": 1.979,
+      "step": 620
+    },
+    {
+      "entropy": 0.7424944408237935,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.2898184061050415,
+      "learning_rate": 0.000361495013613564,
+      "loss": 0.6913110733032226,
+      "mean_token_accuracy": 0.8009015507996082,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7255479529332579,
+      "eval_loss": 0.7345991730690002,
+      "eval_mean_token_accuracy": 0.8013592945056015,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 89.817,
+      "eval_samples_per_second": 15.81,
+      "eval_steps_per_second": 1.982,
+      "step": 640
+    },
+    {
+      "entropy": 0.7303367637097835,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.1310476064682007,
+      "learning_rate": 0.00036089510669270666,
+      "loss": 0.6875617027282714,
+      "mean_token_accuracy": 0.8035397171974182,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7101308248016271,
+      "eval_loss": 0.7394917011260986,
+      "eval_mean_token_accuracy": 0.7990792942850777,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 89.7431,
+      "eval_samples_per_second": 15.823,
+      "eval_steps_per_second": 1.983,
+      "step": 660
+    },
+    {
+      "entropy": 0.7387678384780884,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.9645389914512634,
+      "learning_rate": 0.0003602446767678725,
+      "loss": 0.6911204338073731,
+      "mean_token_accuracy": 0.8056333385407924,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7281073737010527,
+      "eval_loss": 0.7376708984375,
+      "eval_mean_token_accuracy": 0.7997356415464637,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 90.0338,
+      "eval_samples_per_second": 15.772,
+      "eval_steps_per_second": 1.977,
+      "step": 680
+    },
+    {
+      "entropy": 0.7449554048478604,
+      "epoch": 1.687990343995172,
+      "grad_norm": 1.9069629907608032,
+      "learning_rate": 0.0003595439079030364,
+      "loss": 0.7042286396026611,
+      "mean_token_accuracy": 0.8018639378249646,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7022633817088738,
+      "eval_loss": 0.7320939898490906,
+      "eval_mean_token_accuracy": 0.8006051541044471,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 89.8948,
+      "eval_samples_per_second": 15.796,
+      "eval_steps_per_second": 1.98,
+      "step": 700
+    },
+    {
+      "entropy": 0.723249051719904,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.8333877325057983,
+      "learning_rate": 0.00035879299840749777,
+      "loss": 0.7043714046478271,
+      "mean_token_accuracy": 0.8028345607221127,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.672436946898364,
+      "eval_loss": 0.713233470916748,
+      "eval_mean_token_accuracy": 0.8047760577684038,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 89.6394,
+      "eval_samples_per_second": 15.841,
+      "eval_steps_per_second": 1.986,
+      "step": 720
+    },
+    {
+      "entropy": 0.730014206469059,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.4580037593841553,
+      "learning_rate": 0.00035799216077976137,
+      "loss": 0.6985176563262939,
+      "mean_token_accuracy": 0.8015547141432762,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.6908873795123582,
+      "eval_loss": 0.7198202013969421,
+      "eval_mean_token_accuracy": 0.8024131682481659,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 89.3771,
+      "eval_samples_per_second": 15.888,
+      "eval_steps_per_second": 1.992,
+      "step": 740
+    },
+    {
+      "entropy": 0.7219605796039105,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.51682710647583,
+      "learning_rate": 0.000357141621647403,
+      "loss": 0.6844424724578857,
+      "mean_token_accuracy": 0.8032813094556331,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7113401109582922,
+      "eval_loss": 0.7195309400558472,
+      "eval_mean_token_accuracy": 0.8019187035185568,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 89.4658,
+      "eval_samples_per_second": 15.872,
+      "eval_steps_per_second": 1.99,
+      "step": 760
+    },
+    {
+      "entropy": 0.729826743900776,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 1.0838000774383545,
+      "learning_rate": 0.0003562416217029361,
+      "loss": 0.685858678817749,
+      "mean_token_accuracy": 0.8050567395985126,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.649370267484965,
+      "eval_loss": 0.7177029252052307,
+      "eval_mean_token_accuracy": 0.8050174043419656,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 89.2287,
+      "eval_samples_per_second": 15.914,
+      "eval_steps_per_second": 1.995,
+      "step": 780
+    },
+    {
+      "entropy": 0.6799364153295755,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 1.1364924907684326,
+      "learning_rate": 0.00035529241563569903,
+      "loss": 0.6714428901672364,
+      "mean_token_accuracy": 0.8087341658771038,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6603799047095052,
+      "eval_loss": 0.7201809287071228,
+      "eval_mean_token_accuracy": 0.802863868769635,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 89.2718,
+      "eval_samples_per_second": 15.906,
+      "eval_steps_per_second": 1.994,
+      "step": 800
+    },
+    {
+      "entropy": 0.7078610375523567,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 1.2191942930221558,
+      "learning_rate": 0.0003542942720597808,
+      "loss": 0.6996590614318847,
+      "mean_token_accuracy": 0.8049512051045895,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.7240545264120852,
+      "eval_loss": 0.7301309108734131,
+      "eval_mean_token_accuracy": 0.8015973166133581,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 89.3438,
+      "eval_samples_per_second": 15.894,
+      "eval_steps_per_second": 1.992,
+      "step": 820
+    },
+    {
+      "entropy": 0.6557391556826505,
+      "epoch": 2.024140012070006,
+      "grad_norm": 1.36203134059906,
+      "learning_rate": 0.0003532474734380064,
+      "loss": 0.6327113628387451,
+      "mean_token_accuracy": 0.8142829847026181,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6573808648613062,
+      "eval_loss": 0.7334365844726562,
+      "eval_mean_token_accuracy": 0.8043633893634496,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 89.3359,
+      "eval_samples_per_second": 15.895,
+      "eval_steps_per_second": 1.992,
+      "step": 840
+    },
+    {
+      "entropy": 0.6409128502011299,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 1.5804955959320068,
+      "learning_rate": 0.0003521523160020035,
+      "loss": 0.6023559093475341,
+      "mean_token_accuracy": 0.8222251623868942,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6983234788594621,
+      "eval_loss": 0.7232135534286499,
+      "eval_mean_token_accuracy": 0.8010654268639811,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 89.2615,
+      "eval_samples_per_second": 15.908,
+      "eval_steps_per_second": 1.994,
+      "step": 860
+    },
+    {
+      "entropy": 0.6151149723678827,
+      "epoch": 2.12070006035003,
+      "grad_norm": 1.1975436210632324,
+      "learning_rate": 0.00035100910966837193,
+      "loss": 0.5803564548492431,
+      "mean_token_accuracy": 0.8256325736641884,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6636555573243773,
+      "eval_loss": 0.7205662727355957,
+      "eval_mean_token_accuracy": 0.8036768369460374,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 89.2957,
+      "eval_samples_per_second": 15.902,
+      "eval_steps_per_second": 1.993,
+      "step": 880
+    },
+    {
+      "entropy": 0.6201727617532015,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9907131195068359,
+      "learning_rate": 0.0003498181779509813,
+      "loss": 0.6101872444152832,
+      "mean_token_accuracy": 0.8196512959897518,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.6473788086617931,
+      "eval_loss": 0.7062397003173828,
+      "eval_mean_token_accuracy": 0.8078362184963869,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 88.6198,
+      "eval_samples_per_second": 16.024,
+      "eval_steps_per_second": 2.009,
+      "step": 900
+    },
+    {
+      "entropy": 0.6097332935780286,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.9055633544921875,
+      "learning_rate": 0.00034857985786942026,
+      "loss": 0.5956094264984131,
+      "mean_token_accuracy": 0.8240575887262821,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.612629868341296,
+      "eval_loss": 0.7106770277023315,
+      "eval_mean_token_accuracy": 0.805324277181304,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 89.4383,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 920
+    },
+    {
+      "entropy": 0.6182701248675585,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.9513736367225647,
+      "learning_rate": 0.00034729449985362404,
+      "loss": 0.6019269466400147,
+      "mean_token_accuracy": 0.8187106013298034,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6437647456533453,
+      "eval_loss": 0.7173502445220947,
+      "eval_mean_token_accuracy": 0.8056439573175451,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.9568,
+      "eval_samples_per_second": 15.963,
+      "eval_steps_per_second": 2.001,
+      "step": 940
+    },
+    {
+      "entropy": 0.6508570514619351,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.9606207013130188,
+      "learning_rate": 0.0003459624676447067,
+      "loss": 0.6296409606933594,
+      "mean_token_accuracy": 0.8167732007801533,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6288247135248077,
+      "eval_loss": 0.7160675525665283,
+      "eval_mean_token_accuracy": 0.8066010689467527,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 89.0044,
+      "eval_samples_per_second": 15.954,
+      "eval_steps_per_second": 2.0,
+      "step": 960
+    },
+    {
+      "entropy": 0.6380081418901682,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.9677181839942932,
+      "learning_rate": 0.000344584138192027,
+      "loss": 0.6210546493530273,
+      "mean_token_accuracy": 0.8186135642230511,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6407805031604981,
+      "eval_loss": 0.7055043578147888,
+      "eval_mean_token_accuracy": 0.8070103157772107,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 89.5606,
+      "eval_samples_per_second": 15.855,
+      "eval_steps_per_second": 1.987,
+      "step": 980
+    },
+    {
+      "entropy": 0.6310251638293266,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 2.1046407222747803,
+      "learning_rate": 0.000343159901546516,
+      "loss": 0.6165830135345459,
+      "mean_token_accuracy": 0.8207855716347694,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6461507470420237,
+      "eval_loss": 0.7292136549949646,
+      "eval_mean_token_accuracy": 0.8030418261383356,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 89.315,
+      "eval_samples_per_second": 15.899,
+      "eval_steps_per_second": 1.993,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6256617344915867,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 1.1768349409103394,
+      "learning_rate": 0.0003416901607502972,
+      "loss": 0.6142709255218506,
+      "mean_token_accuracy": 0.8220669947564602,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6150523993406403,
+      "eval_loss": 0.7019563317298889,
+      "eval_mean_token_accuracy": 0.8094323099998946,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 89.419,
+      "eval_samples_per_second": 15.88,
+      "eval_steps_per_second": 1.991,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6126735735684633,
+      "epoch": 2.506940253470127,
+      "grad_norm": 1.8065496683120728,
+      "learning_rate": 0.00034017533172263055,
+      "loss": 0.6056458473205566,
+      "mean_token_accuracy": 0.8203762136399746,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6544345456562685,
+      "eval_loss": 0.7086517214775085,
+      "eval_mean_token_accuracy": 0.806709805901131,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 88.981,
+      "eval_samples_per_second": 15.958,
+      "eval_steps_per_second": 2.0,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6311754353344441,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 1.0576369762420654,
+      "learning_rate": 0.00033861584314221236,
+      "loss": 0.6176392555236816,
+      "mean_token_accuracy": 0.8177410490810871,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6273799200406235,
+      "eval_loss": 0.708163321018219,
+      "eval_mean_token_accuracy": 0.8032017099053672,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 89.7668,
+      "eval_samples_per_second": 15.819,
+      "eval_steps_per_second": 1.983,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6200249589979648,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.997065007686615,
+      "learning_rate": 0.0003370121363258637,
+      "loss": 0.6203888893127442,
+      "mean_token_accuracy": 0.8196275025606156,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6457576811983344,
+      "eval_loss": 0.6972290277481079,
+      "eval_mean_token_accuracy": 0.8095980849158898,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 89.2027,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 1080
+    },
+    {
+      "entropy": 0.626810473203659,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.865475058555603,
+      "learning_rate": 0.0003353646651036438,
+      "loss": 0.6102585792541504,
+      "mean_token_accuracy": 0.819708751142025,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.6412022431914726,
+      "eval_loss": 0.7036139965057373,
+      "eval_mean_token_accuracy": 0.8070317135098275,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 89.4733,
+      "eval_samples_per_second": 15.871,
+      "eval_steps_per_second": 1.989,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6581619590520859,
+      "epoch": 2.700060350030175,
+      "grad_norm": 1.00307035446167,
+      "learning_rate": 0.00033367389569042064,
+      "loss": 0.6397720336914062,
+      "mean_token_accuracy": 0.81413309648633,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6288736402318719,
+      "eval_loss": 0.6966825723648071,
+      "eval_mean_token_accuracy": 0.8094039669867312,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 88.7341,
+      "eval_samples_per_second": 16.003,
+      "eval_steps_per_second": 2.006,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6399412982165813,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.8589329719543457,
+      "learning_rate": 0.0003319403065539384,
+      "loss": 0.6347338199615479,
+      "mean_token_accuracy": 0.8139534957706929,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.683491913455256,
+      "eval_loss": 0.6984149217605591,
+      "eval_mean_token_accuracy": 0.8083989469522841,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 89.3929,
+      "eval_samples_per_second": 15.885,
+      "eval_steps_per_second": 1.991,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6264835461974144,
+      "epoch": 2.796620398310199,
+      "grad_norm": 0.7807853817939758,
+      "learning_rate": 0.0003301643882794163,
+      "loss": 0.6162629127502441,
+      "mean_token_accuracy": 0.8216292977333068,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6578792159476977,
+      "eval_loss": 0.6921441555023193,
+      "eval_mean_token_accuracy": 0.8094299177775223,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 89.2356,
+      "eval_samples_per_second": 15.913,
+      "eval_steps_per_second": 1.995,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6269129011780024,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 1.1675593852996826,
+      "learning_rate": 0.0003283466434307189,
+      "loss": 0.6235532760620117,
+      "mean_token_accuracy": 0.8188897788524627,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6533248447970058,
+      "eval_loss": 0.696849524974823,
+      "eval_mean_token_accuracy": 0.8102703713968898,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2602,
+      "eval_samples_per_second": 16.089,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.6259873781353236,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.9596486687660217,
+      "learning_rate": 0.00032648758640813655,
+      "loss": 0.6118728160858155,
+      "mean_token_accuracy": 0.8171859130263328,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6212078754821521,
+      "eval_loss": 0.6942870020866394,
+      "eval_mean_token_accuracy": 0.810484343365337,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 89.4375,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 1200
+    },
+    {
+      "entropy": 0.6142537571489811,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 0.9625058770179749,
+      "learning_rate": 0.00032458774330281615,
+      "loss": 0.6060261249542236,
+      "mean_token_accuracy": 0.8223113484680653,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6553665667437436,
+      "eval_loss": 0.6835269331932068,
+      "eval_mean_token_accuracy": 0.8134723968720168,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 89.5092,
+      "eval_samples_per_second": 15.864,
+      "eval_steps_per_second": 1.989,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6301972426474094,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.9657333493232727,
+      "learning_rate": 0.0003226476517478835,
+      "loss": 0.6277759075164795,
+      "mean_token_accuracy": 0.8164266042411328,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6594253118788258,
+      "eval_loss": 0.6820935606956482,
+      "eval_mean_token_accuracy": 0.8132368198941263,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 89.6211,
+      "eval_samples_per_second": 15.844,
+      "eval_steps_per_second": 1.986,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5346100655469027,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8541185259819031,
+      "learning_rate": 0.0003206678607662996,
+      "loss": 0.5261692047119141,
+      "mean_token_accuracy": 0.8382289425119177,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5626645662476507,
+      "eval_loss": 0.7028542160987854,
+      "eval_mean_token_accuracy": 0.8133395572056931,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 89.1267,
+      "eval_samples_per_second": 15.932,
+      "eval_steps_per_second": 1.997,
+      "step": 1260
+    },
+    {
+      "entropy": 0.5001101832836866,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.8208815455436707,
+      "learning_rate": 0.0003186489306154935,
+      "loss": 0.4850132942199707,
+      "mean_token_accuracy": 0.8468983843922615,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5889462046743779,
+      "eval_loss": 0.691015899181366,
+      "eval_mean_token_accuracy": 0.8135226466012805,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.9677,
+      "eval_samples_per_second": 15.961,
+      "eval_steps_per_second": 2.001,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5242766339331866,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 0.7893072962760925,
+      "learning_rate": 0.0003165914326288163,
+      "loss": 0.5061577320098877,
+      "mean_token_accuracy": 0.8425608821213245,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5543691962957382,
+      "eval_loss": 0.6913734078407288,
+      "eval_mean_token_accuracy": 0.8154889680026622,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 89.6963,
+      "eval_samples_per_second": 15.831,
+      "eval_steps_per_second": 1.984,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5327721010893584,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.8992080092430115,
+      "learning_rate": 0.0003144959490538604,
+      "loss": 0.5068036556243897,
+      "mean_token_accuracy": 0.8420542575418949,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5734540803378887,
+      "eval_loss": 0.6948716044425964,
+      "eval_mean_token_accuracy": 0.8137217944257715,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 89.4597,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5247377116233111,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.9291845560073853,
+      "learning_rate": 0.0003123630728876902,
+      "loss": 0.5023272037506104,
+      "mean_token_accuracy": 0.8435182586312294,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5492362935891312,
+      "eval_loss": 0.6980633735656738,
+      "eval_mean_token_accuracy": 0.8155082919624415,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 89.2124,
+      "eval_samples_per_second": 15.917,
+      "eval_steps_per_second": 1.995,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5380499072372913,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 1.0477524995803833,
+      "learning_rate": 0.00031019340770903136,
+      "loss": 0.5264591693878173,
+      "mean_token_accuracy": 0.8391959741711617,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5896503307511297,
+      "eval_loss": 0.6997144222259521,
+      "eval_mean_token_accuracy": 0.8124373062942805,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 89.1147,
+      "eval_samples_per_second": 15.935,
+      "eval_steps_per_second": 1.997,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5460653610527515,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 1.2473069429397583,
+      "learning_rate": 0.00030798756750746476,
+      "loss": 0.5319677829742432,
+      "mean_token_accuracy": 0.8364320226013661,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5659052182114526,
+      "eval_loss": 0.6924049258232117,
+      "eval_mean_token_accuracy": 0.8153053585732921,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 89.6468,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5367407951503992,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 1.2679712772369385,
+      "learning_rate": 0.00030574617650967485,
+      "loss": 0.5190927028656006,
+      "mean_token_accuracy": 0.8399469174444676,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5581759144081159,
+      "eval_loss": 0.6933317184448242,
+      "eval_mean_token_accuracy": 0.8139757862251796,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 89.0749,
+      "eval_samples_per_second": 15.942,
+      "eval_steps_per_second": 1.998,
+      "step": 1400
+    },
+    {
+      "entropy": 0.530736118927598,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.9717612862586975,
+      "learning_rate": 0.0003034698690028009,
+      "loss": 0.512363052368164,
+      "mean_token_accuracy": 0.8419127956032753,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5233335836549823,
+      "eval_loss": 0.6958550214767456,
+      "eval_mean_token_accuracy": 0.8158778501360604,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 88.8664,
+      "eval_samples_per_second": 15.979,
+      "eval_steps_per_second": 2.003,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5327631575986743,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7649208903312683,
+      "learning_rate": 0.00030115928915494116,
+      "loss": 0.5202179431915284,
+      "mean_token_accuracy": 0.8384683802723885,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.5960972657364406,
+      "eval_loss": 0.6984783411026001,
+      "eval_mean_token_accuracy": 0.8095930597085631,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 89.5168,
+      "eval_samples_per_second": 15.863,
+      "eval_steps_per_second": 1.988,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5428155034780502,
+      "epoch": 3.51901025950513,
+      "grad_norm": 1.2790874242782593,
+      "learning_rate": 0.00029881509083286116,
+      "loss": 0.524717378616333,
+      "mean_token_accuracy": 0.8346644505858422,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.550318562917495,
+      "eval_loss": 0.6988471746444702,
+      "eval_mean_token_accuracy": 0.8132969897784544,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 89.2598,
+      "eval_samples_per_second": 15.909,
+      "eval_steps_per_second": 1.994,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5489993002265692,
+      "epoch": 3.567290283645142,
+      "grad_norm": 1.055821418762207,
+      "learning_rate": 0.00029643793741695687,
+      "loss": 0.5421150207519532,
+      "mean_token_accuracy": 0.8345710933208466,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5816821718818685,
+      "eval_loss": 0.6853668689727783,
+      "eval_mean_token_accuracy": 0.8142332360985574,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 89.0424,
+      "eval_samples_per_second": 15.947,
+      "eval_steps_per_second": 1.999,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5451365381479263,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.9027532935142517,
+      "learning_rate": 0.00029402850161352583,
+      "loss": 0.5463609218597412,
+      "mean_token_accuracy": 0.8369908526539802,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5344233209832331,
+      "eval_loss": 0.6794531345367432,
+      "eval_mean_token_accuracy": 0.8177759824843889,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 89.123,
+      "eval_samples_per_second": 15.933,
+      "eval_steps_per_second": 1.997,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5521467242389917,
+      "epoch": 3.663850331925166,
+      "grad_norm": 1.4606311321258545,
+      "learning_rate": 0.0002915874652643997,
+      "loss": 0.5421902179718018,
+      "mean_token_accuracy": 0.8339455373585224,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5807709094513668,
+      "eval_loss": 0.6949691772460938,
+      "eval_mean_token_accuracy": 0.8121141439743256,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.5437,
+      "eval_samples_per_second": 16.037,
+      "eval_steps_per_second": 2.01,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5455610822886229,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7269843220710754,
+      "learning_rate": 0.0002891155191539905,
+      "loss": 0.5407682895660401,
+      "mean_token_accuracy": 0.8393648222088814,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.6032205542151847,
+      "eval_loss": 0.6790974140167236,
+      "eval_mean_token_accuracy": 0.8165463112043531,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 89.517,
+      "eval_samples_per_second": 15.863,
+      "eval_steps_per_second": 1.988,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5474163968116045,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.8148866295814514,
+      "learning_rate": 0.00028661336281380717,
+      "loss": 0.5394441604614257,
+      "mean_token_accuracy": 0.8336028568446636,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5797032238392348,
+      "eval_loss": 0.670870304107666,
+      "eval_mean_token_accuracy": 0.8182691593518417,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 89.4468,
+      "eval_samples_per_second": 15.875,
+      "eval_steps_per_second": 1.99,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5511750865727663,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.3441656827926636,
+      "learning_rate": 0.0002840817043244964,
+      "loss": 0.5378933906555176,
+      "mean_token_accuracy": 0.8349151819944381,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5297858750217417,
+      "eval_loss": 0.680833637714386,
+      "eval_mean_token_accuracy": 0.8180924275617921,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 89.5417,
+      "eval_samples_per_second": 15.859,
+      "eval_steps_per_second": 1.988,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5379613988101483,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9813932776451111,
+      "learning_rate": 0.00028152126011546396,
+      "loss": 0.5340342044830322,
+      "mean_token_accuracy": 0.8346010789275169,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5811898510777549,
+      "eval_loss": 0.6764267683029175,
+      "eval_mean_token_accuracy": 0.8180528236239144,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 88.7831,
+      "eval_samples_per_second": 15.994,
+      "eval_steps_per_second": 2.005,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5607876226305961,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.9617928266525269,
+      "learning_rate": 0.00027893275476213383,
+      "loss": 0.5434338569641113,
+      "mean_token_accuracy": 0.8332898400723934,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5474644318390428,
+      "eval_loss": 0.6740226745605469,
+      "eval_mean_token_accuracy": 0.8185851306058047,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 89.3246,
+      "eval_samples_per_second": 15.897,
+      "eval_steps_per_second": 1.993,
+      "step": 1620
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.6233042105247334e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1640/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1640/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1640/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1640/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1640/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..7ce4b072fcbe5b636867e13937c56fedee3b68b4
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1640/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.02728394274790723,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1640/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1640/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1640/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1640/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1640/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1640/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1640/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1640/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..d4e98591dd813af138359498f4ffb138190852d2
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1640/trainer_state.json
@@ -0,0 +1,1756 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 3.9535304767652386,
+  "eval_steps": 20,
+  "global_step": 1640,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.024795526266098,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.945375919342041,
+      "learning_rate": 1.6698127428345936e-05,
+      "loss": 1.7825420379638672,
+      "mean_token_accuracy": 0.6315708436071873,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.29921497588747,
+      "eval_loss": 1.169131875038147,
+      "eval_mean_token_accuracy": 0.7185706342204233,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 91.7898,
+      "eval_samples_per_second": 15.47,
+      "eval_steps_per_second": 1.939,
+      "step": 20
+    },
+    {
+      "entropy": 1.0346344470977784,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.7916979789733887,
+      "learning_rate": 3.427510366871008e-05,
+      "loss": 0.9480395317077637,
+      "mean_token_accuracy": 0.7543429024517536,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 0.9673695430326997,
+      "eval_loss": 0.8588430285453796,
+      "eval_mean_token_accuracy": 0.7737233002534073,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 89.8219,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 40
+    },
+    {
+      "entropy": 0.8977566614747048,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7972187995910645,
+      "learning_rate": 5.185207990907422e-05,
+      "loss": 0.8029604911804199,
+      "mean_token_accuracy": 0.7818016350269318,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8436836982041263,
+      "eval_loss": 0.7893033623695374,
+      "eval_mean_token_accuracy": 0.7858914984076211,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 89.9349,
+      "eval_samples_per_second": 15.789,
+      "eval_steps_per_second": 1.979,
+      "step": 60
+    },
+    {
+      "entropy": 0.8650321319699288,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.2128729820251465,
+      "learning_rate": 6.942905614943836e-05,
+      "loss": 0.77417893409729,
+      "mean_token_accuracy": 0.7875479467213153,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8209857180547179,
+      "eval_loss": 0.7572280168533325,
+      "eval_mean_token_accuracy": 0.7925970383574453,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 89.9459,
+      "eval_samples_per_second": 15.787,
+      "eval_steps_per_second": 1.979,
+      "step": 80
+    },
+    {
+      "entropy": 0.8096660938113928,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.1254287958145142,
+      "learning_rate": 8.700603238980251e-05,
+      "loss": 0.7170331001281738,
+      "mean_token_accuracy": 0.7994555257260799,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.792780208788561,
+      "eval_loss": 0.7442984580993652,
+      "eval_mean_token_accuracy": 0.7954704573984896,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 90.294,
+      "eval_samples_per_second": 15.726,
+      "eval_steps_per_second": 1.971,
+      "step": 100
+    },
+    {
+      "entropy": 0.7979829199612141,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2140507698059082,
+      "learning_rate": 0.00010458300863016666,
+      "loss": 0.7141861915588379,
+      "mean_token_accuracy": 0.8000014744699001,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.7938692486018277,
+      "eval_loss": 0.7279797196388245,
+      "eval_mean_token_accuracy": 0.7989798734027348,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 89.7189,
+      "eval_samples_per_second": 15.827,
+      "eval_steps_per_second": 1.984,
+      "step": 120
+    },
+    {
+      "entropy": 0.7916347607970238,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.0506458282470703,
+      "learning_rate": 0.0001221599848705308,
+      "loss": 0.7117055416107178,
+      "mean_token_accuracy": 0.7970046654343605,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.8052781063519167,
+      "eval_loss": 0.722576916217804,
+      "eval_mean_token_accuracy": 0.7982530523551984,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 90.0376,
+      "eval_samples_per_second": 15.771,
+      "eval_steps_per_second": 1.977,
+      "step": 140
+    },
+    {
+      "entropy": 0.7880666613578796,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8409207463264465,
+      "learning_rate": 0.00013973696111089492,
+      "loss": 0.7052523136138916,
+      "mean_token_accuracy": 0.8000123649835587,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7787443134891853,
+      "eval_loss": 0.7137772440910339,
+      "eval_mean_token_accuracy": 0.8014209287220173,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 90.2562,
+      "eval_samples_per_second": 15.733,
+      "eval_steps_per_second": 1.972,
+      "step": 160
+    },
+    {
+      "entropy": 0.7696007996797561,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.0182477235794067,
+      "learning_rate": 0.00015731393735125907,
+      "loss": 0.7008797645568847,
+      "mean_token_accuracy": 0.8007259473204613,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.752259682068664,
+      "eval_loss": 0.7171286344528198,
+      "eval_mean_token_accuracy": 0.80052122574174,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 90.0797,
+      "eval_samples_per_second": 15.764,
+      "eval_steps_per_second": 1.976,
+      "step": 180
+    },
+    {
+      "entropy": 0.7828126326203346,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9553632736206055,
+      "learning_rate": 0.0001748909135916232,
+      "loss": 0.7046853542327881,
+      "mean_token_accuracy": 0.7988562889397144,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7389638169427936,
+      "eval_loss": 0.7187935709953308,
+      "eval_mean_token_accuracy": 0.7995899999409579,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 89.9223,
+      "eval_samples_per_second": 15.791,
+      "eval_steps_per_second": 1.979,
+      "step": 200
+    },
+    {
+      "entropy": 0.7909286297857762,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 1.134459376335144,
+      "learning_rate": 0.00019246788983198735,
+      "loss": 0.6989312171936035,
+      "mean_token_accuracy": 0.7991565138101577,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7496965199374082,
+      "eval_loss": 0.7106321454048157,
+      "eval_mean_token_accuracy": 0.8046183442131857,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 89.6899,
+      "eval_samples_per_second": 15.832,
+      "eval_steps_per_second": 1.985,
+      "step": 220
+    },
+    {
+      "entropy": 0.7651963323354721,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.9557585716247559,
+      "learning_rate": 0.0002100448660723515,
+      "loss": 0.6984559059143066,
+      "mean_token_accuracy": 0.8019823372364044,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7615563079212488,
+      "eval_loss": 0.7098539471626282,
+      "eval_mean_token_accuracy": 0.8047233508544022,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 90.3752,
+      "eval_samples_per_second": 15.712,
+      "eval_steps_per_second": 1.97,
+      "step": 240
+    },
+    {
+      "entropy": 0.787507726252079,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.9279311299324036,
+      "learning_rate": 0.00022762184231271568,
+      "loss": 0.708138370513916,
+      "mean_token_accuracy": 0.8013308703899383,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7856640896100676,
+      "eval_loss": 0.7191570401191711,
+      "eval_mean_token_accuracy": 0.8024677249153008,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 93.7336,
+      "eval_samples_per_second": 15.149,
+      "eval_steps_per_second": 1.899,
+      "step": 260
+    },
+    {
+      "entropy": 0.768489234894514,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.7992776036262512,
+      "learning_rate": 0.0002451988185530798,
+      "loss": 0.7035849571228028,
+      "mean_token_accuracy": 0.8031105332076549,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7682498740346244,
+      "eval_loss": 0.7149022221565247,
+      "eval_mean_token_accuracy": 0.7973497995499814,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 89.7309,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 280
+    },
+    {
+      "entropy": 0.7720955617725849,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.1931514739990234,
+      "learning_rate": 0.0002627757947934439,
+      "loss": 0.6965402603149414,
+      "mean_token_accuracy": 0.8040348663926125,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7214708304807042,
+      "eval_loss": 0.7350823879241943,
+      "eval_mean_token_accuracy": 0.7942911925610532,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 89.6451,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 300
+    },
+    {
+      "entropy": 0.7930095426738262,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8994467258453369,
+      "learning_rate": 0.0002803527710338081,
+      "loss": 0.7331492900848389,
+      "mean_token_accuracy": 0.796185302734375,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.8142535488926963,
+      "eval_loss": 0.7237834334373474,
+      "eval_mean_token_accuracy": 0.799495273761535,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 90.6759,
+      "eval_samples_per_second": 15.66,
+      "eval_steps_per_second": 1.963,
+      "step": 320
+    },
+    {
+      "entropy": 0.7825102441012859,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 1.8224239349365234,
+      "learning_rate": 0.00029792974727417223,
+      "loss": 0.7331175804138184,
+      "mean_token_accuracy": 0.7977667152881622,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7327135443017724,
+      "eval_loss": 0.7388784289360046,
+      "eval_mean_token_accuracy": 0.8007041252730938,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 89.7692,
+      "eval_samples_per_second": 15.818,
+      "eval_steps_per_second": 1.983,
+      "step": 340
+    },
+    {
+      "entropy": 0.819331557303667,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 1.0407779216766357,
+      "learning_rate": 0.0003155067235145364,
+      "loss": 0.7456695556640625,
+      "mean_token_accuracy": 0.7936980701982975,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.8065585420372781,
+      "eval_loss": 0.735519528388977,
+      "eval_mean_token_accuracy": 0.7974206296245704,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 89.4336,
+      "eval_samples_per_second": 15.878,
+      "eval_steps_per_second": 1.99,
+      "step": 360
+    },
+    {
+      "entropy": 0.8114653021097183,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.1209490299224854,
+      "learning_rate": 0.0003330836997549005,
+      "loss": 0.7340899467468261,
+      "mean_token_accuracy": 0.7914272703230381,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.8025527285056168,
+      "eval_loss": 0.7295576930046082,
+      "eval_mean_token_accuracy": 0.7990576628218876,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 89.9579,
+      "eval_samples_per_second": 15.785,
+      "eval_steps_per_second": 1.979,
+      "step": 380
+    },
+    {
+      "entropy": 0.8042497783899307,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 1.2437798976898193,
+      "learning_rate": 0.0003506606759952647,
+      "loss": 0.7381744384765625,
+      "mean_token_accuracy": 0.7967362694442273,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7717011336530193,
+      "eval_loss": 0.7436173558235168,
+      "eval_mean_token_accuracy": 0.7975849482450592,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 89.8411,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 400
+    },
+    {
+      "entropy": 0.7987750010056929,
+      "epoch": 1.012070006035003,
+      "grad_norm": 1.2390918731689453,
+      "learning_rate": 0.000364721224843344,
+      "loss": 0.7317790031433106,
+      "mean_token_accuracy": 0.7986087551364651,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7333170604170038,
+      "eval_loss": 0.7456948757171631,
+      "eval_mean_token_accuracy": 0.7964897577682238,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 89.7288,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 420
+    },
+    {
+      "entropy": 0.7372976846992969,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.1552740335464478,
+      "learning_rate": 0.00036468510102269575,
+      "loss": 0.6734249114990234,
+      "mean_token_accuracy": 0.805834824591875,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.7130355526891987,
+      "eval_loss": 0.7555333375930786,
+      "eval_mean_token_accuracy": 0.7961652781186479,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 89.8713,
+      "eval_samples_per_second": 15.8,
+      "eval_steps_per_second": 1.981,
+      "step": 440
+    },
+    {
+      "entropy": 0.7465668715536594,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 1.1991469860076904,
+      "learning_rate": 0.0003645973816745066,
+      "loss": 0.6871879577636719,
+      "mean_token_accuracy": 0.8042450234293937,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7772979542110743,
+      "eval_loss": 0.7561826705932617,
+      "eval_mean_token_accuracy": 0.7953876427720102,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 89.8395,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 460
+    },
+    {
+      "entropy": 0.7477855801582336,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 1.325486660003662,
+      "learning_rate": 0.00036445809162231435,
+      "loss": 0.6836830139160156,
+      "mean_token_accuracy": 0.8021619468927383,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.7098110676481483,
+      "eval_loss": 0.758007287979126,
+      "eval_mean_token_accuracy": 0.7909953189030122,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 89.6101,
+      "eval_samples_per_second": 15.846,
+      "eval_steps_per_second": 1.986,
+      "step": 480
+    },
+    {
+      "entropy": 0.7632691070437432,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 1.0122262239456177,
+      "learning_rate": 0.0003642672702835562,
+      "loss": 0.7071213245391845,
+      "mean_token_accuracy": 0.8000680930912495,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7645382720432924,
+      "eval_loss": 0.7492606043815613,
+      "eval_mean_token_accuracy": 0.7976858676149604,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 89.6092,
+      "eval_samples_per_second": 15.847,
+      "eval_steps_per_second": 1.986,
+      "step": 500
+    },
+    {
+      "entropy": 0.7623726457357407,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.367774486541748,
+      "learning_rate": 0.00036402497165841384,
+      "loss": 0.6965175151824952,
+      "mean_token_accuracy": 0.8016501650214195,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7868083715438843,
+      "eval_loss": 0.7481449842453003,
+      "eval_mean_token_accuracy": 0.799330582779445,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 89.8241,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 520
+    },
+    {
+      "entropy": 0.7535207405686378,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 1.213614821434021,
+      "learning_rate": 0.00036373126431453207,
+      "loss": 0.7078310489654541,
+      "mean_token_accuracy": 0.7998907208442688,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7829091950748743,
+      "eval_loss": 0.7416072487831116,
+      "eval_mean_token_accuracy": 0.7969146250339036,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 89.4607,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 540
+    },
+    {
+      "entropy": 0.7527363449335098,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 1.0929982662200928,
+      "learning_rate": 0.00036338623136761495,
+      "loss": 0.7058369636535644,
+      "mean_token_accuracy": 0.7997087433934211,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.728419938114252,
+      "eval_loss": 0.744873046875,
+      "eval_mean_token_accuracy": 0.7966634296299366,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 89.6597,
+      "eval_samples_per_second": 15.838,
+      "eval_steps_per_second": 1.985,
+      "step": 560
+    },
+    {
+      "entropy": 0.7406099259853363,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.409364104270935,
+      "learning_rate": 0.00036298997045790515,
+      "loss": 0.6902852058410645,
+      "mean_token_accuracy": 0.805386807769537,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.6991632758231645,
+      "eval_loss": 0.7480319738388062,
+      "eval_mean_token_accuracy": 0.7970068006033308,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 89.8329,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 580
+    },
+    {
+      "entropy": 0.7459449704736472,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.7053111791610718,
+      "learning_rate": 0.00036254259372255277,
+      "loss": 0.7044640064239502,
+      "mean_token_accuracy": 0.8005018755793571,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7454566660891758,
+      "eval_loss": 0.7551484704017639,
+      "eval_mean_token_accuracy": 0.7897093731365846,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 89.8313,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 600
+    },
+    {
+      "entropy": 0.7621391348540782,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 1.174443244934082,
+      "learning_rate": 0.000362044227763882,
+      "loss": 0.7111573696136475,
+      "mean_token_accuracy": 0.798752411454916,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7664387949397055,
+      "eval_loss": 0.7365804314613342,
+      "eval_mean_token_accuracy": 0.7963101200843126,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 89.9623,
+      "eval_samples_per_second": 15.784,
+      "eval_steps_per_second": 1.979,
+      "step": 620
+    },
+    {
+      "entropy": 0.7424944408237935,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.2898184061050415,
+      "learning_rate": 0.000361495013613564,
+      "loss": 0.6913110733032226,
+      "mean_token_accuracy": 0.8009015507996082,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7255479529332579,
+      "eval_loss": 0.7345991730690002,
+      "eval_mean_token_accuracy": 0.8013592945056015,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 89.817,
+      "eval_samples_per_second": 15.81,
+      "eval_steps_per_second": 1.982,
+      "step": 640
+    },
+    {
+      "entropy": 0.7303367637097835,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.1310476064682007,
+      "learning_rate": 0.00036089510669270666,
+      "loss": 0.6875617027282714,
+      "mean_token_accuracy": 0.8035397171974182,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7101308248016271,
+      "eval_loss": 0.7394917011260986,
+      "eval_mean_token_accuracy": 0.7990792942850777,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 89.7431,
+      "eval_samples_per_second": 15.823,
+      "eval_steps_per_second": 1.983,
+      "step": 660
+    },
+    {
+      "entropy": 0.7387678384780884,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.9645389914512634,
+      "learning_rate": 0.0003602446767678725,
+      "loss": 0.6911204338073731,
+      "mean_token_accuracy": 0.8056333385407924,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7281073737010527,
+      "eval_loss": 0.7376708984375,
+      "eval_mean_token_accuracy": 0.7997356415464637,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 90.0338,
+      "eval_samples_per_second": 15.772,
+      "eval_steps_per_second": 1.977,
+      "step": 680
+    },
+    {
+      "entropy": 0.7449554048478604,
+      "epoch": 1.687990343995172,
+      "grad_norm": 1.9069629907608032,
+      "learning_rate": 0.0003595439079030364,
+      "loss": 0.7042286396026611,
+      "mean_token_accuracy": 0.8018639378249646,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7022633817088738,
+      "eval_loss": 0.7320939898490906,
+      "eval_mean_token_accuracy": 0.8006051541044471,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 89.8948,
+      "eval_samples_per_second": 15.796,
+      "eval_steps_per_second": 1.98,
+      "step": 700
+    },
+    {
+      "entropy": 0.723249051719904,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.8333877325057983,
+      "learning_rate": 0.00035879299840749777,
+      "loss": 0.7043714046478271,
+      "mean_token_accuracy": 0.8028345607221127,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.672436946898364,
+      "eval_loss": 0.713233470916748,
+      "eval_mean_token_accuracy": 0.8047760577684038,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 89.6394,
+      "eval_samples_per_second": 15.841,
+      "eval_steps_per_second": 1.986,
+      "step": 720
+    },
+    {
+      "entropy": 0.730014206469059,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.4580037593841553,
+      "learning_rate": 0.00035799216077976137,
+      "loss": 0.6985176563262939,
+      "mean_token_accuracy": 0.8015547141432762,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.6908873795123582,
+      "eval_loss": 0.7198202013969421,
+      "eval_mean_token_accuracy": 0.8024131682481659,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 89.3771,
+      "eval_samples_per_second": 15.888,
+      "eval_steps_per_second": 1.992,
+      "step": 740
+    },
+    {
+      "entropy": 0.7219605796039105,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.51682710647583,
+      "learning_rate": 0.000357141621647403,
+      "loss": 0.6844424724578857,
+      "mean_token_accuracy": 0.8032813094556331,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7113401109582922,
+      "eval_loss": 0.7195309400558472,
+      "eval_mean_token_accuracy": 0.8019187035185568,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 89.4658,
+      "eval_samples_per_second": 15.872,
+      "eval_steps_per_second": 1.99,
+      "step": 760
+    },
+    {
+      "entropy": 0.729826743900776,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 1.0838000774383545,
+      "learning_rate": 0.0003562416217029361,
+      "loss": 0.685858678817749,
+      "mean_token_accuracy": 0.8050567395985126,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.649370267484965,
+      "eval_loss": 0.7177029252052307,
+      "eval_mean_token_accuracy": 0.8050174043419656,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 89.2287,
+      "eval_samples_per_second": 15.914,
+      "eval_steps_per_second": 1.995,
+      "step": 780
+    },
+    {
+      "entropy": 0.6799364153295755,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 1.1364924907684326,
+      "learning_rate": 0.00035529241563569903,
+      "loss": 0.6714428901672364,
+      "mean_token_accuracy": 0.8087341658771038,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6603799047095052,
+      "eval_loss": 0.7201809287071228,
+      "eval_mean_token_accuracy": 0.802863868769635,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 89.2718,
+      "eval_samples_per_second": 15.906,
+      "eval_steps_per_second": 1.994,
+      "step": 800
+    },
+    {
+      "entropy": 0.7078610375523567,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 1.2191942930221558,
+      "learning_rate": 0.0003542942720597808,
+      "loss": 0.6996590614318847,
+      "mean_token_accuracy": 0.8049512051045895,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.7240545264120852,
+      "eval_loss": 0.7301309108734131,
+      "eval_mean_token_accuracy": 0.8015973166133581,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 89.3438,
+      "eval_samples_per_second": 15.894,
+      "eval_steps_per_second": 1.992,
+      "step": 820
+    },
+    {
+      "entropy": 0.6557391556826505,
+      "epoch": 2.024140012070006,
+      "grad_norm": 1.36203134059906,
+      "learning_rate": 0.0003532474734380064,
+      "loss": 0.6327113628387451,
+      "mean_token_accuracy": 0.8142829847026181,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6573808648613062,
+      "eval_loss": 0.7334365844726562,
+      "eval_mean_token_accuracy": 0.8043633893634496,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 89.3359,
+      "eval_samples_per_second": 15.895,
+      "eval_steps_per_second": 1.992,
+      "step": 840
+    },
+    {
+      "entropy": 0.6409128502011299,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 1.5804955959320068,
+      "learning_rate": 0.0003521523160020035,
+      "loss": 0.6023559093475341,
+      "mean_token_accuracy": 0.8222251623868942,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6983234788594621,
+      "eval_loss": 0.7232135534286499,
+      "eval_mean_token_accuracy": 0.8010654268639811,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 89.2615,
+      "eval_samples_per_second": 15.908,
+      "eval_steps_per_second": 1.994,
+      "step": 860
+    },
+    {
+      "entropy": 0.6151149723678827,
+      "epoch": 2.12070006035003,
+      "grad_norm": 1.1975436210632324,
+      "learning_rate": 0.00035100910966837193,
+      "loss": 0.5803564548492431,
+      "mean_token_accuracy": 0.8256325736641884,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6636555573243773,
+      "eval_loss": 0.7205662727355957,
+      "eval_mean_token_accuracy": 0.8036768369460374,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 89.2957,
+      "eval_samples_per_second": 15.902,
+      "eval_steps_per_second": 1.993,
+      "step": 880
+    },
+    {
+      "entropy": 0.6201727617532015,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9907131195068359,
+      "learning_rate": 0.0003498181779509813,
+      "loss": 0.6101872444152832,
+      "mean_token_accuracy": 0.8196512959897518,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.6473788086617931,
+      "eval_loss": 0.7062397003173828,
+      "eval_mean_token_accuracy": 0.8078362184963869,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 88.6198,
+      "eval_samples_per_second": 16.024,
+      "eval_steps_per_second": 2.009,
+      "step": 900
+    },
+    {
+      "entropy": 0.6097332935780286,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.9055633544921875,
+      "learning_rate": 0.00034857985786942026,
+      "loss": 0.5956094264984131,
+      "mean_token_accuracy": 0.8240575887262821,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.612629868341296,
+      "eval_loss": 0.7106770277023315,
+      "eval_mean_token_accuracy": 0.805324277181304,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 89.4383,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 920
+    },
+    {
+      "entropy": 0.6182701248675585,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.9513736367225647,
+      "learning_rate": 0.00034729449985362404,
+      "loss": 0.6019269466400147,
+      "mean_token_accuracy": 0.8187106013298034,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6437647456533453,
+      "eval_loss": 0.7173502445220947,
+      "eval_mean_token_accuracy": 0.8056439573175451,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.9568,
+      "eval_samples_per_second": 15.963,
+      "eval_steps_per_second": 2.001,
+      "step": 940
+    },
+    {
+      "entropy": 0.6508570514619351,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.9606207013130188,
+      "learning_rate": 0.0003459624676447067,
+      "loss": 0.6296409606933594,
+      "mean_token_accuracy": 0.8167732007801533,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6288247135248077,
+      "eval_loss": 0.7160675525665283,
+      "eval_mean_token_accuracy": 0.8066010689467527,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 89.0044,
+      "eval_samples_per_second": 15.954,
+      "eval_steps_per_second": 2.0,
+      "step": 960
+    },
+    {
+      "entropy": 0.6380081418901682,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.9677181839942932,
+      "learning_rate": 0.000344584138192027,
+      "loss": 0.6210546493530273,
+      "mean_token_accuracy": 0.8186135642230511,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6407805031604981,
+      "eval_loss": 0.7055043578147888,
+      "eval_mean_token_accuracy": 0.8070103157772107,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 89.5606,
+      "eval_samples_per_second": 15.855,
+      "eval_steps_per_second": 1.987,
+      "step": 980
+    },
+    {
+      "entropy": 0.6310251638293266,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 2.1046407222747803,
+      "learning_rate": 0.000343159901546516,
+      "loss": 0.6165830135345459,
+      "mean_token_accuracy": 0.8207855716347694,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6461507470420237,
+      "eval_loss": 0.7292136549949646,
+      "eval_mean_token_accuracy": 0.8030418261383356,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 89.315,
+      "eval_samples_per_second": 15.899,
+      "eval_steps_per_second": 1.993,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6256617344915867,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 1.1768349409103394,
+      "learning_rate": 0.0003416901607502972,
+      "loss": 0.6142709255218506,
+      "mean_token_accuracy": 0.8220669947564602,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6150523993406403,
+      "eval_loss": 0.7019563317298889,
+      "eval_mean_token_accuracy": 0.8094323099998946,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 89.419,
+      "eval_samples_per_second": 15.88,
+      "eval_steps_per_second": 1.991,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6126735735684633,
+      "epoch": 2.506940253470127,
+      "grad_norm": 1.8065496683120728,
+      "learning_rate": 0.00034017533172263055,
+      "loss": 0.6056458473205566,
+      "mean_token_accuracy": 0.8203762136399746,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6544345456562685,
+      "eval_loss": 0.7086517214775085,
+      "eval_mean_token_accuracy": 0.806709805901131,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 88.981,
+      "eval_samples_per_second": 15.958,
+      "eval_steps_per_second": 2.0,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6311754353344441,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 1.0576369762420654,
+      "learning_rate": 0.00033861584314221236,
+      "loss": 0.6176392555236816,
+      "mean_token_accuracy": 0.8177410490810871,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6273799200406235,
+      "eval_loss": 0.708163321018219,
+      "eval_mean_token_accuracy": 0.8032017099053672,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 89.7668,
+      "eval_samples_per_second": 15.819,
+      "eval_steps_per_second": 1.983,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6200249589979648,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.997065007686615,
+      "learning_rate": 0.0003370121363258637,
+      "loss": 0.6203888893127442,
+      "mean_token_accuracy": 0.8196275025606156,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6457576811983344,
+      "eval_loss": 0.6972290277481079,
+      "eval_mean_token_accuracy": 0.8095980849158898,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 89.2027,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 1080
+    },
+    {
+      "entropy": 0.626810473203659,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.865475058555603,
+      "learning_rate": 0.0003353646651036438,
+      "loss": 0.6102585792541504,
+      "mean_token_accuracy": 0.819708751142025,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.6412022431914726,
+      "eval_loss": 0.7036139965057373,
+      "eval_mean_token_accuracy": 0.8070317135098275,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 89.4733,
+      "eval_samples_per_second": 15.871,
+      "eval_steps_per_second": 1.989,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6581619590520859,
+      "epoch": 2.700060350030175,
+      "grad_norm": 1.00307035446167,
+      "learning_rate": 0.00033367389569042064,
+      "loss": 0.6397720336914062,
+      "mean_token_accuracy": 0.81413309648633,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6288736402318719,
+      "eval_loss": 0.6966825723648071,
+      "eval_mean_token_accuracy": 0.8094039669867312,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 88.7341,
+      "eval_samples_per_second": 16.003,
+      "eval_steps_per_second": 2.006,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6399412982165813,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.8589329719543457,
+      "learning_rate": 0.0003319403065539384,
+      "loss": 0.6347338199615479,
+      "mean_token_accuracy": 0.8139534957706929,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.683491913455256,
+      "eval_loss": 0.6984149217605591,
+      "eval_mean_token_accuracy": 0.8083989469522841,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 89.3929,
+      "eval_samples_per_second": 15.885,
+      "eval_steps_per_second": 1.991,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6264835461974144,
+      "epoch": 2.796620398310199,
+      "grad_norm": 0.7807853817939758,
+      "learning_rate": 0.0003301643882794163,
+      "loss": 0.6162629127502441,
+      "mean_token_accuracy": 0.8216292977333068,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6578792159476977,
+      "eval_loss": 0.6921441555023193,
+      "eval_mean_token_accuracy": 0.8094299177775223,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 89.2356,
+      "eval_samples_per_second": 15.913,
+      "eval_steps_per_second": 1.995,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6269129011780024,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 1.1675593852996826,
+      "learning_rate": 0.0003283466434307189,
+      "loss": 0.6235532760620117,
+      "mean_token_accuracy": 0.8188897788524627,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6533248447970058,
+      "eval_loss": 0.696849524974823,
+      "eval_mean_token_accuracy": 0.8102703713968898,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2602,
+      "eval_samples_per_second": 16.089,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.6259873781353236,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.9596486687660217,
+      "learning_rate": 0.00032648758640813655,
+      "loss": 0.6118728160858155,
+      "mean_token_accuracy": 0.8171859130263328,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6212078754821521,
+      "eval_loss": 0.6942870020866394,
+      "eval_mean_token_accuracy": 0.810484343365337,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 89.4375,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 1200
+    },
+    {
+      "entropy": 0.6142537571489811,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 0.9625058770179749,
+      "learning_rate": 0.00032458774330281615,
+      "loss": 0.6060261249542236,
+      "mean_token_accuracy": 0.8223113484680653,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6553665667437436,
+      "eval_loss": 0.6835269331932068,
+      "eval_mean_token_accuracy": 0.8134723968720168,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 89.5092,
+      "eval_samples_per_second": 15.864,
+      "eval_steps_per_second": 1.989,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6301972426474094,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.9657333493232727,
+      "learning_rate": 0.0003226476517478835,
+      "loss": 0.6277759075164795,
+      "mean_token_accuracy": 0.8164266042411328,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6594253118788258,
+      "eval_loss": 0.6820935606956482,
+      "eval_mean_token_accuracy": 0.8132368198941263,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 89.6211,
+      "eval_samples_per_second": 15.844,
+      "eval_steps_per_second": 1.986,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5346100655469027,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8541185259819031,
+      "learning_rate": 0.0003206678607662996,
+      "loss": 0.5261692047119141,
+      "mean_token_accuracy": 0.8382289425119177,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5626645662476507,
+      "eval_loss": 0.7028542160987854,
+      "eval_mean_token_accuracy": 0.8133395572056931,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 89.1267,
+      "eval_samples_per_second": 15.932,
+      "eval_steps_per_second": 1.997,
+      "step": 1260
+    },
+    {
+      "entropy": 0.5001101832836866,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.8208815455436707,
+      "learning_rate": 0.0003186489306154935,
+      "loss": 0.4850132942199707,
+      "mean_token_accuracy": 0.8468983843922615,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5889462046743779,
+      "eval_loss": 0.691015899181366,
+      "eval_mean_token_accuracy": 0.8135226466012805,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.9677,
+      "eval_samples_per_second": 15.961,
+      "eval_steps_per_second": 2.001,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5242766339331866,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 0.7893072962760925,
+      "learning_rate": 0.0003165914326288163,
+      "loss": 0.5061577320098877,
+      "mean_token_accuracy": 0.8425608821213245,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5543691962957382,
+      "eval_loss": 0.6913734078407288,
+      "eval_mean_token_accuracy": 0.8154889680026622,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 89.6963,
+      "eval_samples_per_second": 15.831,
+      "eval_steps_per_second": 1.984,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5327721010893584,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.8992080092430115,
+      "learning_rate": 0.0003144959490538604,
+      "loss": 0.5068036556243897,
+      "mean_token_accuracy": 0.8420542575418949,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5734540803378887,
+      "eval_loss": 0.6948716044425964,
+      "eval_mean_token_accuracy": 0.8137217944257715,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 89.4597,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5247377116233111,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.9291845560073853,
+      "learning_rate": 0.0003123630728876902,
+      "loss": 0.5023272037506104,
+      "mean_token_accuracy": 0.8435182586312294,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5492362935891312,
+      "eval_loss": 0.6980633735656738,
+      "eval_mean_token_accuracy": 0.8155082919624415,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 89.2124,
+      "eval_samples_per_second": 15.917,
+      "eval_steps_per_second": 1.995,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5380499072372913,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 1.0477524995803833,
+      "learning_rate": 0.00031019340770903136,
+      "loss": 0.5264591693878173,
+      "mean_token_accuracy": 0.8391959741711617,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5896503307511297,
+      "eval_loss": 0.6997144222259521,
+      "eval_mean_token_accuracy": 0.8124373062942805,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 89.1147,
+      "eval_samples_per_second": 15.935,
+      "eval_steps_per_second": 1.997,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5460653610527515,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 1.2473069429397583,
+      "learning_rate": 0.00030798756750746476,
+      "loss": 0.5319677829742432,
+      "mean_token_accuracy": 0.8364320226013661,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5659052182114526,
+      "eval_loss": 0.6924049258232117,
+      "eval_mean_token_accuracy": 0.8153053585732921,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 89.6468,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5367407951503992,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 1.2679712772369385,
+      "learning_rate": 0.00030574617650967485,
+      "loss": 0.5190927028656006,
+      "mean_token_accuracy": 0.8399469174444676,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5581759144081159,
+      "eval_loss": 0.6933317184448242,
+      "eval_mean_token_accuracy": 0.8139757862251796,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 89.0749,
+      "eval_samples_per_second": 15.942,
+      "eval_steps_per_second": 1.998,
+      "step": 1400
+    },
+    {
+      "entropy": 0.530736118927598,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.9717612862586975,
+      "learning_rate": 0.0003034698690028009,
+      "loss": 0.512363052368164,
+      "mean_token_accuracy": 0.8419127956032753,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5233335836549823,
+      "eval_loss": 0.6958550214767456,
+      "eval_mean_token_accuracy": 0.8158778501360604,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 88.8664,
+      "eval_samples_per_second": 15.979,
+      "eval_steps_per_second": 2.003,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5327631575986743,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7649208903312683,
+      "learning_rate": 0.00030115928915494116,
+      "loss": 0.5202179431915284,
+      "mean_token_accuracy": 0.8384683802723885,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.5960972657364406,
+      "eval_loss": 0.6984783411026001,
+      "eval_mean_token_accuracy": 0.8095930597085631,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 89.5168,
+      "eval_samples_per_second": 15.863,
+      "eval_steps_per_second": 1.988,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5428155034780502,
+      "epoch": 3.51901025950513,
+      "grad_norm": 1.2790874242782593,
+      "learning_rate": 0.00029881509083286116,
+      "loss": 0.524717378616333,
+      "mean_token_accuracy": 0.8346644505858422,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.550318562917495,
+      "eval_loss": 0.6988471746444702,
+      "eval_mean_token_accuracy": 0.8132969897784544,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 89.2598,
+      "eval_samples_per_second": 15.909,
+      "eval_steps_per_second": 1.994,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5489993002265692,
+      "epoch": 3.567290283645142,
+      "grad_norm": 1.055821418762207,
+      "learning_rate": 0.00029643793741695687,
+      "loss": 0.5421150207519532,
+      "mean_token_accuracy": 0.8345710933208466,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5816821718818685,
+      "eval_loss": 0.6853668689727783,
+      "eval_mean_token_accuracy": 0.8142332360985574,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 89.0424,
+      "eval_samples_per_second": 15.947,
+      "eval_steps_per_second": 1.999,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5451365381479263,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.9027532935142517,
+      "learning_rate": 0.00029402850161352583,
+      "loss": 0.5463609218597412,
+      "mean_token_accuracy": 0.8369908526539802,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5344233209832331,
+      "eval_loss": 0.6794531345367432,
+      "eval_mean_token_accuracy": 0.8177759824843889,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 89.123,
+      "eval_samples_per_second": 15.933,
+      "eval_steps_per_second": 1.997,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5521467242389917,
+      "epoch": 3.663850331925166,
+      "grad_norm": 1.4606311321258545,
+      "learning_rate": 0.0002915874652643997,
+      "loss": 0.5421902179718018,
+      "mean_token_accuracy": 0.8339455373585224,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5807709094513668,
+      "eval_loss": 0.6949691772460938,
+      "eval_mean_token_accuracy": 0.8121141439743256,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.5437,
+      "eval_samples_per_second": 16.037,
+      "eval_steps_per_second": 2.01,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5455610822886229,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7269843220710754,
+      "learning_rate": 0.0002891155191539905,
+      "loss": 0.5407682895660401,
+      "mean_token_accuracy": 0.8393648222088814,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.6032205542151847,
+      "eval_loss": 0.6790974140167236,
+      "eval_mean_token_accuracy": 0.8165463112043531,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 89.517,
+      "eval_samples_per_second": 15.863,
+      "eval_steps_per_second": 1.988,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5474163968116045,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.8148866295814514,
+      "learning_rate": 0.00028661336281380717,
+      "loss": 0.5394441604614257,
+      "mean_token_accuracy": 0.8336028568446636,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5797032238392348,
+      "eval_loss": 0.670870304107666,
+      "eval_mean_token_accuracy": 0.8182691593518417,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 89.4468,
+      "eval_samples_per_second": 15.875,
+      "eval_steps_per_second": 1.99,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5511750865727663,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.3441656827926636,
+      "learning_rate": 0.0002840817043244964,
+      "loss": 0.5378933906555176,
+      "mean_token_accuracy": 0.8349151819944381,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5297858750217417,
+      "eval_loss": 0.680833637714386,
+      "eval_mean_token_accuracy": 0.8180924275617921,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 89.5417,
+      "eval_samples_per_second": 15.859,
+      "eval_steps_per_second": 1.988,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5379613988101483,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9813932776451111,
+      "learning_rate": 0.00028152126011546396,
+      "loss": 0.5340342044830322,
+      "mean_token_accuracy": 0.8346010789275169,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5811898510777549,
+      "eval_loss": 0.6764267683029175,
+      "eval_mean_token_accuracy": 0.8180528236239144,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 88.7831,
+      "eval_samples_per_second": 15.994,
+      "eval_steps_per_second": 2.005,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5607876226305961,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.9617928266525269,
+      "learning_rate": 0.00027893275476213383,
+      "loss": 0.5434338569641113,
+      "mean_token_accuracy": 0.8332898400723934,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5474644318390428,
+      "eval_loss": 0.6740226745605469,
+      "eval_mean_token_accuracy": 0.8185851306058047,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 89.3246,
+      "eval_samples_per_second": 15.897,
+      "eval_steps_per_second": 1.993,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5354843523353339,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.5320990085601807,
+      "learning_rate": 0.0002763169207809021,
+      "loss": 0.5251852989196777,
+      "mean_token_accuracy": 0.8416872084140777,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.5436856410141742,
+      "eval_loss": 0.6675190925598145,
+      "eval_mean_token_accuracy": 0.8202396507343549,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 89.6328,
+      "eval_samples_per_second": 15.842,
+      "eval_steps_per_second": 1.986,
+      "step": 1640
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.6434764490183475e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1660/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1660/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1660/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1660/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1660/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..7ce4b072fcbe5b636867e13937c56fedee3b68b4
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1660/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.02728394274790723,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1660/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1660/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1660/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1660/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1660/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1660/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1660/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1660/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..da84d5f35ebd04d3e3010e4a2ea6c306fadc3b5f
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1660/trainer_state.json
@@ -0,0 +1,1777 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 4.0,
+  "eval_steps": 20,
+  "global_step": 1660,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.024795526266098,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.945375919342041,
+      "learning_rate": 1.6698127428345936e-05,
+      "loss": 1.7825420379638672,
+      "mean_token_accuracy": 0.6315708436071873,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.29921497588747,
+      "eval_loss": 1.169131875038147,
+      "eval_mean_token_accuracy": 0.7185706342204233,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 91.7898,
+      "eval_samples_per_second": 15.47,
+      "eval_steps_per_second": 1.939,
+      "step": 20
+    },
+    {
+      "entropy": 1.0346344470977784,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.7916979789733887,
+      "learning_rate": 3.427510366871008e-05,
+      "loss": 0.9480395317077637,
+      "mean_token_accuracy": 0.7543429024517536,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 0.9673695430326997,
+      "eval_loss": 0.8588430285453796,
+      "eval_mean_token_accuracy": 0.7737233002534073,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 89.8219,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 40
+    },
+    {
+      "entropy": 0.8977566614747048,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7972187995910645,
+      "learning_rate": 5.185207990907422e-05,
+      "loss": 0.8029604911804199,
+      "mean_token_accuracy": 0.7818016350269318,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8436836982041263,
+      "eval_loss": 0.7893033623695374,
+      "eval_mean_token_accuracy": 0.7858914984076211,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 89.9349,
+      "eval_samples_per_second": 15.789,
+      "eval_steps_per_second": 1.979,
+      "step": 60
+    },
+    {
+      "entropy": 0.8650321319699288,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.2128729820251465,
+      "learning_rate": 6.942905614943836e-05,
+      "loss": 0.77417893409729,
+      "mean_token_accuracy": 0.7875479467213153,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8209857180547179,
+      "eval_loss": 0.7572280168533325,
+      "eval_mean_token_accuracy": 0.7925970383574453,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 89.9459,
+      "eval_samples_per_second": 15.787,
+      "eval_steps_per_second": 1.979,
+      "step": 80
+    },
+    {
+      "entropy": 0.8096660938113928,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.1254287958145142,
+      "learning_rate": 8.700603238980251e-05,
+      "loss": 0.7170331001281738,
+      "mean_token_accuracy": 0.7994555257260799,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.792780208788561,
+      "eval_loss": 0.7442984580993652,
+      "eval_mean_token_accuracy": 0.7954704573984896,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 90.294,
+      "eval_samples_per_second": 15.726,
+      "eval_steps_per_second": 1.971,
+      "step": 100
+    },
+    {
+      "entropy": 0.7979829199612141,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2140507698059082,
+      "learning_rate": 0.00010458300863016666,
+      "loss": 0.7141861915588379,
+      "mean_token_accuracy": 0.8000014744699001,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.7938692486018277,
+      "eval_loss": 0.7279797196388245,
+      "eval_mean_token_accuracy": 0.7989798734027348,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 89.7189,
+      "eval_samples_per_second": 15.827,
+      "eval_steps_per_second": 1.984,
+      "step": 120
+    },
+    {
+      "entropy": 0.7916347607970238,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.0506458282470703,
+      "learning_rate": 0.0001221599848705308,
+      "loss": 0.7117055416107178,
+      "mean_token_accuracy": 0.7970046654343605,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.8052781063519167,
+      "eval_loss": 0.722576916217804,
+      "eval_mean_token_accuracy": 0.7982530523551984,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 90.0376,
+      "eval_samples_per_second": 15.771,
+      "eval_steps_per_second": 1.977,
+      "step": 140
+    },
+    {
+      "entropy": 0.7880666613578796,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8409207463264465,
+      "learning_rate": 0.00013973696111089492,
+      "loss": 0.7052523136138916,
+      "mean_token_accuracy": 0.8000123649835587,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7787443134891853,
+      "eval_loss": 0.7137772440910339,
+      "eval_mean_token_accuracy": 0.8014209287220173,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 90.2562,
+      "eval_samples_per_second": 15.733,
+      "eval_steps_per_second": 1.972,
+      "step": 160
+    },
+    {
+      "entropy": 0.7696007996797561,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.0182477235794067,
+      "learning_rate": 0.00015731393735125907,
+      "loss": 0.7008797645568847,
+      "mean_token_accuracy": 0.8007259473204613,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.752259682068664,
+      "eval_loss": 0.7171286344528198,
+      "eval_mean_token_accuracy": 0.80052122574174,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 90.0797,
+      "eval_samples_per_second": 15.764,
+      "eval_steps_per_second": 1.976,
+      "step": 180
+    },
+    {
+      "entropy": 0.7828126326203346,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9553632736206055,
+      "learning_rate": 0.0001748909135916232,
+      "loss": 0.7046853542327881,
+      "mean_token_accuracy": 0.7988562889397144,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7389638169427936,
+      "eval_loss": 0.7187935709953308,
+      "eval_mean_token_accuracy": 0.7995899999409579,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 89.9223,
+      "eval_samples_per_second": 15.791,
+      "eval_steps_per_second": 1.979,
+      "step": 200
+    },
+    {
+      "entropy": 0.7909286297857762,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 1.134459376335144,
+      "learning_rate": 0.00019246788983198735,
+      "loss": 0.6989312171936035,
+      "mean_token_accuracy": 0.7991565138101577,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7496965199374082,
+      "eval_loss": 0.7106321454048157,
+      "eval_mean_token_accuracy": 0.8046183442131857,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 89.6899,
+      "eval_samples_per_second": 15.832,
+      "eval_steps_per_second": 1.985,
+      "step": 220
+    },
+    {
+      "entropy": 0.7651963323354721,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.9557585716247559,
+      "learning_rate": 0.0002100448660723515,
+      "loss": 0.6984559059143066,
+      "mean_token_accuracy": 0.8019823372364044,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7615563079212488,
+      "eval_loss": 0.7098539471626282,
+      "eval_mean_token_accuracy": 0.8047233508544022,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 90.3752,
+      "eval_samples_per_second": 15.712,
+      "eval_steps_per_second": 1.97,
+      "step": 240
+    },
+    {
+      "entropy": 0.787507726252079,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.9279311299324036,
+      "learning_rate": 0.00022762184231271568,
+      "loss": 0.708138370513916,
+      "mean_token_accuracy": 0.8013308703899383,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7856640896100676,
+      "eval_loss": 0.7191570401191711,
+      "eval_mean_token_accuracy": 0.8024677249153008,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 93.7336,
+      "eval_samples_per_second": 15.149,
+      "eval_steps_per_second": 1.899,
+      "step": 260
+    },
+    {
+      "entropy": 0.768489234894514,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.7992776036262512,
+      "learning_rate": 0.0002451988185530798,
+      "loss": 0.7035849571228028,
+      "mean_token_accuracy": 0.8031105332076549,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7682498740346244,
+      "eval_loss": 0.7149022221565247,
+      "eval_mean_token_accuracy": 0.7973497995499814,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 89.7309,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 280
+    },
+    {
+      "entropy": 0.7720955617725849,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.1931514739990234,
+      "learning_rate": 0.0002627757947934439,
+      "loss": 0.6965402603149414,
+      "mean_token_accuracy": 0.8040348663926125,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7214708304807042,
+      "eval_loss": 0.7350823879241943,
+      "eval_mean_token_accuracy": 0.7942911925610532,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 89.6451,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 300
+    },
+    {
+      "entropy": 0.7930095426738262,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8994467258453369,
+      "learning_rate": 0.0002803527710338081,
+      "loss": 0.7331492900848389,
+      "mean_token_accuracy": 0.796185302734375,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.8142535488926963,
+      "eval_loss": 0.7237834334373474,
+      "eval_mean_token_accuracy": 0.799495273761535,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 90.6759,
+      "eval_samples_per_second": 15.66,
+      "eval_steps_per_second": 1.963,
+      "step": 320
+    },
+    {
+      "entropy": 0.7825102441012859,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 1.8224239349365234,
+      "learning_rate": 0.00029792974727417223,
+      "loss": 0.7331175804138184,
+      "mean_token_accuracy": 0.7977667152881622,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7327135443017724,
+      "eval_loss": 0.7388784289360046,
+      "eval_mean_token_accuracy": 0.8007041252730938,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 89.7692,
+      "eval_samples_per_second": 15.818,
+      "eval_steps_per_second": 1.983,
+      "step": 340
+    },
+    {
+      "entropy": 0.819331557303667,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 1.0407779216766357,
+      "learning_rate": 0.0003155067235145364,
+      "loss": 0.7456695556640625,
+      "mean_token_accuracy": 0.7936980701982975,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.8065585420372781,
+      "eval_loss": 0.735519528388977,
+      "eval_mean_token_accuracy": 0.7974206296245704,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 89.4336,
+      "eval_samples_per_second": 15.878,
+      "eval_steps_per_second": 1.99,
+      "step": 360
+    },
+    {
+      "entropy": 0.8114653021097183,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.1209490299224854,
+      "learning_rate": 0.0003330836997549005,
+      "loss": 0.7340899467468261,
+      "mean_token_accuracy": 0.7914272703230381,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.8025527285056168,
+      "eval_loss": 0.7295576930046082,
+      "eval_mean_token_accuracy": 0.7990576628218876,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 89.9579,
+      "eval_samples_per_second": 15.785,
+      "eval_steps_per_second": 1.979,
+      "step": 380
+    },
+    {
+      "entropy": 0.8042497783899307,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 1.2437798976898193,
+      "learning_rate": 0.0003506606759952647,
+      "loss": 0.7381744384765625,
+      "mean_token_accuracy": 0.7967362694442273,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7717011336530193,
+      "eval_loss": 0.7436173558235168,
+      "eval_mean_token_accuracy": 0.7975849482450592,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 89.8411,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 400
+    },
+    {
+      "entropy": 0.7987750010056929,
+      "epoch": 1.012070006035003,
+      "grad_norm": 1.2390918731689453,
+      "learning_rate": 0.000364721224843344,
+      "loss": 0.7317790031433106,
+      "mean_token_accuracy": 0.7986087551364651,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7333170604170038,
+      "eval_loss": 0.7456948757171631,
+      "eval_mean_token_accuracy": 0.7964897577682238,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 89.7288,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 420
+    },
+    {
+      "entropy": 0.7372976846992969,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.1552740335464478,
+      "learning_rate": 0.00036468510102269575,
+      "loss": 0.6734249114990234,
+      "mean_token_accuracy": 0.805834824591875,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.7130355526891987,
+      "eval_loss": 0.7555333375930786,
+      "eval_mean_token_accuracy": 0.7961652781186479,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 89.8713,
+      "eval_samples_per_second": 15.8,
+      "eval_steps_per_second": 1.981,
+      "step": 440
+    },
+    {
+      "entropy": 0.7465668715536594,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 1.1991469860076904,
+      "learning_rate": 0.0003645973816745066,
+      "loss": 0.6871879577636719,
+      "mean_token_accuracy": 0.8042450234293937,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7772979542110743,
+      "eval_loss": 0.7561826705932617,
+      "eval_mean_token_accuracy": 0.7953876427720102,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 89.8395,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 460
+    },
+    {
+      "entropy": 0.7477855801582336,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 1.325486660003662,
+      "learning_rate": 0.00036445809162231435,
+      "loss": 0.6836830139160156,
+      "mean_token_accuracy": 0.8021619468927383,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.7098110676481483,
+      "eval_loss": 0.758007287979126,
+      "eval_mean_token_accuracy": 0.7909953189030122,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 89.6101,
+      "eval_samples_per_second": 15.846,
+      "eval_steps_per_second": 1.986,
+      "step": 480
+    },
+    {
+      "entropy": 0.7632691070437432,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 1.0122262239456177,
+      "learning_rate": 0.0003642672702835562,
+      "loss": 0.7071213245391845,
+      "mean_token_accuracy": 0.8000680930912495,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7645382720432924,
+      "eval_loss": 0.7492606043815613,
+      "eval_mean_token_accuracy": 0.7976858676149604,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 89.6092,
+      "eval_samples_per_second": 15.847,
+      "eval_steps_per_second": 1.986,
+      "step": 500
+    },
+    {
+      "entropy": 0.7623726457357407,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.367774486541748,
+      "learning_rate": 0.00036402497165841384,
+      "loss": 0.6965175151824952,
+      "mean_token_accuracy": 0.8016501650214195,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7868083715438843,
+      "eval_loss": 0.7481449842453003,
+      "eval_mean_token_accuracy": 0.799330582779445,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 89.8241,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 520
+    },
+    {
+      "entropy": 0.7535207405686378,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 1.213614821434021,
+      "learning_rate": 0.00036373126431453207,
+      "loss": 0.7078310489654541,
+      "mean_token_accuracy": 0.7998907208442688,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7829091950748743,
+      "eval_loss": 0.7416072487831116,
+      "eval_mean_token_accuracy": 0.7969146250339036,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 89.4607,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 540
+    },
+    {
+      "entropy": 0.7527363449335098,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 1.0929982662200928,
+      "learning_rate": 0.00036338623136761495,
+      "loss": 0.7058369636535644,
+      "mean_token_accuracy": 0.7997087433934211,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.728419938114252,
+      "eval_loss": 0.744873046875,
+      "eval_mean_token_accuracy": 0.7966634296299366,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 89.6597,
+      "eval_samples_per_second": 15.838,
+      "eval_steps_per_second": 1.985,
+      "step": 560
+    },
+    {
+      "entropy": 0.7406099259853363,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.409364104270935,
+      "learning_rate": 0.00036298997045790515,
+      "loss": 0.6902852058410645,
+      "mean_token_accuracy": 0.805386807769537,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.6991632758231645,
+      "eval_loss": 0.7480319738388062,
+      "eval_mean_token_accuracy": 0.7970068006033308,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 89.8329,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 580
+    },
+    {
+      "entropy": 0.7459449704736472,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.7053111791610718,
+      "learning_rate": 0.00036254259372255277,
+      "loss": 0.7044640064239502,
+      "mean_token_accuracy": 0.8005018755793571,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7454566660891758,
+      "eval_loss": 0.7551484704017639,
+      "eval_mean_token_accuracy": 0.7897093731365846,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 89.8313,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 600
+    },
+    {
+      "entropy": 0.7621391348540782,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 1.174443244934082,
+      "learning_rate": 0.000362044227763882,
+      "loss": 0.7111573696136475,
+      "mean_token_accuracy": 0.798752411454916,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7664387949397055,
+      "eval_loss": 0.7365804314613342,
+      "eval_mean_token_accuracy": 0.7963101200843126,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 89.9623,
+      "eval_samples_per_second": 15.784,
+      "eval_steps_per_second": 1.979,
+      "step": 620
+    },
+    {
+      "entropy": 0.7424944408237935,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.2898184061050415,
+      "learning_rate": 0.000361495013613564,
+      "loss": 0.6913110733032226,
+      "mean_token_accuracy": 0.8009015507996082,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7255479529332579,
+      "eval_loss": 0.7345991730690002,
+      "eval_mean_token_accuracy": 0.8013592945056015,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 89.817,
+      "eval_samples_per_second": 15.81,
+      "eval_steps_per_second": 1.982,
+      "step": 640
+    },
+    {
+      "entropy": 0.7303367637097835,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.1310476064682007,
+      "learning_rate": 0.00036089510669270666,
+      "loss": 0.6875617027282714,
+      "mean_token_accuracy": 0.8035397171974182,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7101308248016271,
+      "eval_loss": 0.7394917011260986,
+      "eval_mean_token_accuracy": 0.7990792942850777,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 89.7431,
+      "eval_samples_per_second": 15.823,
+      "eval_steps_per_second": 1.983,
+      "step": 660
+    },
+    {
+      "entropy": 0.7387678384780884,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.9645389914512634,
+      "learning_rate": 0.0003602446767678725,
+      "loss": 0.6911204338073731,
+      "mean_token_accuracy": 0.8056333385407924,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7281073737010527,
+      "eval_loss": 0.7376708984375,
+      "eval_mean_token_accuracy": 0.7997356415464637,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 90.0338,
+      "eval_samples_per_second": 15.772,
+      "eval_steps_per_second": 1.977,
+      "step": 680
+    },
+    {
+      "entropy": 0.7449554048478604,
+      "epoch": 1.687990343995172,
+      "grad_norm": 1.9069629907608032,
+      "learning_rate": 0.0003595439079030364,
+      "loss": 0.7042286396026611,
+      "mean_token_accuracy": 0.8018639378249646,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7022633817088738,
+      "eval_loss": 0.7320939898490906,
+      "eval_mean_token_accuracy": 0.8006051541044471,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 89.8948,
+      "eval_samples_per_second": 15.796,
+      "eval_steps_per_second": 1.98,
+      "step": 700
+    },
+    {
+      "entropy": 0.723249051719904,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.8333877325057983,
+      "learning_rate": 0.00035879299840749777,
+      "loss": 0.7043714046478271,
+      "mean_token_accuracy": 0.8028345607221127,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.672436946898364,
+      "eval_loss": 0.713233470916748,
+      "eval_mean_token_accuracy": 0.8047760577684038,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 89.6394,
+      "eval_samples_per_second": 15.841,
+      "eval_steps_per_second": 1.986,
+      "step": 720
+    },
+    {
+      "entropy": 0.730014206469059,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.4580037593841553,
+      "learning_rate": 0.00035799216077976137,
+      "loss": 0.6985176563262939,
+      "mean_token_accuracy": 0.8015547141432762,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.6908873795123582,
+      "eval_loss": 0.7198202013969421,
+      "eval_mean_token_accuracy": 0.8024131682481659,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 89.3771,
+      "eval_samples_per_second": 15.888,
+      "eval_steps_per_second": 1.992,
+      "step": 740
+    },
+    {
+      "entropy": 0.7219605796039105,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.51682710647583,
+      "learning_rate": 0.000357141621647403,
+      "loss": 0.6844424724578857,
+      "mean_token_accuracy": 0.8032813094556331,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7113401109582922,
+      "eval_loss": 0.7195309400558472,
+      "eval_mean_token_accuracy": 0.8019187035185568,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 89.4658,
+      "eval_samples_per_second": 15.872,
+      "eval_steps_per_second": 1.99,
+      "step": 760
+    },
+    {
+      "entropy": 0.729826743900776,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 1.0838000774383545,
+      "learning_rate": 0.0003562416217029361,
+      "loss": 0.685858678817749,
+      "mean_token_accuracy": 0.8050567395985126,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.649370267484965,
+      "eval_loss": 0.7177029252052307,
+      "eval_mean_token_accuracy": 0.8050174043419656,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 89.2287,
+      "eval_samples_per_second": 15.914,
+      "eval_steps_per_second": 1.995,
+      "step": 780
+    },
+    {
+      "entropy": 0.6799364153295755,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 1.1364924907684326,
+      "learning_rate": 0.00035529241563569903,
+      "loss": 0.6714428901672364,
+      "mean_token_accuracy": 0.8087341658771038,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6603799047095052,
+      "eval_loss": 0.7201809287071228,
+      "eval_mean_token_accuracy": 0.802863868769635,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 89.2718,
+      "eval_samples_per_second": 15.906,
+      "eval_steps_per_second": 1.994,
+      "step": 800
+    },
+    {
+      "entropy": 0.7078610375523567,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 1.2191942930221558,
+      "learning_rate": 0.0003542942720597808,
+      "loss": 0.6996590614318847,
+      "mean_token_accuracy": 0.8049512051045895,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.7240545264120852,
+      "eval_loss": 0.7301309108734131,
+      "eval_mean_token_accuracy": 0.8015973166133581,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 89.3438,
+      "eval_samples_per_second": 15.894,
+      "eval_steps_per_second": 1.992,
+      "step": 820
+    },
+    {
+      "entropy": 0.6557391556826505,
+      "epoch": 2.024140012070006,
+      "grad_norm": 1.36203134059906,
+      "learning_rate": 0.0003532474734380064,
+      "loss": 0.6327113628387451,
+      "mean_token_accuracy": 0.8142829847026181,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6573808648613062,
+      "eval_loss": 0.7334365844726562,
+      "eval_mean_token_accuracy": 0.8043633893634496,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 89.3359,
+      "eval_samples_per_second": 15.895,
+      "eval_steps_per_second": 1.992,
+      "step": 840
+    },
+    {
+      "entropy": 0.6409128502011299,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 1.5804955959320068,
+      "learning_rate": 0.0003521523160020035,
+      "loss": 0.6023559093475341,
+      "mean_token_accuracy": 0.8222251623868942,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6983234788594621,
+      "eval_loss": 0.7232135534286499,
+      "eval_mean_token_accuracy": 0.8010654268639811,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 89.2615,
+      "eval_samples_per_second": 15.908,
+      "eval_steps_per_second": 1.994,
+      "step": 860
+    },
+    {
+      "entropy": 0.6151149723678827,
+      "epoch": 2.12070006035003,
+      "grad_norm": 1.1975436210632324,
+      "learning_rate": 0.00035100910966837193,
+      "loss": 0.5803564548492431,
+      "mean_token_accuracy": 0.8256325736641884,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6636555573243773,
+      "eval_loss": 0.7205662727355957,
+      "eval_mean_token_accuracy": 0.8036768369460374,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 89.2957,
+      "eval_samples_per_second": 15.902,
+      "eval_steps_per_second": 1.993,
+      "step": 880
+    },
+    {
+      "entropy": 0.6201727617532015,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9907131195068359,
+      "learning_rate": 0.0003498181779509813,
+      "loss": 0.6101872444152832,
+      "mean_token_accuracy": 0.8196512959897518,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.6473788086617931,
+      "eval_loss": 0.7062397003173828,
+      "eval_mean_token_accuracy": 0.8078362184963869,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 88.6198,
+      "eval_samples_per_second": 16.024,
+      "eval_steps_per_second": 2.009,
+      "step": 900
+    },
+    {
+      "entropy": 0.6097332935780286,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.9055633544921875,
+      "learning_rate": 0.00034857985786942026,
+      "loss": 0.5956094264984131,
+      "mean_token_accuracy": 0.8240575887262821,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.612629868341296,
+      "eval_loss": 0.7106770277023315,
+      "eval_mean_token_accuracy": 0.805324277181304,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 89.4383,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 920
+    },
+    {
+      "entropy": 0.6182701248675585,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.9513736367225647,
+      "learning_rate": 0.00034729449985362404,
+      "loss": 0.6019269466400147,
+      "mean_token_accuracy": 0.8187106013298034,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6437647456533453,
+      "eval_loss": 0.7173502445220947,
+      "eval_mean_token_accuracy": 0.8056439573175451,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.9568,
+      "eval_samples_per_second": 15.963,
+      "eval_steps_per_second": 2.001,
+      "step": 940
+    },
+    {
+      "entropy": 0.6508570514619351,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.9606207013130188,
+      "learning_rate": 0.0003459624676447067,
+      "loss": 0.6296409606933594,
+      "mean_token_accuracy": 0.8167732007801533,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6288247135248077,
+      "eval_loss": 0.7160675525665283,
+      "eval_mean_token_accuracy": 0.8066010689467527,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 89.0044,
+      "eval_samples_per_second": 15.954,
+      "eval_steps_per_second": 2.0,
+      "step": 960
+    },
+    {
+      "entropy": 0.6380081418901682,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.9677181839942932,
+      "learning_rate": 0.000344584138192027,
+      "loss": 0.6210546493530273,
+      "mean_token_accuracy": 0.8186135642230511,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6407805031604981,
+      "eval_loss": 0.7055043578147888,
+      "eval_mean_token_accuracy": 0.8070103157772107,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 89.5606,
+      "eval_samples_per_second": 15.855,
+      "eval_steps_per_second": 1.987,
+      "step": 980
+    },
+    {
+      "entropy": 0.6310251638293266,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 2.1046407222747803,
+      "learning_rate": 0.000343159901546516,
+      "loss": 0.6165830135345459,
+      "mean_token_accuracy": 0.8207855716347694,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6461507470420237,
+      "eval_loss": 0.7292136549949646,
+      "eval_mean_token_accuracy": 0.8030418261383356,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 89.315,
+      "eval_samples_per_second": 15.899,
+      "eval_steps_per_second": 1.993,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6256617344915867,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 1.1768349409103394,
+      "learning_rate": 0.0003416901607502972,
+      "loss": 0.6142709255218506,
+      "mean_token_accuracy": 0.8220669947564602,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6150523993406403,
+      "eval_loss": 0.7019563317298889,
+      "eval_mean_token_accuracy": 0.8094323099998946,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 89.419,
+      "eval_samples_per_second": 15.88,
+      "eval_steps_per_second": 1.991,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6126735735684633,
+      "epoch": 2.506940253470127,
+      "grad_norm": 1.8065496683120728,
+      "learning_rate": 0.00034017533172263055,
+      "loss": 0.6056458473205566,
+      "mean_token_accuracy": 0.8203762136399746,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6544345456562685,
+      "eval_loss": 0.7086517214775085,
+      "eval_mean_token_accuracy": 0.806709805901131,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 88.981,
+      "eval_samples_per_second": 15.958,
+      "eval_steps_per_second": 2.0,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6311754353344441,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 1.0576369762420654,
+      "learning_rate": 0.00033861584314221236,
+      "loss": 0.6176392555236816,
+      "mean_token_accuracy": 0.8177410490810871,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6273799200406235,
+      "eval_loss": 0.708163321018219,
+      "eval_mean_token_accuracy": 0.8032017099053672,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 89.7668,
+      "eval_samples_per_second": 15.819,
+      "eval_steps_per_second": 1.983,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6200249589979648,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.997065007686615,
+      "learning_rate": 0.0003370121363258637,
+      "loss": 0.6203888893127442,
+      "mean_token_accuracy": 0.8196275025606156,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6457576811983344,
+      "eval_loss": 0.6972290277481079,
+      "eval_mean_token_accuracy": 0.8095980849158898,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 89.2027,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 1080
+    },
+    {
+      "entropy": 0.626810473203659,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.865475058555603,
+      "learning_rate": 0.0003353646651036438,
+      "loss": 0.6102585792541504,
+      "mean_token_accuracy": 0.819708751142025,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.6412022431914726,
+      "eval_loss": 0.7036139965057373,
+      "eval_mean_token_accuracy": 0.8070317135098275,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 89.4733,
+      "eval_samples_per_second": 15.871,
+      "eval_steps_per_second": 1.989,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6581619590520859,
+      "epoch": 2.700060350030175,
+      "grad_norm": 1.00307035446167,
+      "learning_rate": 0.00033367389569042064,
+      "loss": 0.6397720336914062,
+      "mean_token_accuracy": 0.81413309648633,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6288736402318719,
+      "eval_loss": 0.6966825723648071,
+      "eval_mean_token_accuracy": 0.8094039669867312,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 88.7341,
+      "eval_samples_per_second": 16.003,
+      "eval_steps_per_second": 2.006,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6399412982165813,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.8589329719543457,
+      "learning_rate": 0.0003319403065539384,
+      "loss": 0.6347338199615479,
+      "mean_token_accuracy": 0.8139534957706929,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.683491913455256,
+      "eval_loss": 0.6984149217605591,
+      "eval_mean_token_accuracy": 0.8083989469522841,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 89.3929,
+      "eval_samples_per_second": 15.885,
+      "eval_steps_per_second": 1.991,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6264835461974144,
+      "epoch": 2.796620398310199,
+      "grad_norm": 0.7807853817939758,
+      "learning_rate": 0.0003301643882794163,
+      "loss": 0.6162629127502441,
+      "mean_token_accuracy": 0.8216292977333068,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6578792159476977,
+      "eval_loss": 0.6921441555023193,
+      "eval_mean_token_accuracy": 0.8094299177775223,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 89.2356,
+      "eval_samples_per_second": 15.913,
+      "eval_steps_per_second": 1.995,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6269129011780024,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 1.1675593852996826,
+      "learning_rate": 0.0003283466434307189,
+      "loss": 0.6235532760620117,
+      "mean_token_accuracy": 0.8188897788524627,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6533248447970058,
+      "eval_loss": 0.696849524974823,
+      "eval_mean_token_accuracy": 0.8102703713968898,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2602,
+      "eval_samples_per_second": 16.089,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.6259873781353236,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.9596486687660217,
+      "learning_rate": 0.00032648758640813655,
+      "loss": 0.6118728160858155,
+      "mean_token_accuracy": 0.8171859130263328,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6212078754821521,
+      "eval_loss": 0.6942870020866394,
+      "eval_mean_token_accuracy": 0.810484343365337,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 89.4375,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 1200
+    },
+    {
+      "entropy": 0.6142537571489811,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 0.9625058770179749,
+      "learning_rate": 0.00032458774330281615,
+      "loss": 0.6060261249542236,
+      "mean_token_accuracy": 0.8223113484680653,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6553665667437436,
+      "eval_loss": 0.6835269331932068,
+      "eval_mean_token_accuracy": 0.8134723968720168,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 89.5092,
+      "eval_samples_per_second": 15.864,
+      "eval_steps_per_second": 1.989,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6301972426474094,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.9657333493232727,
+      "learning_rate": 0.0003226476517478835,
+      "loss": 0.6277759075164795,
+      "mean_token_accuracy": 0.8164266042411328,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6594253118788258,
+      "eval_loss": 0.6820935606956482,
+      "eval_mean_token_accuracy": 0.8132368198941263,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 89.6211,
+      "eval_samples_per_second": 15.844,
+      "eval_steps_per_second": 1.986,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5346100655469027,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8541185259819031,
+      "learning_rate": 0.0003206678607662996,
+      "loss": 0.5261692047119141,
+      "mean_token_accuracy": 0.8382289425119177,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5626645662476507,
+      "eval_loss": 0.7028542160987854,
+      "eval_mean_token_accuracy": 0.8133395572056931,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 89.1267,
+      "eval_samples_per_second": 15.932,
+      "eval_steps_per_second": 1.997,
+      "step": 1260
+    },
+    {
+      "entropy": 0.5001101832836866,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.8208815455436707,
+      "learning_rate": 0.0003186489306154935,
+      "loss": 0.4850132942199707,
+      "mean_token_accuracy": 0.8468983843922615,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5889462046743779,
+      "eval_loss": 0.691015899181366,
+      "eval_mean_token_accuracy": 0.8135226466012805,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.9677,
+      "eval_samples_per_second": 15.961,
+      "eval_steps_per_second": 2.001,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5242766339331866,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 0.7893072962760925,
+      "learning_rate": 0.0003165914326288163,
+      "loss": 0.5061577320098877,
+      "mean_token_accuracy": 0.8425608821213245,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5543691962957382,
+      "eval_loss": 0.6913734078407288,
+      "eval_mean_token_accuracy": 0.8154889680026622,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 89.6963,
+      "eval_samples_per_second": 15.831,
+      "eval_steps_per_second": 1.984,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5327721010893584,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.8992080092430115,
+      "learning_rate": 0.0003144959490538604,
+      "loss": 0.5068036556243897,
+      "mean_token_accuracy": 0.8420542575418949,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5734540803378887,
+      "eval_loss": 0.6948716044425964,
+      "eval_mean_token_accuracy": 0.8137217944257715,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 89.4597,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5247377116233111,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.9291845560073853,
+      "learning_rate": 0.0003123630728876902,
+      "loss": 0.5023272037506104,
+      "mean_token_accuracy": 0.8435182586312294,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5492362935891312,
+      "eval_loss": 0.6980633735656738,
+      "eval_mean_token_accuracy": 0.8155082919624415,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 89.2124,
+      "eval_samples_per_second": 15.917,
+      "eval_steps_per_second": 1.995,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5380499072372913,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 1.0477524995803833,
+      "learning_rate": 0.00031019340770903136,
+      "loss": 0.5264591693878173,
+      "mean_token_accuracy": 0.8391959741711617,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5896503307511297,
+      "eval_loss": 0.6997144222259521,
+      "eval_mean_token_accuracy": 0.8124373062942805,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 89.1147,
+      "eval_samples_per_second": 15.935,
+      "eval_steps_per_second": 1.997,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5460653610527515,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 1.2473069429397583,
+      "learning_rate": 0.00030798756750746476,
+      "loss": 0.5319677829742432,
+      "mean_token_accuracy": 0.8364320226013661,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5659052182114526,
+      "eval_loss": 0.6924049258232117,
+      "eval_mean_token_accuracy": 0.8153053585732921,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 89.6468,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5367407951503992,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 1.2679712772369385,
+      "learning_rate": 0.00030574617650967485,
+      "loss": 0.5190927028656006,
+      "mean_token_accuracy": 0.8399469174444676,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5581759144081159,
+      "eval_loss": 0.6933317184448242,
+      "eval_mean_token_accuracy": 0.8139757862251796,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 89.0749,
+      "eval_samples_per_second": 15.942,
+      "eval_steps_per_second": 1.998,
+      "step": 1400
+    },
+    {
+      "entropy": 0.530736118927598,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.9717612862586975,
+      "learning_rate": 0.0003034698690028009,
+      "loss": 0.512363052368164,
+      "mean_token_accuracy": 0.8419127956032753,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5233335836549823,
+      "eval_loss": 0.6958550214767456,
+      "eval_mean_token_accuracy": 0.8158778501360604,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 88.8664,
+      "eval_samples_per_second": 15.979,
+      "eval_steps_per_second": 2.003,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5327631575986743,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7649208903312683,
+      "learning_rate": 0.00030115928915494116,
+      "loss": 0.5202179431915284,
+      "mean_token_accuracy": 0.8384683802723885,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.5960972657364406,
+      "eval_loss": 0.6984783411026001,
+      "eval_mean_token_accuracy": 0.8095930597085631,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 89.5168,
+      "eval_samples_per_second": 15.863,
+      "eval_steps_per_second": 1.988,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5428155034780502,
+      "epoch": 3.51901025950513,
+      "grad_norm": 1.2790874242782593,
+      "learning_rate": 0.00029881509083286116,
+      "loss": 0.524717378616333,
+      "mean_token_accuracy": 0.8346644505858422,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.550318562917495,
+      "eval_loss": 0.6988471746444702,
+      "eval_mean_token_accuracy": 0.8132969897784544,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 89.2598,
+      "eval_samples_per_second": 15.909,
+      "eval_steps_per_second": 1.994,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5489993002265692,
+      "epoch": 3.567290283645142,
+      "grad_norm": 1.055821418762207,
+      "learning_rate": 0.00029643793741695687,
+      "loss": 0.5421150207519532,
+      "mean_token_accuracy": 0.8345710933208466,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5816821718818685,
+      "eval_loss": 0.6853668689727783,
+      "eval_mean_token_accuracy": 0.8142332360985574,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 89.0424,
+      "eval_samples_per_second": 15.947,
+      "eval_steps_per_second": 1.999,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5451365381479263,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.9027532935142517,
+      "learning_rate": 0.00029402850161352583,
+      "loss": 0.5463609218597412,
+      "mean_token_accuracy": 0.8369908526539802,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5344233209832331,
+      "eval_loss": 0.6794531345367432,
+      "eval_mean_token_accuracy": 0.8177759824843889,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 89.123,
+      "eval_samples_per_second": 15.933,
+      "eval_steps_per_second": 1.997,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5521467242389917,
+      "epoch": 3.663850331925166,
+      "grad_norm": 1.4606311321258545,
+      "learning_rate": 0.0002915874652643997,
+      "loss": 0.5421902179718018,
+      "mean_token_accuracy": 0.8339455373585224,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5807709094513668,
+      "eval_loss": 0.6949691772460938,
+      "eval_mean_token_accuracy": 0.8121141439743256,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.5437,
+      "eval_samples_per_second": 16.037,
+      "eval_steps_per_second": 2.01,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5455610822886229,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7269843220710754,
+      "learning_rate": 0.0002891155191539905,
+      "loss": 0.5407682895660401,
+      "mean_token_accuracy": 0.8393648222088814,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.6032205542151847,
+      "eval_loss": 0.6790974140167236,
+      "eval_mean_token_accuracy": 0.8165463112043531,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 89.517,
+      "eval_samples_per_second": 15.863,
+      "eval_steps_per_second": 1.988,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5474163968116045,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.8148866295814514,
+      "learning_rate": 0.00028661336281380717,
+      "loss": 0.5394441604614257,
+      "mean_token_accuracy": 0.8336028568446636,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5797032238392348,
+      "eval_loss": 0.670870304107666,
+      "eval_mean_token_accuracy": 0.8182691593518417,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 89.4468,
+      "eval_samples_per_second": 15.875,
+      "eval_steps_per_second": 1.99,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5511750865727663,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.3441656827926636,
+      "learning_rate": 0.0002840817043244964,
+      "loss": 0.5378933906555176,
+      "mean_token_accuracy": 0.8349151819944381,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5297858750217417,
+      "eval_loss": 0.680833637714386,
+      "eval_mean_token_accuracy": 0.8180924275617921,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 89.5417,
+      "eval_samples_per_second": 15.859,
+      "eval_steps_per_second": 1.988,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5379613988101483,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9813932776451111,
+      "learning_rate": 0.00028152126011546396,
+      "loss": 0.5340342044830322,
+      "mean_token_accuracy": 0.8346010789275169,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5811898510777549,
+      "eval_loss": 0.6764267683029175,
+      "eval_mean_token_accuracy": 0.8180528236239144,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 88.7831,
+      "eval_samples_per_second": 15.994,
+      "eval_steps_per_second": 2.005,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5607876226305961,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.9617928266525269,
+      "learning_rate": 0.00027893275476213383,
+      "loss": 0.5434338569641113,
+      "mean_token_accuracy": 0.8332898400723934,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5474644318390428,
+      "eval_loss": 0.6740226745605469,
+      "eval_mean_token_accuracy": 0.8185851306058047,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 89.3246,
+      "eval_samples_per_second": 15.897,
+      "eval_steps_per_second": 1.993,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5354843523353339,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.5320990085601807,
+      "learning_rate": 0.0002763169207809021,
+      "loss": 0.5251852989196777,
+      "mean_token_accuracy": 0.8416872084140777,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.5436856410141742,
+      "eval_loss": 0.6675190925598145,
+      "eval_mean_token_accuracy": 0.8202396507343549,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 89.6328,
+      "eval_samples_per_second": 15.842,
+      "eval_steps_per_second": 1.986,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5418571938167919,
+      "epoch": 4.0,
+      "grad_norm": 4.690354347229004,
+      "learning_rate": 0.000273674498421843,
+      "loss": 0.548275089263916,
+      "mean_token_accuracy": 0.836557479647847,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5497228983747825,
+      "eval_loss": 0.6621812582015991,
+      "eval_mean_token_accuracy": 0.8205679802412398,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 89.6272,
+      "eval_samples_per_second": 15.843,
+      "eval_steps_per_second": 1.986,
+      "step": 1660
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.6623734491102618e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1680/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1680/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1680/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1680/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1680/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..7ce4b072fcbe5b636867e13937c56fedee3b68b4
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1680/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.02728394274790723,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1680/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1680/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1680/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1680/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1680/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1680/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1680/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1680/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..2aadd1d233280d318896048ad0322373d755ff98
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1680/trainer_state.json
@@ -0,0 +1,1798 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 4.048280024140012,
+  "eval_steps": 20,
+  "global_step": 1680,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.024795526266098,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.945375919342041,
+      "learning_rate": 1.6698127428345936e-05,
+      "loss": 1.7825420379638672,
+      "mean_token_accuracy": 0.6315708436071873,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.29921497588747,
+      "eval_loss": 1.169131875038147,
+      "eval_mean_token_accuracy": 0.7185706342204233,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 91.7898,
+      "eval_samples_per_second": 15.47,
+      "eval_steps_per_second": 1.939,
+      "step": 20
+    },
+    {
+      "entropy": 1.0346344470977784,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.7916979789733887,
+      "learning_rate": 3.427510366871008e-05,
+      "loss": 0.9480395317077637,
+      "mean_token_accuracy": 0.7543429024517536,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 0.9673695430326997,
+      "eval_loss": 0.8588430285453796,
+      "eval_mean_token_accuracy": 0.7737233002534073,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 89.8219,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 40
+    },
+    {
+      "entropy": 0.8977566614747048,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7972187995910645,
+      "learning_rate": 5.185207990907422e-05,
+      "loss": 0.8029604911804199,
+      "mean_token_accuracy": 0.7818016350269318,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8436836982041263,
+      "eval_loss": 0.7893033623695374,
+      "eval_mean_token_accuracy": 0.7858914984076211,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 89.9349,
+      "eval_samples_per_second": 15.789,
+      "eval_steps_per_second": 1.979,
+      "step": 60
+    },
+    {
+      "entropy": 0.8650321319699288,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.2128729820251465,
+      "learning_rate": 6.942905614943836e-05,
+      "loss": 0.77417893409729,
+      "mean_token_accuracy": 0.7875479467213153,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8209857180547179,
+      "eval_loss": 0.7572280168533325,
+      "eval_mean_token_accuracy": 0.7925970383574453,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 89.9459,
+      "eval_samples_per_second": 15.787,
+      "eval_steps_per_second": 1.979,
+      "step": 80
+    },
+    {
+      "entropy": 0.8096660938113928,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.1254287958145142,
+      "learning_rate": 8.700603238980251e-05,
+      "loss": 0.7170331001281738,
+      "mean_token_accuracy": 0.7994555257260799,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.792780208788561,
+      "eval_loss": 0.7442984580993652,
+      "eval_mean_token_accuracy": 0.7954704573984896,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 90.294,
+      "eval_samples_per_second": 15.726,
+      "eval_steps_per_second": 1.971,
+      "step": 100
+    },
+    {
+      "entropy": 0.7979829199612141,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2140507698059082,
+      "learning_rate": 0.00010458300863016666,
+      "loss": 0.7141861915588379,
+      "mean_token_accuracy": 0.8000014744699001,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.7938692486018277,
+      "eval_loss": 0.7279797196388245,
+      "eval_mean_token_accuracy": 0.7989798734027348,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 89.7189,
+      "eval_samples_per_second": 15.827,
+      "eval_steps_per_second": 1.984,
+      "step": 120
+    },
+    {
+      "entropy": 0.7916347607970238,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.0506458282470703,
+      "learning_rate": 0.0001221599848705308,
+      "loss": 0.7117055416107178,
+      "mean_token_accuracy": 0.7970046654343605,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.8052781063519167,
+      "eval_loss": 0.722576916217804,
+      "eval_mean_token_accuracy": 0.7982530523551984,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 90.0376,
+      "eval_samples_per_second": 15.771,
+      "eval_steps_per_second": 1.977,
+      "step": 140
+    },
+    {
+      "entropy": 0.7880666613578796,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8409207463264465,
+      "learning_rate": 0.00013973696111089492,
+      "loss": 0.7052523136138916,
+      "mean_token_accuracy": 0.8000123649835587,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7787443134891853,
+      "eval_loss": 0.7137772440910339,
+      "eval_mean_token_accuracy": 0.8014209287220173,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 90.2562,
+      "eval_samples_per_second": 15.733,
+      "eval_steps_per_second": 1.972,
+      "step": 160
+    },
+    {
+      "entropy": 0.7696007996797561,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.0182477235794067,
+      "learning_rate": 0.00015731393735125907,
+      "loss": 0.7008797645568847,
+      "mean_token_accuracy": 0.8007259473204613,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.752259682068664,
+      "eval_loss": 0.7171286344528198,
+      "eval_mean_token_accuracy": 0.80052122574174,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 90.0797,
+      "eval_samples_per_second": 15.764,
+      "eval_steps_per_second": 1.976,
+      "step": 180
+    },
+    {
+      "entropy": 0.7828126326203346,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9553632736206055,
+      "learning_rate": 0.0001748909135916232,
+      "loss": 0.7046853542327881,
+      "mean_token_accuracy": 0.7988562889397144,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7389638169427936,
+      "eval_loss": 0.7187935709953308,
+      "eval_mean_token_accuracy": 0.7995899999409579,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 89.9223,
+      "eval_samples_per_second": 15.791,
+      "eval_steps_per_second": 1.979,
+      "step": 200
+    },
+    {
+      "entropy": 0.7909286297857762,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 1.134459376335144,
+      "learning_rate": 0.00019246788983198735,
+      "loss": 0.6989312171936035,
+      "mean_token_accuracy": 0.7991565138101577,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7496965199374082,
+      "eval_loss": 0.7106321454048157,
+      "eval_mean_token_accuracy": 0.8046183442131857,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 89.6899,
+      "eval_samples_per_second": 15.832,
+      "eval_steps_per_second": 1.985,
+      "step": 220
+    },
+    {
+      "entropy": 0.7651963323354721,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.9557585716247559,
+      "learning_rate": 0.0002100448660723515,
+      "loss": 0.6984559059143066,
+      "mean_token_accuracy": 0.8019823372364044,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7615563079212488,
+      "eval_loss": 0.7098539471626282,
+      "eval_mean_token_accuracy": 0.8047233508544022,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 90.3752,
+      "eval_samples_per_second": 15.712,
+      "eval_steps_per_second": 1.97,
+      "step": 240
+    },
+    {
+      "entropy": 0.787507726252079,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.9279311299324036,
+      "learning_rate": 0.00022762184231271568,
+      "loss": 0.708138370513916,
+      "mean_token_accuracy": 0.8013308703899383,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7856640896100676,
+      "eval_loss": 0.7191570401191711,
+      "eval_mean_token_accuracy": 0.8024677249153008,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 93.7336,
+      "eval_samples_per_second": 15.149,
+      "eval_steps_per_second": 1.899,
+      "step": 260
+    },
+    {
+      "entropy": 0.768489234894514,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.7992776036262512,
+      "learning_rate": 0.0002451988185530798,
+      "loss": 0.7035849571228028,
+      "mean_token_accuracy": 0.8031105332076549,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7682498740346244,
+      "eval_loss": 0.7149022221565247,
+      "eval_mean_token_accuracy": 0.7973497995499814,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 89.7309,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 280
+    },
+    {
+      "entropy": 0.7720955617725849,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.1931514739990234,
+      "learning_rate": 0.0002627757947934439,
+      "loss": 0.6965402603149414,
+      "mean_token_accuracy": 0.8040348663926125,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7214708304807042,
+      "eval_loss": 0.7350823879241943,
+      "eval_mean_token_accuracy": 0.7942911925610532,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 89.6451,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 300
+    },
+    {
+      "entropy": 0.7930095426738262,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8994467258453369,
+      "learning_rate": 0.0002803527710338081,
+      "loss": 0.7331492900848389,
+      "mean_token_accuracy": 0.796185302734375,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.8142535488926963,
+      "eval_loss": 0.7237834334373474,
+      "eval_mean_token_accuracy": 0.799495273761535,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 90.6759,
+      "eval_samples_per_second": 15.66,
+      "eval_steps_per_second": 1.963,
+      "step": 320
+    },
+    {
+      "entropy": 0.7825102441012859,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 1.8224239349365234,
+      "learning_rate": 0.00029792974727417223,
+      "loss": 0.7331175804138184,
+      "mean_token_accuracy": 0.7977667152881622,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7327135443017724,
+      "eval_loss": 0.7388784289360046,
+      "eval_mean_token_accuracy": 0.8007041252730938,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 89.7692,
+      "eval_samples_per_second": 15.818,
+      "eval_steps_per_second": 1.983,
+      "step": 340
+    },
+    {
+      "entropy": 0.819331557303667,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 1.0407779216766357,
+      "learning_rate": 0.0003155067235145364,
+      "loss": 0.7456695556640625,
+      "mean_token_accuracy": 0.7936980701982975,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.8065585420372781,
+      "eval_loss": 0.735519528388977,
+      "eval_mean_token_accuracy": 0.7974206296245704,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 89.4336,
+      "eval_samples_per_second": 15.878,
+      "eval_steps_per_second": 1.99,
+      "step": 360
+    },
+    {
+      "entropy": 0.8114653021097183,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.1209490299224854,
+      "learning_rate": 0.0003330836997549005,
+      "loss": 0.7340899467468261,
+      "mean_token_accuracy": 0.7914272703230381,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.8025527285056168,
+      "eval_loss": 0.7295576930046082,
+      "eval_mean_token_accuracy": 0.7990576628218876,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 89.9579,
+      "eval_samples_per_second": 15.785,
+      "eval_steps_per_second": 1.979,
+      "step": 380
+    },
+    {
+      "entropy": 0.8042497783899307,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 1.2437798976898193,
+      "learning_rate": 0.0003506606759952647,
+      "loss": 0.7381744384765625,
+      "mean_token_accuracy": 0.7967362694442273,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7717011336530193,
+      "eval_loss": 0.7436173558235168,
+      "eval_mean_token_accuracy": 0.7975849482450592,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 89.8411,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 400
+    },
+    {
+      "entropy": 0.7987750010056929,
+      "epoch": 1.012070006035003,
+      "grad_norm": 1.2390918731689453,
+      "learning_rate": 0.000364721224843344,
+      "loss": 0.7317790031433106,
+      "mean_token_accuracy": 0.7986087551364651,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7333170604170038,
+      "eval_loss": 0.7456948757171631,
+      "eval_mean_token_accuracy": 0.7964897577682238,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 89.7288,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 420
+    },
+    {
+      "entropy": 0.7372976846992969,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.1552740335464478,
+      "learning_rate": 0.00036468510102269575,
+      "loss": 0.6734249114990234,
+      "mean_token_accuracy": 0.805834824591875,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.7130355526891987,
+      "eval_loss": 0.7555333375930786,
+      "eval_mean_token_accuracy": 0.7961652781186479,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 89.8713,
+      "eval_samples_per_second": 15.8,
+      "eval_steps_per_second": 1.981,
+      "step": 440
+    },
+    {
+      "entropy": 0.7465668715536594,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 1.1991469860076904,
+      "learning_rate": 0.0003645973816745066,
+      "loss": 0.6871879577636719,
+      "mean_token_accuracy": 0.8042450234293937,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7772979542110743,
+      "eval_loss": 0.7561826705932617,
+      "eval_mean_token_accuracy": 0.7953876427720102,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 89.8395,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 460
+    },
+    {
+      "entropy": 0.7477855801582336,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 1.325486660003662,
+      "learning_rate": 0.00036445809162231435,
+      "loss": 0.6836830139160156,
+      "mean_token_accuracy": 0.8021619468927383,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.7098110676481483,
+      "eval_loss": 0.758007287979126,
+      "eval_mean_token_accuracy": 0.7909953189030122,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 89.6101,
+      "eval_samples_per_second": 15.846,
+      "eval_steps_per_second": 1.986,
+      "step": 480
+    },
+    {
+      "entropy": 0.7632691070437432,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 1.0122262239456177,
+      "learning_rate": 0.0003642672702835562,
+      "loss": 0.7071213245391845,
+      "mean_token_accuracy": 0.8000680930912495,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7645382720432924,
+      "eval_loss": 0.7492606043815613,
+      "eval_mean_token_accuracy": 0.7976858676149604,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 89.6092,
+      "eval_samples_per_second": 15.847,
+      "eval_steps_per_second": 1.986,
+      "step": 500
+    },
+    {
+      "entropy": 0.7623726457357407,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.367774486541748,
+      "learning_rate": 0.00036402497165841384,
+      "loss": 0.6965175151824952,
+      "mean_token_accuracy": 0.8016501650214195,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7868083715438843,
+      "eval_loss": 0.7481449842453003,
+      "eval_mean_token_accuracy": 0.799330582779445,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 89.8241,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 520
+    },
+    {
+      "entropy": 0.7535207405686378,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 1.213614821434021,
+      "learning_rate": 0.00036373126431453207,
+      "loss": 0.7078310489654541,
+      "mean_token_accuracy": 0.7998907208442688,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7829091950748743,
+      "eval_loss": 0.7416072487831116,
+      "eval_mean_token_accuracy": 0.7969146250339036,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 89.4607,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 540
+    },
+    {
+      "entropy": 0.7527363449335098,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 1.0929982662200928,
+      "learning_rate": 0.00036338623136761495,
+      "loss": 0.7058369636535644,
+      "mean_token_accuracy": 0.7997087433934211,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.728419938114252,
+      "eval_loss": 0.744873046875,
+      "eval_mean_token_accuracy": 0.7966634296299366,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 89.6597,
+      "eval_samples_per_second": 15.838,
+      "eval_steps_per_second": 1.985,
+      "step": 560
+    },
+    {
+      "entropy": 0.7406099259853363,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.409364104270935,
+      "learning_rate": 0.00036298997045790515,
+      "loss": 0.6902852058410645,
+      "mean_token_accuracy": 0.805386807769537,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.6991632758231645,
+      "eval_loss": 0.7480319738388062,
+      "eval_mean_token_accuracy": 0.7970068006033308,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 89.8329,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 580
+    },
+    {
+      "entropy": 0.7459449704736472,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.7053111791610718,
+      "learning_rate": 0.00036254259372255277,
+      "loss": 0.7044640064239502,
+      "mean_token_accuracy": 0.8005018755793571,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7454566660891758,
+      "eval_loss": 0.7551484704017639,
+      "eval_mean_token_accuracy": 0.7897093731365846,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 89.8313,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 600
+    },
+    {
+      "entropy": 0.7621391348540782,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 1.174443244934082,
+      "learning_rate": 0.000362044227763882,
+      "loss": 0.7111573696136475,
+      "mean_token_accuracy": 0.798752411454916,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7664387949397055,
+      "eval_loss": 0.7365804314613342,
+      "eval_mean_token_accuracy": 0.7963101200843126,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 89.9623,
+      "eval_samples_per_second": 15.784,
+      "eval_steps_per_second": 1.979,
+      "step": 620
+    },
+    {
+      "entropy": 0.7424944408237935,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.2898184061050415,
+      "learning_rate": 0.000361495013613564,
+      "loss": 0.6913110733032226,
+      "mean_token_accuracy": 0.8009015507996082,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7255479529332579,
+      "eval_loss": 0.7345991730690002,
+      "eval_mean_token_accuracy": 0.8013592945056015,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 89.817,
+      "eval_samples_per_second": 15.81,
+      "eval_steps_per_second": 1.982,
+      "step": 640
+    },
+    {
+      "entropy": 0.7303367637097835,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.1310476064682007,
+      "learning_rate": 0.00036089510669270666,
+      "loss": 0.6875617027282714,
+      "mean_token_accuracy": 0.8035397171974182,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7101308248016271,
+      "eval_loss": 0.7394917011260986,
+      "eval_mean_token_accuracy": 0.7990792942850777,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 89.7431,
+      "eval_samples_per_second": 15.823,
+      "eval_steps_per_second": 1.983,
+      "step": 660
+    },
+    {
+      "entropy": 0.7387678384780884,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.9645389914512634,
+      "learning_rate": 0.0003602446767678725,
+      "loss": 0.6911204338073731,
+      "mean_token_accuracy": 0.8056333385407924,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7281073737010527,
+      "eval_loss": 0.7376708984375,
+      "eval_mean_token_accuracy": 0.7997356415464637,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 90.0338,
+      "eval_samples_per_second": 15.772,
+      "eval_steps_per_second": 1.977,
+      "step": 680
+    },
+    {
+      "entropy": 0.7449554048478604,
+      "epoch": 1.687990343995172,
+      "grad_norm": 1.9069629907608032,
+      "learning_rate": 0.0003595439079030364,
+      "loss": 0.7042286396026611,
+      "mean_token_accuracy": 0.8018639378249646,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7022633817088738,
+      "eval_loss": 0.7320939898490906,
+      "eval_mean_token_accuracy": 0.8006051541044471,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 89.8948,
+      "eval_samples_per_second": 15.796,
+      "eval_steps_per_second": 1.98,
+      "step": 700
+    },
+    {
+      "entropy": 0.723249051719904,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.8333877325057983,
+      "learning_rate": 0.00035879299840749777,
+      "loss": 0.7043714046478271,
+      "mean_token_accuracy": 0.8028345607221127,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.672436946898364,
+      "eval_loss": 0.713233470916748,
+      "eval_mean_token_accuracy": 0.8047760577684038,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 89.6394,
+      "eval_samples_per_second": 15.841,
+      "eval_steps_per_second": 1.986,
+      "step": 720
+    },
+    {
+      "entropy": 0.730014206469059,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.4580037593841553,
+      "learning_rate": 0.00035799216077976137,
+      "loss": 0.6985176563262939,
+      "mean_token_accuracy": 0.8015547141432762,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.6908873795123582,
+      "eval_loss": 0.7198202013969421,
+      "eval_mean_token_accuracy": 0.8024131682481659,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 89.3771,
+      "eval_samples_per_second": 15.888,
+      "eval_steps_per_second": 1.992,
+      "step": 740
+    },
+    {
+      "entropy": 0.7219605796039105,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.51682710647583,
+      "learning_rate": 0.000357141621647403,
+      "loss": 0.6844424724578857,
+      "mean_token_accuracy": 0.8032813094556331,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7113401109582922,
+      "eval_loss": 0.7195309400558472,
+      "eval_mean_token_accuracy": 0.8019187035185568,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 89.4658,
+      "eval_samples_per_second": 15.872,
+      "eval_steps_per_second": 1.99,
+      "step": 760
+    },
+    {
+      "entropy": 0.729826743900776,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 1.0838000774383545,
+      "learning_rate": 0.0003562416217029361,
+      "loss": 0.685858678817749,
+      "mean_token_accuracy": 0.8050567395985126,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.649370267484965,
+      "eval_loss": 0.7177029252052307,
+      "eval_mean_token_accuracy": 0.8050174043419656,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 89.2287,
+      "eval_samples_per_second": 15.914,
+      "eval_steps_per_second": 1.995,
+      "step": 780
+    },
+    {
+      "entropy": 0.6799364153295755,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 1.1364924907684326,
+      "learning_rate": 0.00035529241563569903,
+      "loss": 0.6714428901672364,
+      "mean_token_accuracy": 0.8087341658771038,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6603799047095052,
+      "eval_loss": 0.7201809287071228,
+      "eval_mean_token_accuracy": 0.802863868769635,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 89.2718,
+      "eval_samples_per_second": 15.906,
+      "eval_steps_per_second": 1.994,
+      "step": 800
+    },
+    {
+      "entropy": 0.7078610375523567,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 1.2191942930221558,
+      "learning_rate": 0.0003542942720597808,
+      "loss": 0.6996590614318847,
+      "mean_token_accuracy": 0.8049512051045895,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.7240545264120852,
+      "eval_loss": 0.7301309108734131,
+      "eval_mean_token_accuracy": 0.8015973166133581,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 89.3438,
+      "eval_samples_per_second": 15.894,
+      "eval_steps_per_second": 1.992,
+      "step": 820
+    },
+    {
+      "entropy": 0.6557391556826505,
+      "epoch": 2.024140012070006,
+      "grad_norm": 1.36203134059906,
+      "learning_rate": 0.0003532474734380064,
+      "loss": 0.6327113628387451,
+      "mean_token_accuracy": 0.8142829847026181,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6573808648613062,
+      "eval_loss": 0.7334365844726562,
+      "eval_mean_token_accuracy": 0.8043633893634496,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 89.3359,
+      "eval_samples_per_second": 15.895,
+      "eval_steps_per_second": 1.992,
+      "step": 840
+    },
+    {
+      "entropy": 0.6409128502011299,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 1.5804955959320068,
+      "learning_rate": 0.0003521523160020035,
+      "loss": 0.6023559093475341,
+      "mean_token_accuracy": 0.8222251623868942,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6983234788594621,
+      "eval_loss": 0.7232135534286499,
+      "eval_mean_token_accuracy": 0.8010654268639811,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 89.2615,
+      "eval_samples_per_second": 15.908,
+      "eval_steps_per_second": 1.994,
+      "step": 860
+    },
+    {
+      "entropy": 0.6151149723678827,
+      "epoch": 2.12070006035003,
+      "grad_norm": 1.1975436210632324,
+      "learning_rate": 0.00035100910966837193,
+      "loss": 0.5803564548492431,
+      "mean_token_accuracy": 0.8256325736641884,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6636555573243773,
+      "eval_loss": 0.7205662727355957,
+      "eval_mean_token_accuracy": 0.8036768369460374,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 89.2957,
+      "eval_samples_per_second": 15.902,
+      "eval_steps_per_second": 1.993,
+      "step": 880
+    },
+    {
+      "entropy": 0.6201727617532015,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9907131195068359,
+      "learning_rate": 0.0003498181779509813,
+      "loss": 0.6101872444152832,
+      "mean_token_accuracy": 0.8196512959897518,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.6473788086617931,
+      "eval_loss": 0.7062397003173828,
+      "eval_mean_token_accuracy": 0.8078362184963869,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 88.6198,
+      "eval_samples_per_second": 16.024,
+      "eval_steps_per_second": 2.009,
+      "step": 900
+    },
+    {
+      "entropy": 0.6097332935780286,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.9055633544921875,
+      "learning_rate": 0.00034857985786942026,
+      "loss": 0.5956094264984131,
+      "mean_token_accuracy": 0.8240575887262821,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.612629868341296,
+      "eval_loss": 0.7106770277023315,
+      "eval_mean_token_accuracy": 0.805324277181304,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 89.4383,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 920
+    },
+    {
+      "entropy": 0.6182701248675585,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.9513736367225647,
+      "learning_rate": 0.00034729449985362404,
+      "loss": 0.6019269466400147,
+      "mean_token_accuracy": 0.8187106013298034,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6437647456533453,
+      "eval_loss": 0.7173502445220947,
+      "eval_mean_token_accuracy": 0.8056439573175451,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.9568,
+      "eval_samples_per_second": 15.963,
+      "eval_steps_per_second": 2.001,
+      "step": 940
+    },
+    {
+      "entropy": 0.6508570514619351,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.9606207013130188,
+      "learning_rate": 0.0003459624676447067,
+      "loss": 0.6296409606933594,
+      "mean_token_accuracy": 0.8167732007801533,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6288247135248077,
+      "eval_loss": 0.7160675525665283,
+      "eval_mean_token_accuracy": 0.8066010689467527,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 89.0044,
+      "eval_samples_per_second": 15.954,
+      "eval_steps_per_second": 2.0,
+      "step": 960
+    },
+    {
+      "entropy": 0.6380081418901682,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.9677181839942932,
+      "learning_rate": 0.000344584138192027,
+      "loss": 0.6210546493530273,
+      "mean_token_accuracy": 0.8186135642230511,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6407805031604981,
+      "eval_loss": 0.7055043578147888,
+      "eval_mean_token_accuracy": 0.8070103157772107,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 89.5606,
+      "eval_samples_per_second": 15.855,
+      "eval_steps_per_second": 1.987,
+      "step": 980
+    },
+    {
+      "entropy": 0.6310251638293266,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 2.1046407222747803,
+      "learning_rate": 0.000343159901546516,
+      "loss": 0.6165830135345459,
+      "mean_token_accuracy": 0.8207855716347694,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6461507470420237,
+      "eval_loss": 0.7292136549949646,
+      "eval_mean_token_accuracy": 0.8030418261383356,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 89.315,
+      "eval_samples_per_second": 15.899,
+      "eval_steps_per_second": 1.993,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6256617344915867,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 1.1768349409103394,
+      "learning_rate": 0.0003416901607502972,
+      "loss": 0.6142709255218506,
+      "mean_token_accuracy": 0.8220669947564602,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6150523993406403,
+      "eval_loss": 0.7019563317298889,
+      "eval_mean_token_accuracy": 0.8094323099998946,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 89.419,
+      "eval_samples_per_second": 15.88,
+      "eval_steps_per_second": 1.991,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6126735735684633,
+      "epoch": 2.506940253470127,
+      "grad_norm": 1.8065496683120728,
+      "learning_rate": 0.00034017533172263055,
+      "loss": 0.6056458473205566,
+      "mean_token_accuracy": 0.8203762136399746,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6544345456562685,
+      "eval_loss": 0.7086517214775085,
+      "eval_mean_token_accuracy": 0.806709805901131,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 88.981,
+      "eval_samples_per_second": 15.958,
+      "eval_steps_per_second": 2.0,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6311754353344441,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 1.0576369762420654,
+      "learning_rate": 0.00033861584314221236,
+      "loss": 0.6176392555236816,
+      "mean_token_accuracy": 0.8177410490810871,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6273799200406235,
+      "eval_loss": 0.708163321018219,
+      "eval_mean_token_accuracy": 0.8032017099053672,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 89.7668,
+      "eval_samples_per_second": 15.819,
+      "eval_steps_per_second": 1.983,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6200249589979648,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.997065007686615,
+      "learning_rate": 0.0003370121363258637,
+      "loss": 0.6203888893127442,
+      "mean_token_accuracy": 0.8196275025606156,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6457576811983344,
+      "eval_loss": 0.6972290277481079,
+      "eval_mean_token_accuracy": 0.8095980849158898,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 89.2027,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 1080
+    },
+    {
+      "entropy": 0.626810473203659,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.865475058555603,
+      "learning_rate": 0.0003353646651036438,
+      "loss": 0.6102585792541504,
+      "mean_token_accuracy": 0.819708751142025,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.6412022431914726,
+      "eval_loss": 0.7036139965057373,
+      "eval_mean_token_accuracy": 0.8070317135098275,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 89.4733,
+      "eval_samples_per_second": 15.871,
+      "eval_steps_per_second": 1.989,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6581619590520859,
+      "epoch": 2.700060350030175,
+      "grad_norm": 1.00307035446167,
+      "learning_rate": 0.00033367389569042064,
+      "loss": 0.6397720336914062,
+      "mean_token_accuracy": 0.81413309648633,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6288736402318719,
+      "eval_loss": 0.6966825723648071,
+      "eval_mean_token_accuracy": 0.8094039669867312,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 88.7341,
+      "eval_samples_per_second": 16.003,
+      "eval_steps_per_second": 2.006,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6399412982165813,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.8589329719543457,
+      "learning_rate": 0.0003319403065539384,
+      "loss": 0.6347338199615479,
+      "mean_token_accuracy": 0.8139534957706929,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.683491913455256,
+      "eval_loss": 0.6984149217605591,
+      "eval_mean_token_accuracy": 0.8083989469522841,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 89.3929,
+      "eval_samples_per_second": 15.885,
+      "eval_steps_per_second": 1.991,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6264835461974144,
+      "epoch": 2.796620398310199,
+      "grad_norm": 0.7807853817939758,
+      "learning_rate": 0.0003301643882794163,
+      "loss": 0.6162629127502441,
+      "mean_token_accuracy": 0.8216292977333068,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6578792159476977,
+      "eval_loss": 0.6921441555023193,
+      "eval_mean_token_accuracy": 0.8094299177775223,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 89.2356,
+      "eval_samples_per_second": 15.913,
+      "eval_steps_per_second": 1.995,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6269129011780024,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 1.1675593852996826,
+      "learning_rate": 0.0003283466434307189,
+      "loss": 0.6235532760620117,
+      "mean_token_accuracy": 0.8188897788524627,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6533248447970058,
+      "eval_loss": 0.696849524974823,
+      "eval_mean_token_accuracy": 0.8102703713968898,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2602,
+      "eval_samples_per_second": 16.089,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.6259873781353236,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.9596486687660217,
+      "learning_rate": 0.00032648758640813655,
+      "loss": 0.6118728160858155,
+      "mean_token_accuracy": 0.8171859130263328,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6212078754821521,
+      "eval_loss": 0.6942870020866394,
+      "eval_mean_token_accuracy": 0.810484343365337,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 89.4375,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 1200
+    },
+    {
+      "entropy": 0.6142537571489811,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 0.9625058770179749,
+      "learning_rate": 0.00032458774330281615,
+      "loss": 0.6060261249542236,
+      "mean_token_accuracy": 0.8223113484680653,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6553665667437436,
+      "eval_loss": 0.6835269331932068,
+      "eval_mean_token_accuracy": 0.8134723968720168,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 89.5092,
+      "eval_samples_per_second": 15.864,
+      "eval_steps_per_second": 1.989,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6301972426474094,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.9657333493232727,
+      "learning_rate": 0.0003226476517478835,
+      "loss": 0.6277759075164795,
+      "mean_token_accuracy": 0.8164266042411328,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6594253118788258,
+      "eval_loss": 0.6820935606956482,
+      "eval_mean_token_accuracy": 0.8132368198941263,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 89.6211,
+      "eval_samples_per_second": 15.844,
+      "eval_steps_per_second": 1.986,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5346100655469027,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8541185259819031,
+      "learning_rate": 0.0003206678607662996,
+      "loss": 0.5261692047119141,
+      "mean_token_accuracy": 0.8382289425119177,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5626645662476507,
+      "eval_loss": 0.7028542160987854,
+      "eval_mean_token_accuracy": 0.8133395572056931,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 89.1267,
+      "eval_samples_per_second": 15.932,
+      "eval_steps_per_second": 1.997,
+      "step": 1260
+    },
+    {
+      "entropy": 0.5001101832836866,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.8208815455436707,
+      "learning_rate": 0.0003186489306154935,
+      "loss": 0.4850132942199707,
+      "mean_token_accuracy": 0.8468983843922615,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5889462046743779,
+      "eval_loss": 0.691015899181366,
+      "eval_mean_token_accuracy": 0.8135226466012805,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.9677,
+      "eval_samples_per_second": 15.961,
+      "eval_steps_per_second": 2.001,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5242766339331866,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 0.7893072962760925,
+      "learning_rate": 0.0003165914326288163,
+      "loss": 0.5061577320098877,
+      "mean_token_accuracy": 0.8425608821213245,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5543691962957382,
+      "eval_loss": 0.6913734078407288,
+      "eval_mean_token_accuracy": 0.8154889680026622,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 89.6963,
+      "eval_samples_per_second": 15.831,
+      "eval_steps_per_second": 1.984,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5327721010893584,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.8992080092430115,
+      "learning_rate": 0.0003144959490538604,
+      "loss": 0.5068036556243897,
+      "mean_token_accuracy": 0.8420542575418949,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5734540803378887,
+      "eval_loss": 0.6948716044425964,
+      "eval_mean_token_accuracy": 0.8137217944257715,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 89.4597,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5247377116233111,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.9291845560073853,
+      "learning_rate": 0.0003123630728876902,
+      "loss": 0.5023272037506104,
+      "mean_token_accuracy": 0.8435182586312294,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5492362935891312,
+      "eval_loss": 0.6980633735656738,
+      "eval_mean_token_accuracy": 0.8155082919624415,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 89.2124,
+      "eval_samples_per_second": 15.917,
+      "eval_steps_per_second": 1.995,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5380499072372913,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 1.0477524995803833,
+      "learning_rate": 0.00031019340770903136,
+      "loss": 0.5264591693878173,
+      "mean_token_accuracy": 0.8391959741711617,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5896503307511297,
+      "eval_loss": 0.6997144222259521,
+      "eval_mean_token_accuracy": 0.8124373062942805,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 89.1147,
+      "eval_samples_per_second": 15.935,
+      "eval_steps_per_second": 1.997,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5460653610527515,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 1.2473069429397583,
+      "learning_rate": 0.00030798756750746476,
+      "loss": 0.5319677829742432,
+      "mean_token_accuracy": 0.8364320226013661,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5659052182114526,
+      "eval_loss": 0.6924049258232117,
+      "eval_mean_token_accuracy": 0.8153053585732921,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 89.6468,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5367407951503992,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 1.2679712772369385,
+      "learning_rate": 0.00030574617650967485,
+      "loss": 0.5190927028656006,
+      "mean_token_accuracy": 0.8399469174444676,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5581759144081159,
+      "eval_loss": 0.6933317184448242,
+      "eval_mean_token_accuracy": 0.8139757862251796,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 89.0749,
+      "eval_samples_per_second": 15.942,
+      "eval_steps_per_second": 1.998,
+      "step": 1400
+    },
+    {
+      "entropy": 0.530736118927598,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.9717612862586975,
+      "learning_rate": 0.0003034698690028009,
+      "loss": 0.512363052368164,
+      "mean_token_accuracy": 0.8419127956032753,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5233335836549823,
+      "eval_loss": 0.6958550214767456,
+      "eval_mean_token_accuracy": 0.8158778501360604,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 88.8664,
+      "eval_samples_per_second": 15.979,
+      "eval_steps_per_second": 2.003,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5327631575986743,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7649208903312683,
+      "learning_rate": 0.00030115928915494116,
+      "loss": 0.5202179431915284,
+      "mean_token_accuracy": 0.8384683802723885,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.5960972657364406,
+      "eval_loss": 0.6984783411026001,
+      "eval_mean_token_accuracy": 0.8095930597085631,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 89.5168,
+      "eval_samples_per_second": 15.863,
+      "eval_steps_per_second": 1.988,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5428155034780502,
+      "epoch": 3.51901025950513,
+      "grad_norm": 1.2790874242782593,
+      "learning_rate": 0.00029881509083286116,
+      "loss": 0.524717378616333,
+      "mean_token_accuracy": 0.8346644505858422,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.550318562917495,
+      "eval_loss": 0.6988471746444702,
+      "eval_mean_token_accuracy": 0.8132969897784544,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 89.2598,
+      "eval_samples_per_second": 15.909,
+      "eval_steps_per_second": 1.994,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5489993002265692,
+      "epoch": 3.567290283645142,
+      "grad_norm": 1.055821418762207,
+      "learning_rate": 0.00029643793741695687,
+      "loss": 0.5421150207519532,
+      "mean_token_accuracy": 0.8345710933208466,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5816821718818685,
+      "eval_loss": 0.6853668689727783,
+      "eval_mean_token_accuracy": 0.8142332360985574,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 89.0424,
+      "eval_samples_per_second": 15.947,
+      "eval_steps_per_second": 1.999,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5451365381479263,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.9027532935142517,
+      "learning_rate": 0.00029402850161352583,
+      "loss": 0.5463609218597412,
+      "mean_token_accuracy": 0.8369908526539802,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5344233209832331,
+      "eval_loss": 0.6794531345367432,
+      "eval_mean_token_accuracy": 0.8177759824843889,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 89.123,
+      "eval_samples_per_second": 15.933,
+      "eval_steps_per_second": 1.997,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5521467242389917,
+      "epoch": 3.663850331925166,
+      "grad_norm": 1.4606311321258545,
+      "learning_rate": 0.0002915874652643997,
+      "loss": 0.5421902179718018,
+      "mean_token_accuracy": 0.8339455373585224,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5807709094513668,
+      "eval_loss": 0.6949691772460938,
+      "eval_mean_token_accuracy": 0.8121141439743256,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.5437,
+      "eval_samples_per_second": 16.037,
+      "eval_steps_per_second": 2.01,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5455610822886229,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7269843220710754,
+      "learning_rate": 0.0002891155191539905,
+      "loss": 0.5407682895660401,
+      "mean_token_accuracy": 0.8393648222088814,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.6032205542151847,
+      "eval_loss": 0.6790974140167236,
+      "eval_mean_token_accuracy": 0.8165463112043531,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 89.517,
+      "eval_samples_per_second": 15.863,
+      "eval_steps_per_second": 1.988,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5474163968116045,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.8148866295814514,
+      "learning_rate": 0.00028661336281380717,
+      "loss": 0.5394441604614257,
+      "mean_token_accuracy": 0.8336028568446636,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5797032238392348,
+      "eval_loss": 0.670870304107666,
+      "eval_mean_token_accuracy": 0.8182691593518417,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 89.4468,
+      "eval_samples_per_second": 15.875,
+      "eval_steps_per_second": 1.99,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5511750865727663,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.3441656827926636,
+      "learning_rate": 0.0002840817043244964,
+      "loss": 0.5378933906555176,
+      "mean_token_accuracy": 0.8349151819944381,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5297858750217417,
+      "eval_loss": 0.680833637714386,
+      "eval_mean_token_accuracy": 0.8180924275617921,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 89.5417,
+      "eval_samples_per_second": 15.859,
+      "eval_steps_per_second": 1.988,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5379613988101483,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9813932776451111,
+      "learning_rate": 0.00028152126011546396,
+      "loss": 0.5340342044830322,
+      "mean_token_accuracy": 0.8346010789275169,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5811898510777549,
+      "eval_loss": 0.6764267683029175,
+      "eval_mean_token_accuracy": 0.8180528236239144,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 88.7831,
+      "eval_samples_per_second": 15.994,
+      "eval_steps_per_second": 2.005,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5607876226305961,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.9617928266525269,
+      "learning_rate": 0.00027893275476213383,
+      "loss": 0.5434338569641113,
+      "mean_token_accuracy": 0.8332898400723934,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5474644318390428,
+      "eval_loss": 0.6740226745605469,
+      "eval_mean_token_accuracy": 0.8185851306058047,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 89.3246,
+      "eval_samples_per_second": 15.897,
+      "eval_steps_per_second": 1.993,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5354843523353339,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.5320990085601807,
+      "learning_rate": 0.0002763169207809021,
+      "loss": 0.5251852989196777,
+      "mean_token_accuracy": 0.8416872084140777,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.5436856410141742,
+      "eval_loss": 0.6675190925598145,
+      "eval_mean_token_accuracy": 0.8202396507343549,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 89.6328,
+      "eval_samples_per_second": 15.842,
+      "eval_steps_per_second": 1.986,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5418571938167919,
+      "epoch": 4.0,
+      "grad_norm": 4.690354347229004,
+      "learning_rate": 0.000273674498421843,
+      "loss": 0.548275089263916,
+      "mean_token_accuracy": 0.836557479647847,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5497228983747825,
+      "eval_loss": 0.6621812582015991,
+      "eval_mean_token_accuracy": 0.8205679802412398,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 89.6272,
+      "eval_samples_per_second": 15.843,
+      "eval_steps_per_second": 1.986,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4195931971073151,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6678048968315125,
+      "learning_rate": 0.0002710062354592273,
+      "loss": 0.3970795154571533,
+      "mean_token_accuracy": 0.8719995342195034,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.46495268753405367,
+      "eval_loss": 0.7061581611633301,
+      "eval_mean_token_accuracy": 0.8190107995204712,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 89.6387,
+      "eval_samples_per_second": 15.841,
+      "eval_steps_per_second": 1.986,
+      "step": 1680
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.6830788507998003e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1700/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1700/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1700/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1700/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1700/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..7ce4b072fcbe5b636867e13937c56fedee3b68b4
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1700/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.02728394274790723,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1700/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1700/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1700/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1700/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1700/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1700/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1700/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1700/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..fc6a9c8c7684b2886f6b5058f1d88e17175569d6
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1700/trainer_state.json
@@ -0,0 +1,1819 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 4.096560048280024,
+  "eval_steps": 20,
+  "global_step": 1700,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.024795526266098,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.945375919342041,
+      "learning_rate": 1.6698127428345936e-05,
+      "loss": 1.7825420379638672,
+      "mean_token_accuracy": 0.6315708436071873,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.29921497588747,
+      "eval_loss": 1.169131875038147,
+      "eval_mean_token_accuracy": 0.7185706342204233,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 91.7898,
+      "eval_samples_per_second": 15.47,
+      "eval_steps_per_second": 1.939,
+      "step": 20
+    },
+    {
+      "entropy": 1.0346344470977784,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.7916979789733887,
+      "learning_rate": 3.427510366871008e-05,
+      "loss": 0.9480395317077637,
+      "mean_token_accuracy": 0.7543429024517536,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 0.9673695430326997,
+      "eval_loss": 0.8588430285453796,
+      "eval_mean_token_accuracy": 0.7737233002534073,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 89.8219,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 40
+    },
+    {
+      "entropy": 0.8977566614747048,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7972187995910645,
+      "learning_rate": 5.185207990907422e-05,
+      "loss": 0.8029604911804199,
+      "mean_token_accuracy": 0.7818016350269318,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8436836982041263,
+      "eval_loss": 0.7893033623695374,
+      "eval_mean_token_accuracy": 0.7858914984076211,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 89.9349,
+      "eval_samples_per_second": 15.789,
+      "eval_steps_per_second": 1.979,
+      "step": 60
+    },
+    {
+      "entropy": 0.8650321319699288,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.2128729820251465,
+      "learning_rate": 6.942905614943836e-05,
+      "loss": 0.77417893409729,
+      "mean_token_accuracy": 0.7875479467213153,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8209857180547179,
+      "eval_loss": 0.7572280168533325,
+      "eval_mean_token_accuracy": 0.7925970383574453,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 89.9459,
+      "eval_samples_per_second": 15.787,
+      "eval_steps_per_second": 1.979,
+      "step": 80
+    },
+    {
+      "entropy": 0.8096660938113928,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.1254287958145142,
+      "learning_rate": 8.700603238980251e-05,
+      "loss": 0.7170331001281738,
+      "mean_token_accuracy": 0.7994555257260799,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.792780208788561,
+      "eval_loss": 0.7442984580993652,
+      "eval_mean_token_accuracy": 0.7954704573984896,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 90.294,
+      "eval_samples_per_second": 15.726,
+      "eval_steps_per_second": 1.971,
+      "step": 100
+    },
+    {
+      "entropy": 0.7979829199612141,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2140507698059082,
+      "learning_rate": 0.00010458300863016666,
+      "loss": 0.7141861915588379,
+      "mean_token_accuracy": 0.8000014744699001,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.7938692486018277,
+      "eval_loss": 0.7279797196388245,
+      "eval_mean_token_accuracy": 0.7989798734027348,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 89.7189,
+      "eval_samples_per_second": 15.827,
+      "eval_steps_per_second": 1.984,
+      "step": 120
+    },
+    {
+      "entropy": 0.7916347607970238,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.0506458282470703,
+      "learning_rate": 0.0001221599848705308,
+      "loss": 0.7117055416107178,
+      "mean_token_accuracy": 0.7970046654343605,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.8052781063519167,
+      "eval_loss": 0.722576916217804,
+      "eval_mean_token_accuracy": 0.7982530523551984,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 90.0376,
+      "eval_samples_per_second": 15.771,
+      "eval_steps_per_second": 1.977,
+      "step": 140
+    },
+    {
+      "entropy": 0.7880666613578796,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8409207463264465,
+      "learning_rate": 0.00013973696111089492,
+      "loss": 0.7052523136138916,
+      "mean_token_accuracy": 0.8000123649835587,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7787443134891853,
+      "eval_loss": 0.7137772440910339,
+      "eval_mean_token_accuracy": 0.8014209287220173,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 90.2562,
+      "eval_samples_per_second": 15.733,
+      "eval_steps_per_second": 1.972,
+      "step": 160
+    },
+    {
+      "entropy": 0.7696007996797561,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.0182477235794067,
+      "learning_rate": 0.00015731393735125907,
+      "loss": 0.7008797645568847,
+      "mean_token_accuracy": 0.8007259473204613,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.752259682068664,
+      "eval_loss": 0.7171286344528198,
+      "eval_mean_token_accuracy": 0.80052122574174,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 90.0797,
+      "eval_samples_per_second": 15.764,
+      "eval_steps_per_second": 1.976,
+      "step": 180
+    },
+    {
+      "entropy": 0.7828126326203346,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9553632736206055,
+      "learning_rate": 0.0001748909135916232,
+      "loss": 0.7046853542327881,
+      "mean_token_accuracy": 0.7988562889397144,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7389638169427936,
+      "eval_loss": 0.7187935709953308,
+      "eval_mean_token_accuracy": 0.7995899999409579,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 89.9223,
+      "eval_samples_per_second": 15.791,
+      "eval_steps_per_second": 1.979,
+      "step": 200
+    },
+    {
+      "entropy": 0.7909286297857762,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 1.134459376335144,
+      "learning_rate": 0.00019246788983198735,
+      "loss": 0.6989312171936035,
+      "mean_token_accuracy": 0.7991565138101577,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7496965199374082,
+      "eval_loss": 0.7106321454048157,
+      "eval_mean_token_accuracy": 0.8046183442131857,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 89.6899,
+      "eval_samples_per_second": 15.832,
+      "eval_steps_per_second": 1.985,
+      "step": 220
+    },
+    {
+      "entropy": 0.7651963323354721,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.9557585716247559,
+      "learning_rate": 0.0002100448660723515,
+      "loss": 0.6984559059143066,
+      "mean_token_accuracy": 0.8019823372364044,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7615563079212488,
+      "eval_loss": 0.7098539471626282,
+      "eval_mean_token_accuracy": 0.8047233508544022,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 90.3752,
+      "eval_samples_per_second": 15.712,
+      "eval_steps_per_second": 1.97,
+      "step": 240
+    },
+    {
+      "entropy": 0.787507726252079,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.9279311299324036,
+      "learning_rate": 0.00022762184231271568,
+      "loss": 0.708138370513916,
+      "mean_token_accuracy": 0.8013308703899383,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7856640896100676,
+      "eval_loss": 0.7191570401191711,
+      "eval_mean_token_accuracy": 0.8024677249153008,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 93.7336,
+      "eval_samples_per_second": 15.149,
+      "eval_steps_per_second": 1.899,
+      "step": 260
+    },
+    {
+      "entropy": 0.768489234894514,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.7992776036262512,
+      "learning_rate": 0.0002451988185530798,
+      "loss": 0.7035849571228028,
+      "mean_token_accuracy": 0.8031105332076549,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7682498740346244,
+      "eval_loss": 0.7149022221565247,
+      "eval_mean_token_accuracy": 0.7973497995499814,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 89.7309,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 280
+    },
+    {
+      "entropy": 0.7720955617725849,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.1931514739990234,
+      "learning_rate": 0.0002627757947934439,
+      "loss": 0.6965402603149414,
+      "mean_token_accuracy": 0.8040348663926125,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7214708304807042,
+      "eval_loss": 0.7350823879241943,
+      "eval_mean_token_accuracy": 0.7942911925610532,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 89.6451,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 300
+    },
+    {
+      "entropy": 0.7930095426738262,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8994467258453369,
+      "learning_rate": 0.0002803527710338081,
+      "loss": 0.7331492900848389,
+      "mean_token_accuracy": 0.796185302734375,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.8142535488926963,
+      "eval_loss": 0.7237834334373474,
+      "eval_mean_token_accuracy": 0.799495273761535,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 90.6759,
+      "eval_samples_per_second": 15.66,
+      "eval_steps_per_second": 1.963,
+      "step": 320
+    },
+    {
+      "entropy": 0.7825102441012859,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 1.8224239349365234,
+      "learning_rate": 0.00029792974727417223,
+      "loss": 0.7331175804138184,
+      "mean_token_accuracy": 0.7977667152881622,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7327135443017724,
+      "eval_loss": 0.7388784289360046,
+      "eval_mean_token_accuracy": 0.8007041252730938,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 89.7692,
+      "eval_samples_per_second": 15.818,
+      "eval_steps_per_second": 1.983,
+      "step": 340
+    },
+    {
+      "entropy": 0.819331557303667,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 1.0407779216766357,
+      "learning_rate": 0.0003155067235145364,
+      "loss": 0.7456695556640625,
+      "mean_token_accuracy": 0.7936980701982975,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.8065585420372781,
+      "eval_loss": 0.735519528388977,
+      "eval_mean_token_accuracy": 0.7974206296245704,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 89.4336,
+      "eval_samples_per_second": 15.878,
+      "eval_steps_per_second": 1.99,
+      "step": 360
+    },
+    {
+      "entropy": 0.8114653021097183,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.1209490299224854,
+      "learning_rate": 0.0003330836997549005,
+      "loss": 0.7340899467468261,
+      "mean_token_accuracy": 0.7914272703230381,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.8025527285056168,
+      "eval_loss": 0.7295576930046082,
+      "eval_mean_token_accuracy": 0.7990576628218876,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 89.9579,
+      "eval_samples_per_second": 15.785,
+      "eval_steps_per_second": 1.979,
+      "step": 380
+    },
+    {
+      "entropy": 0.8042497783899307,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 1.2437798976898193,
+      "learning_rate": 0.0003506606759952647,
+      "loss": 0.7381744384765625,
+      "mean_token_accuracy": 0.7967362694442273,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7717011336530193,
+      "eval_loss": 0.7436173558235168,
+      "eval_mean_token_accuracy": 0.7975849482450592,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 89.8411,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 400
+    },
+    {
+      "entropy": 0.7987750010056929,
+      "epoch": 1.012070006035003,
+      "grad_norm": 1.2390918731689453,
+      "learning_rate": 0.000364721224843344,
+      "loss": 0.7317790031433106,
+      "mean_token_accuracy": 0.7986087551364651,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7333170604170038,
+      "eval_loss": 0.7456948757171631,
+      "eval_mean_token_accuracy": 0.7964897577682238,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 89.7288,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 420
+    },
+    {
+      "entropy": 0.7372976846992969,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.1552740335464478,
+      "learning_rate": 0.00036468510102269575,
+      "loss": 0.6734249114990234,
+      "mean_token_accuracy": 0.805834824591875,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.7130355526891987,
+      "eval_loss": 0.7555333375930786,
+      "eval_mean_token_accuracy": 0.7961652781186479,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 89.8713,
+      "eval_samples_per_second": 15.8,
+      "eval_steps_per_second": 1.981,
+      "step": 440
+    },
+    {
+      "entropy": 0.7465668715536594,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 1.1991469860076904,
+      "learning_rate": 0.0003645973816745066,
+      "loss": 0.6871879577636719,
+      "mean_token_accuracy": 0.8042450234293937,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7772979542110743,
+      "eval_loss": 0.7561826705932617,
+      "eval_mean_token_accuracy": 0.7953876427720102,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 89.8395,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 460
+    },
+    {
+      "entropy": 0.7477855801582336,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 1.325486660003662,
+      "learning_rate": 0.00036445809162231435,
+      "loss": 0.6836830139160156,
+      "mean_token_accuracy": 0.8021619468927383,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.7098110676481483,
+      "eval_loss": 0.758007287979126,
+      "eval_mean_token_accuracy": 0.7909953189030122,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 89.6101,
+      "eval_samples_per_second": 15.846,
+      "eval_steps_per_second": 1.986,
+      "step": 480
+    },
+    {
+      "entropy": 0.7632691070437432,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 1.0122262239456177,
+      "learning_rate": 0.0003642672702835562,
+      "loss": 0.7071213245391845,
+      "mean_token_accuracy": 0.8000680930912495,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7645382720432924,
+      "eval_loss": 0.7492606043815613,
+      "eval_mean_token_accuracy": 0.7976858676149604,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 89.6092,
+      "eval_samples_per_second": 15.847,
+      "eval_steps_per_second": 1.986,
+      "step": 500
+    },
+    {
+      "entropy": 0.7623726457357407,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.367774486541748,
+      "learning_rate": 0.00036402497165841384,
+      "loss": 0.6965175151824952,
+      "mean_token_accuracy": 0.8016501650214195,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7868083715438843,
+      "eval_loss": 0.7481449842453003,
+      "eval_mean_token_accuracy": 0.799330582779445,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 89.8241,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 520
+    },
+    {
+      "entropy": 0.7535207405686378,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 1.213614821434021,
+      "learning_rate": 0.00036373126431453207,
+      "loss": 0.7078310489654541,
+      "mean_token_accuracy": 0.7998907208442688,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7829091950748743,
+      "eval_loss": 0.7416072487831116,
+      "eval_mean_token_accuracy": 0.7969146250339036,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 89.4607,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 540
+    },
+    {
+      "entropy": 0.7527363449335098,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 1.0929982662200928,
+      "learning_rate": 0.00036338623136761495,
+      "loss": 0.7058369636535644,
+      "mean_token_accuracy": 0.7997087433934211,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.728419938114252,
+      "eval_loss": 0.744873046875,
+      "eval_mean_token_accuracy": 0.7966634296299366,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 89.6597,
+      "eval_samples_per_second": 15.838,
+      "eval_steps_per_second": 1.985,
+      "step": 560
+    },
+    {
+      "entropy": 0.7406099259853363,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.409364104270935,
+      "learning_rate": 0.00036298997045790515,
+      "loss": 0.6902852058410645,
+      "mean_token_accuracy": 0.805386807769537,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.6991632758231645,
+      "eval_loss": 0.7480319738388062,
+      "eval_mean_token_accuracy": 0.7970068006033308,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 89.8329,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 580
+    },
+    {
+      "entropy": 0.7459449704736472,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.7053111791610718,
+      "learning_rate": 0.00036254259372255277,
+      "loss": 0.7044640064239502,
+      "mean_token_accuracy": 0.8005018755793571,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7454566660891758,
+      "eval_loss": 0.7551484704017639,
+      "eval_mean_token_accuracy": 0.7897093731365846,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 89.8313,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 600
+    },
+    {
+      "entropy": 0.7621391348540782,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 1.174443244934082,
+      "learning_rate": 0.000362044227763882,
+      "loss": 0.7111573696136475,
+      "mean_token_accuracy": 0.798752411454916,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7664387949397055,
+      "eval_loss": 0.7365804314613342,
+      "eval_mean_token_accuracy": 0.7963101200843126,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 89.9623,
+      "eval_samples_per_second": 15.784,
+      "eval_steps_per_second": 1.979,
+      "step": 620
+    },
+    {
+      "entropy": 0.7424944408237935,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.2898184061050415,
+      "learning_rate": 0.000361495013613564,
+      "loss": 0.6913110733032226,
+      "mean_token_accuracy": 0.8009015507996082,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7255479529332579,
+      "eval_loss": 0.7345991730690002,
+      "eval_mean_token_accuracy": 0.8013592945056015,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 89.817,
+      "eval_samples_per_second": 15.81,
+      "eval_steps_per_second": 1.982,
+      "step": 640
+    },
+    {
+      "entropy": 0.7303367637097835,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.1310476064682007,
+      "learning_rate": 0.00036089510669270666,
+      "loss": 0.6875617027282714,
+      "mean_token_accuracy": 0.8035397171974182,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7101308248016271,
+      "eval_loss": 0.7394917011260986,
+      "eval_mean_token_accuracy": 0.7990792942850777,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 89.7431,
+      "eval_samples_per_second": 15.823,
+      "eval_steps_per_second": 1.983,
+      "step": 660
+    },
+    {
+      "entropy": 0.7387678384780884,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.9645389914512634,
+      "learning_rate": 0.0003602446767678725,
+      "loss": 0.6911204338073731,
+      "mean_token_accuracy": 0.8056333385407924,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7281073737010527,
+      "eval_loss": 0.7376708984375,
+      "eval_mean_token_accuracy": 0.7997356415464637,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 90.0338,
+      "eval_samples_per_second": 15.772,
+      "eval_steps_per_second": 1.977,
+      "step": 680
+    },
+    {
+      "entropy": 0.7449554048478604,
+      "epoch": 1.687990343995172,
+      "grad_norm": 1.9069629907608032,
+      "learning_rate": 0.0003595439079030364,
+      "loss": 0.7042286396026611,
+      "mean_token_accuracy": 0.8018639378249646,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7022633817088738,
+      "eval_loss": 0.7320939898490906,
+      "eval_mean_token_accuracy": 0.8006051541044471,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 89.8948,
+      "eval_samples_per_second": 15.796,
+      "eval_steps_per_second": 1.98,
+      "step": 700
+    },
+    {
+      "entropy": 0.723249051719904,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.8333877325057983,
+      "learning_rate": 0.00035879299840749777,
+      "loss": 0.7043714046478271,
+      "mean_token_accuracy": 0.8028345607221127,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.672436946898364,
+      "eval_loss": 0.713233470916748,
+      "eval_mean_token_accuracy": 0.8047760577684038,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 89.6394,
+      "eval_samples_per_second": 15.841,
+      "eval_steps_per_second": 1.986,
+      "step": 720
+    },
+    {
+      "entropy": 0.730014206469059,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.4580037593841553,
+      "learning_rate": 0.00035799216077976137,
+      "loss": 0.6985176563262939,
+      "mean_token_accuracy": 0.8015547141432762,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.6908873795123582,
+      "eval_loss": 0.7198202013969421,
+      "eval_mean_token_accuracy": 0.8024131682481659,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 89.3771,
+      "eval_samples_per_second": 15.888,
+      "eval_steps_per_second": 1.992,
+      "step": 740
+    },
+    {
+      "entropy": 0.7219605796039105,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.51682710647583,
+      "learning_rate": 0.000357141621647403,
+      "loss": 0.6844424724578857,
+      "mean_token_accuracy": 0.8032813094556331,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7113401109582922,
+      "eval_loss": 0.7195309400558472,
+      "eval_mean_token_accuracy": 0.8019187035185568,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 89.4658,
+      "eval_samples_per_second": 15.872,
+      "eval_steps_per_second": 1.99,
+      "step": 760
+    },
+    {
+      "entropy": 0.729826743900776,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 1.0838000774383545,
+      "learning_rate": 0.0003562416217029361,
+      "loss": 0.685858678817749,
+      "mean_token_accuracy": 0.8050567395985126,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.649370267484965,
+      "eval_loss": 0.7177029252052307,
+      "eval_mean_token_accuracy": 0.8050174043419656,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 89.2287,
+      "eval_samples_per_second": 15.914,
+      "eval_steps_per_second": 1.995,
+      "step": 780
+    },
+    {
+      "entropy": 0.6799364153295755,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 1.1364924907684326,
+      "learning_rate": 0.00035529241563569903,
+      "loss": 0.6714428901672364,
+      "mean_token_accuracy": 0.8087341658771038,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6603799047095052,
+      "eval_loss": 0.7201809287071228,
+      "eval_mean_token_accuracy": 0.802863868769635,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 89.2718,
+      "eval_samples_per_second": 15.906,
+      "eval_steps_per_second": 1.994,
+      "step": 800
+    },
+    {
+      "entropy": 0.7078610375523567,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 1.2191942930221558,
+      "learning_rate": 0.0003542942720597808,
+      "loss": 0.6996590614318847,
+      "mean_token_accuracy": 0.8049512051045895,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.7240545264120852,
+      "eval_loss": 0.7301309108734131,
+      "eval_mean_token_accuracy": 0.8015973166133581,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 89.3438,
+      "eval_samples_per_second": 15.894,
+      "eval_steps_per_second": 1.992,
+      "step": 820
+    },
+    {
+      "entropy": 0.6557391556826505,
+      "epoch": 2.024140012070006,
+      "grad_norm": 1.36203134059906,
+      "learning_rate": 0.0003532474734380064,
+      "loss": 0.6327113628387451,
+      "mean_token_accuracy": 0.8142829847026181,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6573808648613062,
+      "eval_loss": 0.7334365844726562,
+      "eval_mean_token_accuracy": 0.8043633893634496,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 89.3359,
+      "eval_samples_per_second": 15.895,
+      "eval_steps_per_second": 1.992,
+      "step": 840
+    },
+    {
+      "entropy": 0.6409128502011299,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 1.5804955959320068,
+      "learning_rate": 0.0003521523160020035,
+      "loss": 0.6023559093475341,
+      "mean_token_accuracy": 0.8222251623868942,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6983234788594621,
+      "eval_loss": 0.7232135534286499,
+      "eval_mean_token_accuracy": 0.8010654268639811,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 89.2615,
+      "eval_samples_per_second": 15.908,
+      "eval_steps_per_second": 1.994,
+      "step": 860
+    },
+    {
+      "entropy": 0.6151149723678827,
+      "epoch": 2.12070006035003,
+      "grad_norm": 1.1975436210632324,
+      "learning_rate": 0.00035100910966837193,
+      "loss": 0.5803564548492431,
+      "mean_token_accuracy": 0.8256325736641884,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6636555573243773,
+      "eval_loss": 0.7205662727355957,
+      "eval_mean_token_accuracy": 0.8036768369460374,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 89.2957,
+      "eval_samples_per_second": 15.902,
+      "eval_steps_per_second": 1.993,
+      "step": 880
+    },
+    {
+      "entropy": 0.6201727617532015,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9907131195068359,
+      "learning_rate": 0.0003498181779509813,
+      "loss": 0.6101872444152832,
+      "mean_token_accuracy": 0.8196512959897518,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.6473788086617931,
+      "eval_loss": 0.7062397003173828,
+      "eval_mean_token_accuracy": 0.8078362184963869,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 88.6198,
+      "eval_samples_per_second": 16.024,
+      "eval_steps_per_second": 2.009,
+      "step": 900
+    },
+    {
+      "entropy": 0.6097332935780286,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.9055633544921875,
+      "learning_rate": 0.00034857985786942026,
+      "loss": 0.5956094264984131,
+      "mean_token_accuracy": 0.8240575887262821,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.612629868341296,
+      "eval_loss": 0.7106770277023315,
+      "eval_mean_token_accuracy": 0.805324277181304,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 89.4383,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 920
+    },
+    {
+      "entropy": 0.6182701248675585,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.9513736367225647,
+      "learning_rate": 0.00034729449985362404,
+      "loss": 0.6019269466400147,
+      "mean_token_accuracy": 0.8187106013298034,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6437647456533453,
+      "eval_loss": 0.7173502445220947,
+      "eval_mean_token_accuracy": 0.8056439573175451,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.9568,
+      "eval_samples_per_second": 15.963,
+      "eval_steps_per_second": 2.001,
+      "step": 940
+    },
+    {
+      "entropy": 0.6508570514619351,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.9606207013130188,
+      "learning_rate": 0.0003459624676447067,
+      "loss": 0.6296409606933594,
+      "mean_token_accuracy": 0.8167732007801533,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6288247135248077,
+      "eval_loss": 0.7160675525665283,
+      "eval_mean_token_accuracy": 0.8066010689467527,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 89.0044,
+      "eval_samples_per_second": 15.954,
+      "eval_steps_per_second": 2.0,
+      "step": 960
+    },
+    {
+      "entropy": 0.6380081418901682,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.9677181839942932,
+      "learning_rate": 0.000344584138192027,
+      "loss": 0.6210546493530273,
+      "mean_token_accuracy": 0.8186135642230511,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6407805031604981,
+      "eval_loss": 0.7055043578147888,
+      "eval_mean_token_accuracy": 0.8070103157772107,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 89.5606,
+      "eval_samples_per_second": 15.855,
+      "eval_steps_per_second": 1.987,
+      "step": 980
+    },
+    {
+      "entropy": 0.6310251638293266,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 2.1046407222747803,
+      "learning_rate": 0.000343159901546516,
+      "loss": 0.6165830135345459,
+      "mean_token_accuracy": 0.8207855716347694,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6461507470420237,
+      "eval_loss": 0.7292136549949646,
+      "eval_mean_token_accuracy": 0.8030418261383356,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 89.315,
+      "eval_samples_per_second": 15.899,
+      "eval_steps_per_second": 1.993,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6256617344915867,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 1.1768349409103394,
+      "learning_rate": 0.0003416901607502972,
+      "loss": 0.6142709255218506,
+      "mean_token_accuracy": 0.8220669947564602,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6150523993406403,
+      "eval_loss": 0.7019563317298889,
+      "eval_mean_token_accuracy": 0.8094323099998946,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 89.419,
+      "eval_samples_per_second": 15.88,
+      "eval_steps_per_second": 1.991,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6126735735684633,
+      "epoch": 2.506940253470127,
+      "grad_norm": 1.8065496683120728,
+      "learning_rate": 0.00034017533172263055,
+      "loss": 0.6056458473205566,
+      "mean_token_accuracy": 0.8203762136399746,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6544345456562685,
+      "eval_loss": 0.7086517214775085,
+      "eval_mean_token_accuracy": 0.806709805901131,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 88.981,
+      "eval_samples_per_second": 15.958,
+      "eval_steps_per_second": 2.0,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6311754353344441,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 1.0576369762420654,
+      "learning_rate": 0.00033861584314221236,
+      "loss": 0.6176392555236816,
+      "mean_token_accuracy": 0.8177410490810871,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6273799200406235,
+      "eval_loss": 0.708163321018219,
+      "eval_mean_token_accuracy": 0.8032017099053672,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 89.7668,
+      "eval_samples_per_second": 15.819,
+      "eval_steps_per_second": 1.983,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6200249589979648,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.997065007686615,
+      "learning_rate": 0.0003370121363258637,
+      "loss": 0.6203888893127442,
+      "mean_token_accuracy": 0.8196275025606156,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6457576811983344,
+      "eval_loss": 0.6972290277481079,
+      "eval_mean_token_accuracy": 0.8095980849158898,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 89.2027,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 1080
+    },
+    {
+      "entropy": 0.626810473203659,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.865475058555603,
+      "learning_rate": 0.0003353646651036438,
+      "loss": 0.6102585792541504,
+      "mean_token_accuracy": 0.819708751142025,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.6412022431914726,
+      "eval_loss": 0.7036139965057373,
+      "eval_mean_token_accuracy": 0.8070317135098275,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 89.4733,
+      "eval_samples_per_second": 15.871,
+      "eval_steps_per_second": 1.989,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6581619590520859,
+      "epoch": 2.700060350030175,
+      "grad_norm": 1.00307035446167,
+      "learning_rate": 0.00033367389569042064,
+      "loss": 0.6397720336914062,
+      "mean_token_accuracy": 0.81413309648633,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6288736402318719,
+      "eval_loss": 0.6966825723648071,
+      "eval_mean_token_accuracy": 0.8094039669867312,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 88.7341,
+      "eval_samples_per_second": 16.003,
+      "eval_steps_per_second": 2.006,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6399412982165813,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.8589329719543457,
+      "learning_rate": 0.0003319403065539384,
+      "loss": 0.6347338199615479,
+      "mean_token_accuracy": 0.8139534957706929,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.683491913455256,
+      "eval_loss": 0.6984149217605591,
+      "eval_mean_token_accuracy": 0.8083989469522841,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 89.3929,
+      "eval_samples_per_second": 15.885,
+      "eval_steps_per_second": 1.991,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6264835461974144,
+      "epoch": 2.796620398310199,
+      "grad_norm": 0.7807853817939758,
+      "learning_rate": 0.0003301643882794163,
+      "loss": 0.6162629127502441,
+      "mean_token_accuracy": 0.8216292977333068,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6578792159476977,
+      "eval_loss": 0.6921441555023193,
+      "eval_mean_token_accuracy": 0.8094299177775223,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 89.2356,
+      "eval_samples_per_second": 15.913,
+      "eval_steps_per_second": 1.995,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6269129011780024,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 1.1675593852996826,
+      "learning_rate": 0.0003283466434307189,
+      "loss": 0.6235532760620117,
+      "mean_token_accuracy": 0.8188897788524627,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6533248447970058,
+      "eval_loss": 0.696849524974823,
+      "eval_mean_token_accuracy": 0.8102703713968898,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2602,
+      "eval_samples_per_second": 16.089,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.6259873781353236,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.9596486687660217,
+      "learning_rate": 0.00032648758640813655,
+      "loss": 0.6118728160858155,
+      "mean_token_accuracy": 0.8171859130263328,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6212078754821521,
+      "eval_loss": 0.6942870020866394,
+      "eval_mean_token_accuracy": 0.810484343365337,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 89.4375,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 1200
+    },
+    {
+      "entropy": 0.6142537571489811,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 0.9625058770179749,
+      "learning_rate": 0.00032458774330281615,
+      "loss": 0.6060261249542236,
+      "mean_token_accuracy": 0.8223113484680653,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6553665667437436,
+      "eval_loss": 0.6835269331932068,
+      "eval_mean_token_accuracy": 0.8134723968720168,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 89.5092,
+      "eval_samples_per_second": 15.864,
+      "eval_steps_per_second": 1.989,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6301972426474094,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.9657333493232727,
+      "learning_rate": 0.0003226476517478835,
+      "loss": 0.6277759075164795,
+      "mean_token_accuracy": 0.8164266042411328,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6594253118788258,
+      "eval_loss": 0.6820935606956482,
+      "eval_mean_token_accuracy": 0.8132368198941263,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 89.6211,
+      "eval_samples_per_second": 15.844,
+      "eval_steps_per_second": 1.986,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5346100655469027,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8541185259819031,
+      "learning_rate": 0.0003206678607662996,
+      "loss": 0.5261692047119141,
+      "mean_token_accuracy": 0.8382289425119177,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5626645662476507,
+      "eval_loss": 0.7028542160987854,
+      "eval_mean_token_accuracy": 0.8133395572056931,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 89.1267,
+      "eval_samples_per_second": 15.932,
+      "eval_steps_per_second": 1.997,
+      "step": 1260
+    },
+    {
+      "entropy": 0.5001101832836866,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.8208815455436707,
+      "learning_rate": 0.0003186489306154935,
+      "loss": 0.4850132942199707,
+      "mean_token_accuracy": 0.8468983843922615,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5889462046743779,
+      "eval_loss": 0.691015899181366,
+      "eval_mean_token_accuracy": 0.8135226466012805,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.9677,
+      "eval_samples_per_second": 15.961,
+      "eval_steps_per_second": 2.001,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5242766339331866,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 0.7893072962760925,
+      "learning_rate": 0.0003165914326288163,
+      "loss": 0.5061577320098877,
+      "mean_token_accuracy": 0.8425608821213245,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5543691962957382,
+      "eval_loss": 0.6913734078407288,
+      "eval_mean_token_accuracy": 0.8154889680026622,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 89.6963,
+      "eval_samples_per_second": 15.831,
+      "eval_steps_per_second": 1.984,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5327721010893584,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.8992080092430115,
+      "learning_rate": 0.0003144959490538604,
+      "loss": 0.5068036556243897,
+      "mean_token_accuracy": 0.8420542575418949,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5734540803378887,
+      "eval_loss": 0.6948716044425964,
+      "eval_mean_token_accuracy": 0.8137217944257715,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 89.4597,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5247377116233111,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.9291845560073853,
+      "learning_rate": 0.0003123630728876902,
+      "loss": 0.5023272037506104,
+      "mean_token_accuracy": 0.8435182586312294,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5492362935891312,
+      "eval_loss": 0.6980633735656738,
+      "eval_mean_token_accuracy": 0.8155082919624415,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 89.2124,
+      "eval_samples_per_second": 15.917,
+      "eval_steps_per_second": 1.995,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5380499072372913,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 1.0477524995803833,
+      "learning_rate": 0.00031019340770903136,
+      "loss": 0.5264591693878173,
+      "mean_token_accuracy": 0.8391959741711617,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5896503307511297,
+      "eval_loss": 0.6997144222259521,
+      "eval_mean_token_accuracy": 0.8124373062942805,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 89.1147,
+      "eval_samples_per_second": 15.935,
+      "eval_steps_per_second": 1.997,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5460653610527515,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 1.2473069429397583,
+      "learning_rate": 0.00030798756750746476,
+      "loss": 0.5319677829742432,
+      "mean_token_accuracy": 0.8364320226013661,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5659052182114526,
+      "eval_loss": 0.6924049258232117,
+      "eval_mean_token_accuracy": 0.8153053585732921,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 89.6468,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5367407951503992,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 1.2679712772369385,
+      "learning_rate": 0.00030574617650967485,
+      "loss": 0.5190927028656006,
+      "mean_token_accuracy": 0.8399469174444676,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5581759144081159,
+      "eval_loss": 0.6933317184448242,
+      "eval_mean_token_accuracy": 0.8139757862251796,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 89.0749,
+      "eval_samples_per_second": 15.942,
+      "eval_steps_per_second": 1.998,
+      "step": 1400
+    },
+    {
+      "entropy": 0.530736118927598,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.9717612862586975,
+      "learning_rate": 0.0003034698690028009,
+      "loss": 0.512363052368164,
+      "mean_token_accuracy": 0.8419127956032753,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5233335836549823,
+      "eval_loss": 0.6958550214767456,
+      "eval_mean_token_accuracy": 0.8158778501360604,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 88.8664,
+      "eval_samples_per_second": 15.979,
+      "eval_steps_per_second": 2.003,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5327631575986743,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7649208903312683,
+      "learning_rate": 0.00030115928915494116,
+      "loss": 0.5202179431915284,
+      "mean_token_accuracy": 0.8384683802723885,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.5960972657364406,
+      "eval_loss": 0.6984783411026001,
+      "eval_mean_token_accuracy": 0.8095930597085631,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 89.5168,
+      "eval_samples_per_second": 15.863,
+      "eval_steps_per_second": 1.988,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5428155034780502,
+      "epoch": 3.51901025950513,
+      "grad_norm": 1.2790874242782593,
+      "learning_rate": 0.00029881509083286116,
+      "loss": 0.524717378616333,
+      "mean_token_accuracy": 0.8346644505858422,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.550318562917495,
+      "eval_loss": 0.6988471746444702,
+      "eval_mean_token_accuracy": 0.8132969897784544,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 89.2598,
+      "eval_samples_per_second": 15.909,
+      "eval_steps_per_second": 1.994,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5489993002265692,
+      "epoch": 3.567290283645142,
+      "grad_norm": 1.055821418762207,
+      "learning_rate": 0.00029643793741695687,
+      "loss": 0.5421150207519532,
+      "mean_token_accuracy": 0.8345710933208466,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5816821718818685,
+      "eval_loss": 0.6853668689727783,
+      "eval_mean_token_accuracy": 0.8142332360985574,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 89.0424,
+      "eval_samples_per_second": 15.947,
+      "eval_steps_per_second": 1.999,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5451365381479263,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.9027532935142517,
+      "learning_rate": 0.00029402850161352583,
+      "loss": 0.5463609218597412,
+      "mean_token_accuracy": 0.8369908526539802,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5344233209832331,
+      "eval_loss": 0.6794531345367432,
+      "eval_mean_token_accuracy": 0.8177759824843889,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 89.123,
+      "eval_samples_per_second": 15.933,
+      "eval_steps_per_second": 1.997,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5521467242389917,
+      "epoch": 3.663850331925166,
+      "grad_norm": 1.4606311321258545,
+      "learning_rate": 0.0002915874652643997,
+      "loss": 0.5421902179718018,
+      "mean_token_accuracy": 0.8339455373585224,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5807709094513668,
+      "eval_loss": 0.6949691772460938,
+      "eval_mean_token_accuracy": 0.8121141439743256,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.5437,
+      "eval_samples_per_second": 16.037,
+      "eval_steps_per_second": 2.01,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5455610822886229,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7269843220710754,
+      "learning_rate": 0.0002891155191539905,
+      "loss": 0.5407682895660401,
+      "mean_token_accuracy": 0.8393648222088814,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.6032205542151847,
+      "eval_loss": 0.6790974140167236,
+      "eval_mean_token_accuracy": 0.8165463112043531,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 89.517,
+      "eval_samples_per_second": 15.863,
+      "eval_steps_per_second": 1.988,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5474163968116045,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.8148866295814514,
+      "learning_rate": 0.00028661336281380717,
+      "loss": 0.5394441604614257,
+      "mean_token_accuracy": 0.8336028568446636,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5797032238392348,
+      "eval_loss": 0.670870304107666,
+      "eval_mean_token_accuracy": 0.8182691593518417,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 89.4468,
+      "eval_samples_per_second": 15.875,
+      "eval_steps_per_second": 1.99,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5511750865727663,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.3441656827926636,
+      "learning_rate": 0.0002840817043244964,
+      "loss": 0.5378933906555176,
+      "mean_token_accuracy": 0.8349151819944381,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5297858750217417,
+      "eval_loss": 0.680833637714386,
+      "eval_mean_token_accuracy": 0.8180924275617921,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 89.5417,
+      "eval_samples_per_second": 15.859,
+      "eval_steps_per_second": 1.988,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5379613988101483,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9813932776451111,
+      "learning_rate": 0.00028152126011546396,
+      "loss": 0.5340342044830322,
+      "mean_token_accuracy": 0.8346010789275169,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5811898510777549,
+      "eval_loss": 0.6764267683029175,
+      "eval_mean_token_accuracy": 0.8180528236239144,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 88.7831,
+      "eval_samples_per_second": 15.994,
+      "eval_steps_per_second": 2.005,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5607876226305961,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.9617928266525269,
+      "learning_rate": 0.00027893275476213383,
+      "loss": 0.5434338569641113,
+      "mean_token_accuracy": 0.8332898400723934,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5474644318390428,
+      "eval_loss": 0.6740226745605469,
+      "eval_mean_token_accuracy": 0.8185851306058047,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 89.3246,
+      "eval_samples_per_second": 15.897,
+      "eval_steps_per_second": 1.993,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5354843523353339,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.5320990085601807,
+      "learning_rate": 0.0002763169207809021,
+      "loss": 0.5251852989196777,
+      "mean_token_accuracy": 0.8416872084140777,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.5436856410141742,
+      "eval_loss": 0.6675190925598145,
+      "eval_mean_token_accuracy": 0.8202396507343549,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 89.6328,
+      "eval_samples_per_second": 15.842,
+      "eval_steps_per_second": 1.986,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5418571938167919,
+      "epoch": 4.0,
+      "grad_norm": 4.690354347229004,
+      "learning_rate": 0.000273674498421843,
+      "loss": 0.548275089263916,
+      "mean_token_accuracy": 0.836557479647847,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5497228983747825,
+      "eval_loss": 0.6621812582015991,
+      "eval_mean_token_accuracy": 0.8205679802412398,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 89.6272,
+      "eval_samples_per_second": 15.843,
+      "eval_steps_per_second": 1.986,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4195931971073151,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6678048968315125,
+      "learning_rate": 0.0002710062354592273,
+      "loss": 0.3970795154571533,
+      "mean_token_accuracy": 0.8719995342195034,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.46495268753405367,
+      "eval_loss": 0.7061581611633301,
+      "eval_mean_token_accuracy": 0.8190107995204712,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 89.6387,
+      "eval_samples_per_second": 15.841,
+      "eval_steps_per_second": 1.986,
+      "step": 1680
+    },
+    {
+      "entropy": 0.4263830740004778,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.7026669383049011,
+      "learning_rate": 0.000268312886979911,
+      "loss": 0.398905086517334,
+      "mean_token_accuracy": 0.8691012300550938,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.45755320670229666,
+      "eval_loss": 0.7102291584014893,
+      "eval_mean_token_accuracy": 0.819212955370378,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 89.2777,
+      "eval_samples_per_second": 15.905,
+      "eval_steps_per_second": 1.994,
+      "step": 1700
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.7021515523994624e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1720/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1720/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1720/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1720/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1720/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..7ce4b072fcbe5b636867e13937c56fedee3b68b4
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1720/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.02728394274790723,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1720/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1720/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1720/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1720/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1720/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1720/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1720/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1720/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..0ff7bf87334d99140ff56806457cf9b57c4b81d2
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1720/trainer_state.json
@@ -0,0 +1,1840 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 4.1448400724200365,
+  "eval_steps": 20,
+  "global_step": 1720,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.024795526266098,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.945375919342041,
+      "learning_rate": 1.6698127428345936e-05,
+      "loss": 1.7825420379638672,
+      "mean_token_accuracy": 0.6315708436071873,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.29921497588747,
+      "eval_loss": 1.169131875038147,
+      "eval_mean_token_accuracy": 0.7185706342204233,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 91.7898,
+      "eval_samples_per_second": 15.47,
+      "eval_steps_per_second": 1.939,
+      "step": 20
+    },
+    {
+      "entropy": 1.0346344470977784,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.7916979789733887,
+      "learning_rate": 3.427510366871008e-05,
+      "loss": 0.9480395317077637,
+      "mean_token_accuracy": 0.7543429024517536,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 0.9673695430326997,
+      "eval_loss": 0.8588430285453796,
+      "eval_mean_token_accuracy": 0.7737233002534073,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 89.8219,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 40
+    },
+    {
+      "entropy": 0.8977566614747048,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7972187995910645,
+      "learning_rate": 5.185207990907422e-05,
+      "loss": 0.8029604911804199,
+      "mean_token_accuracy": 0.7818016350269318,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8436836982041263,
+      "eval_loss": 0.7893033623695374,
+      "eval_mean_token_accuracy": 0.7858914984076211,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 89.9349,
+      "eval_samples_per_second": 15.789,
+      "eval_steps_per_second": 1.979,
+      "step": 60
+    },
+    {
+      "entropy": 0.8650321319699288,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.2128729820251465,
+      "learning_rate": 6.942905614943836e-05,
+      "loss": 0.77417893409729,
+      "mean_token_accuracy": 0.7875479467213153,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8209857180547179,
+      "eval_loss": 0.7572280168533325,
+      "eval_mean_token_accuracy": 0.7925970383574453,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 89.9459,
+      "eval_samples_per_second": 15.787,
+      "eval_steps_per_second": 1.979,
+      "step": 80
+    },
+    {
+      "entropy": 0.8096660938113928,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.1254287958145142,
+      "learning_rate": 8.700603238980251e-05,
+      "loss": 0.7170331001281738,
+      "mean_token_accuracy": 0.7994555257260799,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.792780208788561,
+      "eval_loss": 0.7442984580993652,
+      "eval_mean_token_accuracy": 0.7954704573984896,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 90.294,
+      "eval_samples_per_second": 15.726,
+      "eval_steps_per_second": 1.971,
+      "step": 100
+    },
+    {
+      "entropy": 0.7979829199612141,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2140507698059082,
+      "learning_rate": 0.00010458300863016666,
+      "loss": 0.7141861915588379,
+      "mean_token_accuracy": 0.8000014744699001,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.7938692486018277,
+      "eval_loss": 0.7279797196388245,
+      "eval_mean_token_accuracy": 0.7989798734027348,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 89.7189,
+      "eval_samples_per_second": 15.827,
+      "eval_steps_per_second": 1.984,
+      "step": 120
+    },
+    {
+      "entropy": 0.7916347607970238,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.0506458282470703,
+      "learning_rate": 0.0001221599848705308,
+      "loss": 0.7117055416107178,
+      "mean_token_accuracy": 0.7970046654343605,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.8052781063519167,
+      "eval_loss": 0.722576916217804,
+      "eval_mean_token_accuracy": 0.7982530523551984,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 90.0376,
+      "eval_samples_per_second": 15.771,
+      "eval_steps_per_second": 1.977,
+      "step": 140
+    },
+    {
+      "entropy": 0.7880666613578796,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8409207463264465,
+      "learning_rate": 0.00013973696111089492,
+      "loss": 0.7052523136138916,
+      "mean_token_accuracy": 0.8000123649835587,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7787443134891853,
+      "eval_loss": 0.7137772440910339,
+      "eval_mean_token_accuracy": 0.8014209287220173,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 90.2562,
+      "eval_samples_per_second": 15.733,
+      "eval_steps_per_second": 1.972,
+      "step": 160
+    },
+    {
+      "entropy": 0.7696007996797561,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.0182477235794067,
+      "learning_rate": 0.00015731393735125907,
+      "loss": 0.7008797645568847,
+      "mean_token_accuracy": 0.8007259473204613,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.752259682068664,
+      "eval_loss": 0.7171286344528198,
+      "eval_mean_token_accuracy": 0.80052122574174,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 90.0797,
+      "eval_samples_per_second": 15.764,
+      "eval_steps_per_second": 1.976,
+      "step": 180
+    },
+    {
+      "entropy": 0.7828126326203346,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9553632736206055,
+      "learning_rate": 0.0001748909135916232,
+      "loss": 0.7046853542327881,
+      "mean_token_accuracy": 0.7988562889397144,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7389638169427936,
+      "eval_loss": 0.7187935709953308,
+      "eval_mean_token_accuracy": 0.7995899999409579,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 89.9223,
+      "eval_samples_per_second": 15.791,
+      "eval_steps_per_second": 1.979,
+      "step": 200
+    },
+    {
+      "entropy": 0.7909286297857762,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 1.134459376335144,
+      "learning_rate": 0.00019246788983198735,
+      "loss": 0.6989312171936035,
+      "mean_token_accuracy": 0.7991565138101577,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7496965199374082,
+      "eval_loss": 0.7106321454048157,
+      "eval_mean_token_accuracy": 0.8046183442131857,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 89.6899,
+      "eval_samples_per_second": 15.832,
+      "eval_steps_per_second": 1.985,
+      "step": 220
+    },
+    {
+      "entropy": 0.7651963323354721,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.9557585716247559,
+      "learning_rate": 0.0002100448660723515,
+      "loss": 0.6984559059143066,
+      "mean_token_accuracy": 0.8019823372364044,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7615563079212488,
+      "eval_loss": 0.7098539471626282,
+      "eval_mean_token_accuracy": 0.8047233508544022,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 90.3752,
+      "eval_samples_per_second": 15.712,
+      "eval_steps_per_second": 1.97,
+      "step": 240
+    },
+    {
+      "entropy": 0.787507726252079,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.9279311299324036,
+      "learning_rate": 0.00022762184231271568,
+      "loss": 0.708138370513916,
+      "mean_token_accuracy": 0.8013308703899383,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7856640896100676,
+      "eval_loss": 0.7191570401191711,
+      "eval_mean_token_accuracy": 0.8024677249153008,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 93.7336,
+      "eval_samples_per_second": 15.149,
+      "eval_steps_per_second": 1.899,
+      "step": 260
+    },
+    {
+      "entropy": 0.768489234894514,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.7992776036262512,
+      "learning_rate": 0.0002451988185530798,
+      "loss": 0.7035849571228028,
+      "mean_token_accuracy": 0.8031105332076549,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7682498740346244,
+      "eval_loss": 0.7149022221565247,
+      "eval_mean_token_accuracy": 0.7973497995499814,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 89.7309,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 280
+    },
+    {
+      "entropy": 0.7720955617725849,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.1931514739990234,
+      "learning_rate": 0.0002627757947934439,
+      "loss": 0.6965402603149414,
+      "mean_token_accuracy": 0.8040348663926125,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7214708304807042,
+      "eval_loss": 0.7350823879241943,
+      "eval_mean_token_accuracy": 0.7942911925610532,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 89.6451,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 300
+    },
+    {
+      "entropy": 0.7930095426738262,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8994467258453369,
+      "learning_rate": 0.0002803527710338081,
+      "loss": 0.7331492900848389,
+      "mean_token_accuracy": 0.796185302734375,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.8142535488926963,
+      "eval_loss": 0.7237834334373474,
+      "eval_mean_token_accuracy": 0.799495273761535,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 90.6759,
+      "eval_samples_per_second": 15.66,
+      "eval_steps_per_second": 1.963,
+      "step": 320
+    },
+    {
+      "entropy": 0.7825102441012859,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 1.8224239349365234,
+      "learning_rate": 0.00029792974727417223,
+      "loss": 0.7331175804138184,
+      "mean_token_accuracy": 0.7977667152881622,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7327135443017724,
+      "eval_loss": 0.7388784289360046,
+      "eval_mean_token_accuracy": 0.8007041252730938,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 89.7692,
+      "eval_samples_per_second": 15.818,
+      "eval_steps_per_second": 1.983,
+      "step": 340
+    },
+    {
+      "entropy": 0.819331557303667,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 1.0407779216766357,
+      "learning_rate": 0.0003155067235145364,
+      "loss": 0.7456695556640625,
+      "mean_token_accuracy": 0.7936980701982975,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.8065585420372781,
+      "eval_loss": 0.735519528388977,
+      "eval_mean_token_accuracy": 0.7974206296245704,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 89.4336,
+      "eval_samples_per_second": 15.878,
+      "eval_steps_per_second": 1.99,
+      "step": 360
+    },
+    {
+      "entropy": 0.8114653021097183,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.1209490299224854,
+      "learning_rate": 0.0003330836997549005,
+      "loss": 0.7340899467468261,
+      "mean_token_accuracy": 0.7914272703230381,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.8025527285056168,
+      "eval_loss": 0.7295576930046082,
+      "eval_mean_token_accuracy": 0.7990576628218876,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 89.9579,
+      "eval_samples_per_second": 15.785,
+      "eval_steps_per_second": 1.979,
+      "step": 380
+    },
+    {
+      "entropy": 0.8042497783899307,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 1.2437798976898193,
+      "learning_rate": 0.0003506606759952647,
+      "loss": 0.7381744384765625,
+      "mean_token_accuracy": 0.7967362694442273,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7717011336530193,
+      "eval_loss": 0.7436173558235168,
+      "eval_mean_token_accuracy": 0.7975849482450592,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 89.8411,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 400
+    },
+    {
+      "entropy": 0.7987750010056929,
+      "epoch": 1.012070006035003,
+      "grad_norm": 1.2390918731689453,
+      "learning_rate": 0.000364721224843344,
+      "loss": 0.7317790031433106,
+      "mean_token_accuracy": 0.7986087551364651,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7333170604170038,
+      "eval_loss": 0.7456948757171631,
+      "eval_mean_token_accuracy": 0.7964897577682238,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 89.7288,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 420
+    },
+    {
+      "entropy": 0.7372976846992969,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.1552740335464478,
+      "learning_rate": 0.00036468510102269575,
+      "loss": 0.6734249114990234,
+      "mean_token_accuracy": 0.805834824591875,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.7130355526891987,
+      "eval_loss": 0.7555333375930786,
+      "eval_mean_token_accuracy": 0.7961652781186479,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 89.8713,
+      "eval_samples_per_second": 15.8,
+      "eval_steps_per_second": 1.981,
+      "step": 440
+    },
+    {
+      "entropy": 0.7465668715536594,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 1.1991469860076904,
+      "learning_rate": 0.0003645973816745066,
+      "loss": 0.6871879577636719,
+      "mean_token_accuracy": 0.8042450234293937,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7772979542110743,
+      "eval_loss": 0.7561826705932617,
+      "eval_mean_token_accuracy": 0.7953876427720102,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 89.8395,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 460
+    },
+    {
+      "entropy": 0.7477855801582336,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 1.325486660003662,
+      "learning_rate": 0.00036445809162231435,
+      "loss": 0.6836830139160156,
+      "mean_token_accuracy": 0.8021619468927383,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.7098110676481483,
+      "eval_loss": 0.758007287979126,
+      "eval_mean_token_accuracy": 0.7909953189030122,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 89.6101,
+      "eval_samples_per_second": 15.846,
+      "eval_steps_per_second": 1.986,
+      "step": 480
+    },
+    {
+      "entropy": 0.7632691070437432,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 1.0122262239456177,
+      "learning_rate": 0.0003642672702835562,
+      "loss": 0.7071213245391845,
+      "mean_token_accuracy": 0.8000680930912495,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7645382720432924,
+      "eval_loss": 0.7492606043815613,
+      "eval_mean_token_accuracy": 0.7976858676149604,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 89.6092,
+      "eval_samples_per_second": 15.847,
+      "eval_steps_per_second": 1.986,
+      "step": 500
+    },
+    {
+      "entropy": 0.7623726457357407,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.367774486541748,
+      "learning_rate": 0.00036402497165841384,
+      "loss": 0.6965175151824952,
+      "mean_token_accuracy": 0.8016501650214195,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7868083715438843,
+      "eval_loss": 0.7481449842453003,
+      "eval_mean_token_accuracy": 0.799330582779445,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 89.8241,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 520
+    },
+    {
+      "entropy": 0.7535207405686378,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 1.213614821434021,
+      "learning_rate": 0.00036373126431453207,
+      "loss": 0.7078310489654541,
+      "mean_token_accuracy": 0.7998907208442688,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7829091950748743,
+      "eval_loss": 0.7416072487831116,
+      "eval_mean_token_accuracy": 0.7969146250339036,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 89.4607,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 540
+    },
+    {
+      "entropy": 0.7527363449335098,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 1.0929982662200928,
+      "learning_rate": 0.00036338623136761495,
+      "loss": 0.7058369636535644,
+      "mean_token_accuracy": 0.7997087433934211,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.728419938114252,
+      "eval_loss": 0.744873046875,
+      "eval_mean_token_accuracy": 0.7966634296299366,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 89.6597,
+      "eval_samples_per_second": 15.838,
+      "eval_steps_per_second": 1.985,
+      "step": 560
+    },
+    {
+      "entropy": 0.7406099259853363,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.409364104270935,
+      "learning_rate": 0.00036298997045790515,
+      "loss": 0.6902852058410645,
+      "mean_token_accuracy": 0.805386807769537,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.6991632758231645,
+      "eval_loss": 0.7480319738388062,
+      "eval_mean_token_accuracy": 0.7970068006033308,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 89.8329,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 580
+    },
+    {
+      "entropy": 0.7459449704736472,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.7053111791610718,
+      "learning_rate": 0.00036254259372255277,
+      "loss": 0.7044640064239502,
+      "mean_token_accuracy": 0.8005018755793571,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7454566660891758,
+      "eval_loss": 0.7551484704017639,
+      "eval_mean_token_accuracy": 0.7897093731365846,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 89.8313,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 600
+    },
+    {
+      "entropy": 0.7621391348540782,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 1.174443244934082,
+      "learning_rate": 0.000362044227763882,
+      "loss": 0.7111573696136475,
+      "mean_token_accuracy": 0.798752411454916,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7664387949397055,
+      "eval_loss": 0.7365804314613342,
+      "eval_mean_token_accuracy": 0.7963101200843126,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 89.9623,
+      "eval_samples_per_second": 15.784,
+      "eval_steps_per_second": 1.979,
+      "step": 620
+    },
+    {
+      "entropy": 0.7424944408237935,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.2898184061050415,
+      "learning_rate": 0.000361495013613564,
+      "loss": 0.6913110733032226,
+      "mean_token_accuracy": 0.8009015507996082,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7255479529332579,
+      "eval_loss": 0.7345991730690002,
+      "eval_mean_token_accuracy": 0.8013592945056015,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 89.817,
+      "eval_samples_per_second": 15.81,
+      "eval_steps_per_second": 1.982,
+      "step": 640
+    },
+    {
+      "entropy": 0.7303367637097835,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.1310476064682007,
+      "learning_rate": 0.00036089510669270666,
+      "loss": 0.6875617027282714,
+      "mean_token_accuracy": 0.8035397171974182,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7101308248016271,
+      "eval_loss": 0.7394917011260986,
+      "eval_mean_token_accuracy": 0.7990792942850777,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 89.7431,
+      "eval_samples_per_second": 15.823,
+      "eval_steps_per_second": 1.983,
+      "step": 660
+    },
+    {
+      "entropy": 0.7387678384780884,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.9645389914512634,
+      "learning_rate": 0.0003602446767678725,
+      "loss": 0.6911204338073731,
+      "mean_token_accuracy": 0.8056333385407924,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7281073737010527,
+      "eval_loss": 0.7376708984375,
+      "eval_mean_token_accuracy": 0.7997356415464637,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 90.0338,
+      "eval_samples_per_second": 15.772,
+      "eval_steps_per_second": 1.977,
+      "step": 680
+    },
+    {
+      "entropy": 0.7449554048478604,
+      "epoch": 1.687990343995172,
+      "grad_norm": 1.9069629907608032,
+      "learning_rate": 0.0003595439079030364,
+      "loss": 0.7042286396026611,
+      "mean_token_accuracy": 0.8018639378249646,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7022633817088738,
+      "eval_loss": 0.7320939898490906,
+      "eval_mean_token_accuracy": 0.8006051541044471,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 89.8948,
+      "eval_samples_per_second": 15.796,
+      "eval_steps_per_second": 1.98,
+      "step": 700
+    },
+    {
+      "entropy": 0.723249051719904,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.8333877325057983,
+      "learning_rate": 0.00035879299840749777,
+      "loss": 0.7043714046478271,
+      "mean_token_accuracy": 0.8028345607221127,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.672436946898364,
+      "eval_loss": 0.713233470916748,
+      "eval_mean_token_accuracy": 0.8047760577684038,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 89.6394,
+      "eval_samples_per_second": 15.841,
+      "eval_steps_per_second": 1.986,
+      "step": 720
+    },
+    {
+      "entropy": 0.730014206469059,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.4580037593841553,
+      "learning_rate": 0.00035799216077976137,
+      "loss": 0.6985176563262939,
+      "mean_token_accuracy": 0.8015547141432762,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.6908873795123582,
+      "eval_loss": 0.7198202013969421,
+      "eval_mean_token_accuracy": 0.8024131682481659,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 89.3771,
+      "eval_samples_per_second": 15.888,
+      "eval_steps_per_second": 1.992,
+      "step": 740
+    },
+    {
+      "entropy": 0.7219605796039105,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.51682710647583,
+      "learning_rate": 0.000357141621647403,
+      "loss": 0.6844424724578857,
+      "mean_token_accuracy": 0.8032813094556331,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7113401109582922,
+      "eval_loss": 0.7195309400558472,
+      "eval_mean_token_accuracy": 0.8019187035185568,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 89.4658,
+      "eval_samples_per_second": 15.872,
+      "eval_steps_per_second": 1.99,
+      "step": 760
+    },
+    {
+      "entropy": 0.729826743900776,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 1.0838000774383545,
+      "learning_rate": 0.0003562416217029361,
+      "loss": 0.685858678817749,
+      "mean_token_accuracy": 0.8050567395985126,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.649370267484965,
+      "eval_loss": 0.7177029252052307,
+      "eval_mean_token_accuracy": 0.8050174043419656,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 89.2287,
+      "eval_samples_per_second": 15.914,
+      "eval_steps_per_second": 1.995,
+      "step": 780
+    },
+    {
+      "entropy": 0.6799364153295755,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 1.1364924907684326,
+      "learning_rate": 0.00035529241563569903,
+      "loss": 0.6714428901672364,
+      "mean_token_accuracy": 0.8087341658771038,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6603799047095052,
+      "eval_loss": 0.7201809287071228,
+      "eval_mean_token_accuracy": 0.802863868769635,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 89.2718,
+      "eval_samples_per_second": 15.906,
+      "eval_steps_per_second": 1.994,
+      "step": 800
+    },
+    {
+      "entropy": 0.7078610375523567,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 1.2191942930221558,
+      "learning_rate": 0.0003542942720597808,
+      "loss": 0.6996590614318847,
+      "mean_token_accuracy": 0.8049512051045895,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.7240545264120852,
+      "eval_loss": 0.7301309108734131,
+      "eval_mean_token_accuracy": 0.8015973166133581,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 89.3438,
+      "eval_samples_per_second": 15.894,
+      "eval_steps_per_second": 1.992,
+      "step": 820
+    },
+    {
+      "entropy": 0.6557391556826505,
+      "epoch": 2.024140012070006,
+      "grad_norm": 1.36203134059906,
+      "learning_rate": 0.0003532474734380064,
+      "loss": 0.6327113628387451,
+      "mean_token_accuracy": 0.8142829847026181,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6573808648613062,
+      "eval_loss": 0.7334365844726562,
+      "eval_mean_token_accuracy": 0.8043633893634496,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 89.3359,
+      "eval_samples_per_second": 15.895,
+      "eval_steps_per_second": 1.992,
+      "step": 840
+    },
+    {
+      "entropy": 0.6409128502011299,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 1.5804955959320068,
+      "learning_rate": 0.0003521523160020035,
+      "loss": 0.6023559093475341,
+      "mean_token_accuracy": 0.8222251623868942,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6983234788594621,
+      "eval_loss": 0.7232135534286499,
+      "eval_mean_token_accuracy": 0.8010654268639811,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 89.2615,
+      "eval_samples_per_second": 15.908,
+      "eval_steps_per_second": 1.994,
+      "step": 860
+    },
+    {
+      "entropy": 0.6151149723678827,
+      "epoch": 2.12070006035003,
+      "grad_norm": 1.1975436210632324,
+      "learning_rate": 0.00035100910966837193,
+      "loss": 0.5803564548492431,
+      "mean_token_accuracy": 0.8256325736641884,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6636555573243773,
+      "eval_loss": 0.7205662727355957,
+      "eval_mean_token_accuracy": 0.8036768369460374,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 89.2957,
+      "eval_samples_per_second": 15.902,
+      "eval_steps_per_second": 1.993,
+      "step": 880
+    },
+    {
+      "entropy": 0.6201727617532015,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9907131195068359,
+      "learning_rate": 0.0003498181779509813,
+      "loss": 0.6101872444152832,
+      "mean_token_accuracy": 0.8196512959897518,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.6473788086617931,
+      "eval_loss": 0.7062397003173828,
+      "eval_mean_token_accuracy": 0.8078362184963869,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 88.6198,
+      "eval_samples_per_second": 16.024,
+      "eval_steps_per_second": 2.009,
+      "step": 900
+    },
+    {
+      "entropy": 0.6097332935780286,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.9055633544921875,
+      "learning_rate": 0.00034857985786942026,
+      "loss": 0.5956094264984131,
+      "mean_token_accuracy": 0.8240575887262821,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.612629868341296,
+      "eval_loss": 0.7106770277023315,
+      "eval_mean_token_accuracy": 0.805324277181304,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 89.4383,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 920
+    },
+    {
+      "entropy": 0.6182701248675585,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.9513736367225647,
+      "learning_rate": 0.00034729449985362404,
+      "loss": 0.6019269466400147,
+      "mean_token_accuracy": 0.8187106013298034,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6437647456533453,
+      "eval_loss": 0.7173502445220947,
+      "eval_mean_token_accuracy": 0.8056439573175451,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.9568,
+      "eval_samples_per_second": 15.963,
+      "eval_steps_per_second": 2.001,
+      "step": 940
+    },
+    {
+      "entropy": 0.6508570514619351,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.9606207013130188,
+      "learning_rate": 0.0003459624676447067,
+      "loss": 0.6296409606933594,
+      "mean_token_accuracy": 0.8167732007801533,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6288247135248077,
+      "eval_loss": 0.7160675525665283,
+      "eval_mean_token_accuracy": 0.8066010689467527,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 89.0044,
+      "eval_samples_per_second": 15.954,
+      "eval_steps_per_second": 2.0,
+      "step": 960
+    },
+    {
+      "entropy": 0.6380081418901682,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.9677181839942932,
+      "learning_rate": 0.000344584138192027,
+      "loss": 0.6210546493530273,
+      "mean_token_accuracy": 0.8186135642230511,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6407805031604981,
+      "eval_loss": 0.7055043578147888,
+      "eval_mean_token_accuracy": 0.8070103157772107,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 89.5606,
+      "eval_samples_per_second": 15.855,
+      "eval_steps_per_second": 1.987,
+      "step": 980
+    },
+    {
+      "entropy": 0.6310251638293266,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 2.1046407222747803,
+      "learning_rate": 0.000343159901546516,
+      "loss": 0.6165830135345459,
+      "mean_token_accuracy": 0.8207855716347694,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6461507470420237,
+      "eval_loss": 0.7292136549949646,
+      "eval_mean_token_accuracy": 0.8030418261383356,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 89.315,
+      "eval_samples_per_second": 15.899,
+      "eval_steps_per_second": 1.993,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6256617344915867,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 1.1768349409103394,
+      "learning_rate": 0.0003416901607502972,
+      "loss": 0.6142709255218506,
+      "mean_token_accuracy": 0.8220669947564602,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6150523993406403,
+      "eval_loss": 0.7019563317298889,
+      "eval_mean_token_accuracy": 0.8094323099998946,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 89.419,
+      "eval_samples_per_second": 15.88,
+      "eval_steps_per_second": 1.991,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6126735735684633,
+      "epoch": 2.506940253470127,
+      "grad_norm": 1.8065496683120728,
+      "learning_rate": 0.00034017533172263055,
+      "loss": 0.6056458473205566,
+      "mean_token_accuracy": 0.8203762136399746,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6544345456562685,
+      "eval_loss": 0.7086517214775085,
+      "eval_mean_token_accuracy": 0.806709805901131,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 88.981,
+      "eval_samples_per_second": 15.958,
+      "eval_steps_per_second": 2.0,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6311754353344441,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 1.0576369762420654,
+      "learning_rate": 0.00033861584314221236,
+      "loss": 0.6176392555236816,
+      "mean_token_accuracy": 0.8177410490810871,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6273799200406235,
+      "eval_loss": 0.708163321018219,
+      "eval_mean_token_accuracy": 0.8032017099053672,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 89.7668,
+      "eval_samples_per_second": 15.819,
+      "eval_steps_per_second": 1.983,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6200249589979648,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.997065007686615,
+      "learning_rate": 0.0003370121363258637,
+      "loss": 0.6203888893127442,
+      "mean_token_accuracy": 0.8196275025606156,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6457576811983344,
+      "eval_loss": 0.6972290277481079,
+      "eval_mean_token_accuracy": 0.8095980849158898,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 89.2027,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 1080
+    },
+    {
+      "entropy": 0.626810473203659,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.865475058555603,
+      "learning_rate": 0.0003353646651036438,
+      "loss": 0.6102585792541504,
+      "mean_token_accuracy": 0.819708751142025,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.6412022431914726,
+      "eval_loss": 0.7036139965057373,
+      "eval_mean_token_accuracy": 0.8070317135098275,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 89.4733,
+      "eval_samples_per_second": 15.871,
+      "eval_steps_per_second": 1.989,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6581619590520859,
+      "epoch": 2.700060350030175,
+      "grad_norm": 1.00307035446167,
+      "learning_rate": 0.00033367389569042064,
+      "loss": 0.6397720336914062,
+      "mean_token_accuracy": 0.81413309648633,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6288736402318719,
+      "eval_loss": 0.6966825723648071,
+      "eval_mean_token_accuracy": 0.8094039669867312,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 88.7341,
+      "eval_samples_per_second": 16.003,
+      "eval_steps_per_second": 2.006,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6399412982165813,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.8589329719543457,
+      "learning_rate": 0.0003319403065539384,
+      "loss": 0.6347338199615479,
+      "mean_token_accuracy": 0.8139534957706929,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.683491913455256,
+      "eval_loss": 0.6984149217605591,
+      "eval_mean_token_accuracy": 0.8083989469522841,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 89.3929,
+      "eval_samples_per_second": 15.885,
+      "eval_steps_per_second": 1.991,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6264835461974144,
+      "epoch": 2.796620398310199,
+      "grad_norm": 0.7807853817939758,
+      "learning_rate": 0.0003301643882794163,
+      "loss": 0.6162629127502441,
+      "mean_token_accuracy": 0.8216292977333068,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6578792159476977,
+      "eval_loss": 0.6921441555023193,
+      "eval_mean_token_accuracy": 0.8094299177775223,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 89.2356,
+      "eval_samples_per_second": 15.913,
+      "eval_steps_per_second": 1.995,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6269129011780024,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 1.1675593852996826,
+      "learning_rate": 0.0003283466434307189,
+      "loss": 0.6235532760620117,
+      "mean_token_accuracy": 0.8188897788524627,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6533248447970058,
+      "eval_loss": 0.696849524974823,
+      "eval_mean_token_accuracy": 0.8102703713968898,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2602,
+      "eval_samples_per_second": 16.089,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.6259873781353236,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.9596486687660217,
+      "learning_rate": 0.00032648758640813655,
+      "loss": 0.6118728160858155,
+      "mean_token_accuracy": 0.8171859130263328,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6212078754821521,
+      "eval_loss": 0.6942870020866394,
+      "eval_mean_token_accuracy": 0.810484343365337,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 89.4375,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 1200
+    },
+    {
+      "entropy": 0.6142537571489811,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 0.9625058770179749,
+      "learning_rate": 0.00032458774330281615,
+      "loss": 0.6060261249542236,
+      "mean_token_accuracy": 0.8223113484680653,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6553665667437436,
+      "eval_loss": 0.6835269331932068,
+      "eval_mean_token_accuracy": 0.8134723968720168,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 89.5092,
+      "eval_samples_per_second": 15.864,
+      "eval_steps_per_second": 1.989,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6301972426474094,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.9657333493232727,
+      "learning_rate": 0.0003226476517478835,
+      "loss": 0.6277759075164795,
+      "mean_token_accuracy": 0.8164266042411328,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6594253118788258,
+      "eval_loss": 0.6820935606956482,
+      "eval_mean_token_accuracy": 0.8132368198941263,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 89.6211,
+      "eval_samples_per_second": 15.844,
+      "eval_steps_per_second": 1.986,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5346100655469027,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8541185259819031,
+      "learning_rate": 0.0003206678607662996,
+      "loss": 0.5261692047119141,
+      "mean_token_accuracy": 0.8382289425119177,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5626645662476507,
+      "eval_loss": 0.7028542160987854,
+      "eval_mean_token_accuracy": 0.8133395572056931,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 89.1267,
+      "eval_samples_per_second": 15.932,
+      "eval_steps_per_second": 1.997,
+      "step": 1260
+    },
+    {
+      "entropy": 0.5001101832836866,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.8208815455436707,
+      "learning_rate": 0.0003186489306154935,
+      "loss": 0.4850132942199707,
+      "mean_token_accuracy": 0.8468983843922615,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5889462046743779,
+      "eval_loss": 0.691015899181366,
+      "eval_mean_token_accuracy": 0.8135226466012805,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.9677,
+      "eval_samples_per_second": 15.961,
+      "eval_steps_per_second": 2.001,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5242766339331866,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 0.7893072962760925,
+      "learning_rate": 0.0003165914326288163,
+      "loss": 0.5061577320098877,
+      "mean_token_accuracy": 0.8425608821213245,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5543691962957382,
+      "eval_loss": 0.6913734078407288,
+      "eval_mean_token_accuracy": 0.8154889680026622,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 89.6963,
+      "eval_samples_per_second": 15.831,
+      "eval_steps_per_second": 1.984,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5327721010893584,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.8992080092430115,
+      "learning_rate": 0.0003144959490538604,
+      "loss": 0.5068036556243897,
+      "mean_token_accuracy": 0.8420542575418949,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5734540803378887,
+      "eval_loss": 0.6948716044425964,
+      "eval_mean_token_accuracy": 0.8137217944257715,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 89.4597,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5247377116233111,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.9291845560073853,
+      "learning_rate": 0.0003123630728876902,
+      "loss": 0.5023272037506104,
+      "mean_token_accuracy": 0.8435182586312294,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5492362935891312,
+      "eval_loss": 0.6980633735656738,
+      "eval_mean_token_accuracy": 0.8155082919624415,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 89.2124,
+      "eval_samples_per_second": 15.917,
+      "eval_steps_per_second": 1.995,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5380499072372913,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 1.0477524995803833,
+      "learning_rate": 0.00031019340770903136,
+      "loss": 0.5264591693878173,
+      "mean_token_accuracy": 0.8391959741711617,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5896503307511297,
+      "eval_loss": 0.6997144222259521,
+      "eval_mean_token_accuracy": 0.8124373062942805,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 89.1147,
+      "eval_samples_per_second": 15.935,
+      "eval_steps_per_second": 1.997,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5460653610527515,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 1.2473069429397583,
+      "learning_rate": 0.00030798756750746476,
+      "loss": 0.5319677829742432,
+      "mean_token_accuracy": 0.8364320226013661,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5659052182114526,
+      "eval_loss": 0.6924049258232117,
+      "eval_mean_token_accuracy": 0.8153053585732921,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 89.6468,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5367407951503992,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 1.2679712772369385,
+      "learning_rate": 0.00030574617650967485,
+      "loss": 0.5190927028656006,
+      "mean_token_accuracy": 0.8399469174444676,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5581759144081159,
+      "eval_loss": 0.6933317184448242,
+      "eval_mean_token_accuracy": 0.8139757862251796,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 89.0749,
+      "eval_samples_per_second": 15.942,
+      "eval_steps_per_second": 1.998,
+      "step": 1400
+    },
+    {
+      "entropy": 0.530736118927598,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.9717612862586975,
+      "learning_rate": 0.0003034698690028009,
+      "loss": 0.512363052368164,
+      "mean_token_accuracy": 0.8419127956032753,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5233335836549823,
+      "eval_loss": 0.6958550214767456,
+      "eval_mean_token_accuracy": 0.8158778501360604,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 88.8664,
+      "eval_samples_per_second": 15.979,
+      "eval_steps_per_second": 2.003,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5327631575986743,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7649208903312683,
+      "learning_rate": 0.00030115928915494116,
+      "loss": 0.5202179431915284,
+      "mean_token_accuracy": 0.8384683802723885,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.5960972657364406,
+      "eval_loss": 0.6984783411026001,
+      "eval_mean_token_accuracy": 0.8095930597085631,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 89.5168,
+      "eval_samples_per_second": 15.863,
+      "eval_steps_per_second": 1.988,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5428155034780502,
+      "epoch": 3.51901025950513,
+      "grad_norm": 1.2790874242782593,
+      "learning_rate": 0.00029881509083286116,
+      "loss": 0.524717378616333,
+      "mean_token_accuracy": 0.8346644505858422,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.550318562917495,
+      "eval_loss": 0.6988471746444702,
+      "eval_mean_token_accuracy": 0.8132969897784544,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 89.2598,
+      "eval_samples_per_second": 15.909,
+      "eval_steps_per_second": 1.994,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5489993002265692,
+      "epoch": 3.567290283645142,
+      "grad_norm": 1.055821418762207,
+      "learning_rate": 0.00029643793741695687,
+      "loss": 0.5421150207519532,
+      "mean_token_accuracy": 0.8345710933208466,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5816821718818685,
+      "eval_loss": 0.6853668689727783,
+      "eval_mean_token_accuracy": 0.8142332360985574,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 89.0424,
+      "eval_samples_per_second": 15.947,
+      "eval_steps_per_second": 1.999,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5451365381479263,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.9027532935142517,
+      "learning_rate": 0.00029402850161352583,
+      "loss": 0.5463609218597412,
+      "mean_token_accuracy": 0.8369908526539802,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5344233209832331,
+      "eval_loss": 0.6794531345367432,
+      "eval_mean_token_accuracy": 0.8177759824843889,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 89.123,
+      "eval_samples_per_second": 15.933,
+      "eval_steps_per_second": 1.997,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5521467242389917,
+      "epoch": 3.663850331925166,
+      "grad_norm": 1.4606311321258545,
+      "learning_rate": 0.0002915874652643997,
+      "loss": 0.5421902179718018,
+      "mean_token_accuracy": 0.8339455373585224,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5807709094513668,
+      "eval_loss": 0.6949691772460938,
+      "eval_mean_token_accuracy": 0.8121141439743256,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.5437,
+      "eval_samples_per_second": 16.037,
+      "eval_steps_per_second": 2.01,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5455610822886229,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7269843220710754,
+      "learning_rate": 0.0002891155191539905,
+      "loss": 0.5407682895660401,
+      "mean_token_accuracy": 0.8393648222088814,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.6032205542151847,
+      "eval_loss": 0.6790974140167236,
+      "eval_mean_token_accuracy": 0.8165463112043531,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 89.517,
+      "eval_samples_per_second": 15.863,
+      "eval_steps_per_second": 1.988,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5474163968116045,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.8148866295814514,
+      "learning_rate": 0.00028661336281380717,
+      "loss": 0.5394441604614257,
+      "mean_token_accuracy": 0.8336028568446636,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5797032238392348,
+      "eval_loss": 0.670870304107666,
+      "eval_mean_token_accuracy": 0.8182691593518417,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 89.4468,
+      "eval_samples_per_second": 15.875,
+      "eval_steps_per_second": 1.99,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5511750865727663,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.3441656827926636,
+      "learning_rate": 0.0002840817043244964,
+      "loss": 0.5378933906555176,
+      "mean_token_accuracy": 0.8349151819944381,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5297858750217417,
+      "eval_loss": 0.680833637714386,
+      "eval_mean_token_accuracy": 0.8180924275617921,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 89.5417,
+      "eval_samples_per_second": 15.859,
+      "eval_steps_per_second": 1.988,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5379613988101483,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9813932776451111,
+      "learning_rate": 0.00028152126011546396,
+      "loss": 0.5340342044830322,
+      "mean_token_accuracy": 0.8346010789275169,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5811898510777549,
+      "eval_loss": 0.6764267683029175,
+      "eval_mean_token_accuracy": 0.8180528236239144,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 88.7831,
+      "eval_samples_per_second": 15.994,
+      "eval_steps_per_second": 2.005,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5607876226305961,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.9617928266525269,
+      "learning_rate": 0.00027893275476213383,
+      "loss": 0.5434338569641113,
+      "mean_token_accuracy": 0.8332898400723934,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5474644318390428,
+      "eval_loss": 0.6740226745605469,
+      "eval_mean_token_accuracy": 0.8185851306058047,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 89.3246,
+      "eval_samples_per_second": 15.897,
+      "eval_steps_per_second": 1.993,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5354843523353339,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.5320990085601807,
+      "learning_rate": 0.0002763169207809021,
+      "loss": 0.5251852989196777,
+      "mean_token_accuracy": 0.8416872084140777,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.5436856410141742,
+      "eval_loss": 0.6675190925598145,
+      "eval_mean_token_accuracy": 0.8202396507343549,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 89.6328,
+      "eval_samples_per_second": 15.842,
+      "eval_steps_per_second": 1.986,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5418571938167919,
+      "epoch": 4.0,
+      "grad_norm": 4.690354347229004,
+      "learning_rate": 0.000273674498421843,
+      "loss": 0.548275089263916,
+      "mean_token_accuracy": 0.836557479647847,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5497228983747825,
+      "eval_loss": 0.6621812582015991,
+      "eval_mean_token_accuracy": 0.8205679802412398,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 89.6272,
+      "eval_samples_per_second": 15.843,
+      "eval_steps_per_second": 1.986,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4195931971073151,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6678048968315125,
+      "learning_rate": 0.0002710062354592273,
+      "loss": 0.3970795154571533,
+      "mean_token_accuracy": 0.8719995342195034,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.46495268753405367,
+      "eval_loss": 0.7061581611633301,
+      "eval_mean_token_accuracy": 0.8190107995204712,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 89.6387,
+      "eval_samples_per_second": 15.841,
+      "eval_steps_per_second": 1.986,
+      "step": 1680
+    },
+    {
+      "entropy": 0.4263830740004778,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.7026669383049011,
+      "learning_rate": 0.000268312886979911,
+      "loss": 0.398905086517334,
+      "mean_token_accuracy": 0.8691012300550938,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.45755320670229666,
+      "eval_loss": 0.7102291584014893,
+      "eval_mean_token_accuracy": 0.819212955370378,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 89.2777,
+      "eval_samples_per_second": 15.905,
+      "eval_steps_per_second": 1.994,
+      "step": 1700
+    },
+    {
+      "entropy": 0.41640330031514167,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.752216100692749,
+      "learning_rate": 0.00026559521516965437,
+      "loss": 0.39748263359069824,
+      "mean_token_accuracy": 0.8684553548693656,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.469178682297803,
+      "eval_loss": 0.7004125714302063,
+      "eval_mean_token_accuracy": 0.8194386949030201,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 89.8428,
+      "eval_samples_per_second": 15.805,
+      "eval_steps_per_second": 1.981,
+      "step": 1720
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.7232312454235136e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1740/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1740/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1740/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1740/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1740/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..7ce4b072fcbe5b636867e13937c56fedee3b68b4
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1740/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.02728394274790723,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1740/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1740/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1740/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1740/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1740/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1740/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1740/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1740/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..5826899c7584526b8913fb08fb5f8543f662ccfc
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1740/trainer_state.json
@@ -0,0 +1,1861 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 4.193120096560048,
+  "eval_steps": 20,
+  "global_step": 1740,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.024795526266098,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.945375919342041,
+      "learning_rate": 1.6698127428345936e-05,
+      "loss": 1.7825420379638672,
+      "mean_token_accuracy": 0.6315708436071873,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.29921497588747,
+      "eval_loss": 1.169131875038147,
+      "eval_mean_token_accuracy": 0.7185706342204233,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 91.7898,
+      "eval_samples_per_second": 15.47,
+      "eval_steps_per_second": 1.939,
+      "step": 20
+    },
+    {
+      "entropy": 1.0346344470977784,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.7916979789733887,
+      "learning_rate": 3.427510366871008e-05,
+      "loss": 0.9480395317077637,
+      "mean_token_accuracy": 0.7543429024517536,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 0.9673695430326997,
+      "eval_loss": 0.8588430285453796,
+      "eval_mean_token_accuracy": 0.7737233002534073,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 89.8219,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 40
+    },
+    {
+      "entropy": 0.8977566614747048,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7972187995910645,
+      "learning_rate": 5.185207990907422e-05,
+      "loss": 0.8029604911804199,
+      "mean_token_accuracy": 0.7818016350269318,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8436836982041263,
+      "eval_loss": 0.7893033623695374,
+      "eval_mean_token_accuracy": 0.7858914984076211,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 89.9349,
+      "eval_samples_per_second": 15.789,
+      "eval_steps_per_second": 1.979,
+      "step": 60
+    },
+    {
+      "entropy": 0.8650321319699288,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.2128729820251465,
+      "learning_rate": 6.942905614943836e-05,
+      "loss": 0.77417893409729,
+      "mean_token_accuracy": 0.7875479467213153,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8209857180547179,
+      "eval_loss": 0.7572280168533325,
+      "eval_mean_token_accuracy": 0.7925970383574453,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 89.9459,
+      "eval_samples_per_second": 15.787,
+      "eval_steps_per_second": 1.979,
+      "step": 80
+    },
+    {
+      "entropy": 0.8096660938113928,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.1254287958145142,
+      "learning_rate": 8.700603238980251e-05,
+      "loss": 0.7170331001281738,
+      "mean_token_accuracy": 0.7994555257260799,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.792780208788561,
+      "eval_loss": 0.7442984580993652,
+      "eval_mean_token_accuracy": 0.7954704573984896,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 90.294,
+      "eval_samples_per_second": 15.726,
+      "eval_steps_per_second": 1.971,
+      "step": 100
+    },
+    {
+      "entropy": 0.7979829199612141,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2140507698059082,
+      "learning_rate": 0.00010458300863016666,
+      "loss": 0.7141861915588379,
+      "mean_token_accuracy": 0.8000014744699001,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.7938692486018277,
+      "eval_loss": 0.7279797196388245,
+      "eval_mean_token_accuracy": 0.7989798734027348,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 89.7189,
+      "eval_samples_per_second": 15.827,
+      "eval_steps_per_second": 1.984,
+      "step": 120
+    },
+    {
+      "entropy": 0.7916347607970238,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.0506458282470703,
+      "learning_rate": 0.0001221599848705308,
+      "loss": 0.7117055416107178,
+      "mean_token_accuracy": 0.7970046654343605,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.8052781063519167,
+      "eval_loss": 0.722576916217804,
+      "eval_mean_token_accuracy": 0.7982530523551984,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 90.0376,
+      "eval_samples_per_second": 15.771,
+      "eval_steps_per_second": 1.977,
+      "step": 140
+    },
+    {
+      "entropy": 0.7880666613578796,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8409207463264465,
+      "learning_rate": 0.00013973696111089492,
+      "loss": 0.7052523136138916,
+      "mean_token_accuracy": 0.8000123649835587,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7787443134891853,
+      "eval_loss": 0.7137772440910339,
+      "eval_mean_token_accuracy": 0.8014209287220173,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 90.2562,
+      "eval_samples_per_second": 15.733,
+      "eval_steps_per_second": 1.972,
+      "step": 160
+    },
+    {
+      "entropy": 0.7696007996797561,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.0182477235794067,
+      "learning_rate": 0.00015731393735125907,
+      "loss": 0.7008797645568847,
+      "mean_token_accuracy": 0.8007259473204613,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.752259682068664,
+      "eval_loss": 0.7171286344528198,
+      "eval_mean_token_accuracy": 0.80052122574174,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 90.0797,
+      "eval_samples_per_second": 15.764,
+      "eval_steps_per_second": 1.976,
+      "step": 180
+    },
+    {
+      "entropy": 0.7828126326203346,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9553632736206055,
+      "learning_rate": 0.0001748909135916232,
+      "loss": 0.7046853542327881,
+      "mean_token_accuracy": 0.7988562889397144,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7389638169427936,
+      "eval_loss": 0.7187935709953308,
+      "eval_mean_token_accuracy": 0.7995899999409579,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 89.9223,
+      "eval_samples_per_second": 15.791,
+      "eval_steps_per_second": 1.979,
+      "step": 200
+    },
+    {
+      "entropy": 0.7909286297857762,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 1.134459376335144,
+      "learning_rate": 0.00019246788983198735,
+      "loss": 0.6989312171936035,
+      "mean_token_accuracy": 0.7991565138101577,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7496965199374082,
+      "eval_loss": 0.7106321454048157,
+      "eval_mean_token_accuracy": 0.8046183442131857,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 89.6899,
+      "eval_samples_per_second": 15.832,
+      "eval_steps_per_second": 1.985,
+      "step": 220
+    },
+    {
+      "entropy": 0.7651963323354721,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.9557585716247559,
+      "learning_rate": 0.0002100448660723515,
+      "loss": 0.6984559059143066,
+      "mean_token_accuracy": 0.8019823372364044,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7615563079212488,
+      "eval_loss": 0.7098539471626282,
+      "eval_mean_token_accuracy": 0.8047233508544022,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 90.3752,
+      "eval_samples_per_second": 15.712,
+      "eval_steps_per_second": 1.97,
+      "step": 240
+    },
+    {
+      "entropy": 0.787507726252079,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.9279311299324036,
+      "learning_rate": 0.00022762184231271568,
+      "loss": 0.708138370513916,
+      "mean_token_accuracy": 0.8013308703899383,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7856640896100676,
+      "eval_loss": 0.7191570401191711,
+      "eval_mean_token_accuracy": 0.8024677249153008,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 93.7336,
+      "eval_samples_per_second": 15.149,
+      "eval_steps_per_second": 1.899,
+      "step": 260
+    },
+    {
+      "entropy": 0.768489234894514,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.7992776036262512,
+      "learning_rate": 0.0002451988185530798,
+      "loss": 0.7035849571228028,
+      "mean_token_accuracy": 0.8031105332076549,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7682498740346244,
+      "eval_loss": 0.7149022221565247,
+      "eval_mean_token_accuracy": 0.7973497995499814,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 89.7309,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 280
+    },
+    {
+      "entropy": 0.7720955617725849,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.1931514739990234,
+      "learning_rate": 0.0002627757947934439,
+      "loss": 0.6965402603149414,
+      "mean_token_accuracy": 0.8040348663926125,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7214708304807042,
+      "eval_loss": 0.7350823879241943,
+      "eval_mean_token_accuracy": 0.7942911925610532,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 89.6451,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 300
+    },
+    {
+      "entropy": 0.7930095426738262,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8994467258453369,
+      "learning_rate": 0.0002803527710338081,
+      "loss": 0.7331492900848389,
+      "mean_token_accuracy": 0.796185302734375,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.8142535488926963,
+      "eval_loss": 0.7237834334373474,
+      "eval_mean_token_accuracy": 0.799495273761535,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 90.6759,
+      "eval_samples_per_second": 15.66,
+      "eval_steps_per_second": 1.963,
+      "step": 320
+    },
+    {
+      "entropy": 0.7825102441012859,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 1.8224239349365234,
+      "learning_rate": 0.00029792974727417223,
+      "loss": 0.7331175804138184,
+      "mean_token_accuracy": 0.7977667152881622,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7327135443017724,
+      "eval_loss": 0.7388784289360046,
+      "eval_mean_token_accuracy": 0.8007041252730938,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 89.7692,
+      "eval_samples_per_second": 15.818,
+      "eval_steps_per_second": 1.983,
+      "step": 340
+    },
+    {
+      "entropy": 0.819331557303667,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 1.0407779216766357,
+      "learning_rate": 0.0003155067235145364,
+      "loss": 0.7456695556640625,
+      "mean_token_accuracy": 0.7936980701982975,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.8065585420372781,
+      "eval_loss": 0.735519528388977,
+      "eval_mean_token_accuracy": 0.7974206296245704,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 89.4336,
+      "eval_samples_per_second": 15.878,
+      "eval_steps_per_second": 1.99,
+      "step": 360
+    },
+    {
+      "entropy": 0.8114653021097183,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.1209490299224854,
+      "learning_rate": 0.0003330836997549005,
+      "loss": 0.7340899467468261,
+      "mean_token_accuracy": 0.7914272703230381,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.8025527285056168,
+      "eval_loss": 0.7295576930046082,
+      "eval_mean_token_accuracy": 0.7990576628218876,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 89.9579,
+      "eval_samples_per_second": 15.785,
+      "eval_steps_per_second": 1.979,
+      "step": 380
+    },
+    {
+      "entropy": 0.8042497783899307,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 1.2437798976898193,
+      "learning_rate": 0.0003506606759952647,
+      "loss": 0.7381744384765625,
+      "mean_token_accuracy": 0.7967362694442273,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7717011336530193,
+      "eval_loss": 0.7436173558235168,
+      "eval_mean_token_accuracy": 0.7975849482450592,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 89.8411,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 400
+    },
+    {
+      "entropy": 0.7987750010056929,
+      "epoch": 1.012070006035003,
+      "grad_norm": 1.2390918731689453,
+      "learning_rate": 0.000364721224843344,
+      "loss": 0.7317790031433106,
+      "mean_token_accuracy": 0.7986087551364651,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7333170604170038,
+      "eval_loss": 0.7456948757171631,
+      "eval_mean_token_accuracy": 0.7964897577682238,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 89.7288,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 420
+    },
+    {
+      "entropy": 0.7372976846992969,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.1552740335464478,
+      "learning_rate": 0.00036468510102269575,
+      "loss": 0.6734249114990234,
+      "mean_token_accuracy": 0.805834824591875,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.7130355526891987,
+      "eval_loss": 0.7555333375930786,
+      "eval_mean_token_accuracy": 0.7961652781186479,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 89.8713,
+      "eval_samples_per_second": 15.8,
+      "eval_steps_per_second": 1.981,
+      "step": 440
+    },
+    {
+      "entropy": 0.7465668715536594,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 1.1991469860076904,
+      "learning_rate": 0.0003645973816745066,
+      "loss": 0.6871879577636719,
+      "mean_token_accuracy": 0.8042450234293937,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7772979542110743,
+      "eval_loss": 0.7561826705932617,
+      "eval_mean_token_accuracy": 0.7953876427720102,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 89.8395,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 460
+    },
+    {
+      "entropy": 0.7477855801582336,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 1.325486660003662,
+      "learning_rate": 0.00036445809162231435,
+      "loss": 0.6836830139160156,
+      "mean_token_accuracy": 0.8021619468927383,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.7098110676481483,
+      "eval_loss": 0.758007287979126,
+      "eval_mean_token_accuracy": 0.7909953189030122,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 89.6101,
+      "eval_samples_per_second": 15.846,
+      "eval_steps_per_second": 1.986,
+      "step": 480
+    },
+    {
+      "entropy": 0.7632691070437432,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 1.0122262239456177,
+      "learning_rate": 0.0003642672702835562,
+      "loss": 0.7071213245391845,
+      "mean_token_accuracy": 0.8000680930912495,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7645382720432924,
+      "eval_loss": 0.7492606043815613,
+      "eval_mean_token_accuracy": 0.7976858676149604,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 89.6092,
+      "eval_samples_per_second": 15.847,
+      "eval_steps_per_second": 1.986,
+      "step": 500
+    },
+    {
+      "entropy": 0.7623726457357407,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.367774486541748,
+      "learning_rate": 0.00036402497165841384,
+      "loss": 0.6965175151824952,
+      "mean_token_accuracy": 0.8016501650214195,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7868083715438843,
+      "eval_loss": 0.7481449842453003,
+      "eval_mean_token_accuracy": 0.799330582779445,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 89.8241,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 520
+    },
+    {
+      "entropy": 0.7535207405686378,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 1.213614821434021,
+      "learning_rate": 0.00036373126431453207,
+      "loss": 0.7078310489654541,
+      "mean_token_accuracy": 0.7998907208442688,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7829091950748743,
+      "eval_loss": 0.7416072487831116,
+      "eval_mean_token_accuracy": 0.7969146250339036,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 89.4607,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 540
+    },
+    {
+      "entropy": 0.7527363449335098,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 1.0929982662200928,
+      "learning_rate": 0.00036338623136761495,
+      "loss": 0.7058369636535644,
+      "mean_token_accuracy": 0.7997087433934211,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.728419938114252,
+      "eval_loss": 0.744873046875,
+      "eval_mean_token_accuracy": 0.7966634296299366,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 89.6597,
+      "eval_samples_per_second": 15.838,
+      "eval_steps_per_second": 1.985,
+      "step": 560
+    },
+    {
+      "entropy": 0.7406099259853363,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.409364104270935,
+      "learning_rate": 0.00036298997045790515,
+      "loss": 0.6902852058410645,
+      "mean_token_accuracy": 0.805386807769537,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.6991632758231645,
+      "eval_loss": 0.7480319738388062,
+      "eval_mean_token_accuracy": 0.7970068006033308,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 89.8329,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 580
+    },
+    {
+      "entropy": 0.7459449704736472,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.7053111791610718,
+      "learning_rate": 0.00036254259372255277,
+      "loss": 0.7044640064239502,
+      "mean_token_accuracy": 0.8005018755793571,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7454566660891758,
+      "eval_loss": 0.7551484704017639,
+      "eval_mean_token_accuracy": 0.7897093731365846,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 89.8313,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 600
+    },
+    {
+      "entropy": 0.7621391348540782,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 1.174443244934082,
+      "learning_rate": 0.000362044227763882,
+      "loss": 0.7111573696136475,
+      "mean_token_accuracy": 0.798752411454916,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7664387949397055,
+      "eval_loss": 0.7365804314613342,
+      "eval_mean_token_accuracy": 0.7963101200843126,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 89.9623,
+      "eval_samples_per_second": 15.784,
+      "eval_steps_per_second": 1.979,
+      "step": 620
+    },
+    {
+      "entropy": 0.7424944408237935,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.2898184061050415,
+      "learning_rate": 0.000361495013613564,
+      "loss": 0.6913110733032226,
+      "mean_token_accuracy": 0.8009015507996082,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7255479529332579,
+      "eval_loss": 0.7345991730690002,
+      "eval_mean_token_accuracy": 0.8013592945056015,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 89.817,
+      "eval_samples_per_second": 15.81,
+      "eval_steps_per_second": 1.982,
+      "step": 640
+    },
+    {
+      "entropy": 0.7303367637097835,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.1310476064682007,
+      "learning_rate": 0.00036089510669270666,
+      "loss": 0.6875617027282714,
+      "mean_token_accuracy": 0.8035397171974182,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7101308248016271,
+      "eval_loss": 0.7394917011260986,
+      "eval_mean_token_accuracy": 0.7990792942850777,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 89.7431,
+      "eval_samples_per_second": 15.823,
+      "eval_steps_per_second": 1.983,
+      "step": 660
+    },
+    {
+      "entropy": 0.7387678384780884,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.9645389914512634,
+      "learning_rate": 0.0003602446767678725,
+      "loss": 0.6911204338073731,
+      "mean_token_accuracy": 0.8056333385407924,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7281073737010527,
+      "eval_loss": 0.7376708984375,
+      "eval_mean_token_accuracy": 0.7997356415464637,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 90.0338,
+      "eval_samples_per_second": 15.772,
+      "eval_steps_per_second": 1.977,
+      "step": 680
+    },
+    {
+      "entropy": 0.7449554048478604,
+      "epoch": 1.687990343995172,
+      "grad_norm": 1.9069629907608032,
+      "learning_rate": 0.0003595439079030364,
+      "loss": 0.7042286396026611,
+      "mean_token_accuracy": 0.8018639378249646,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7022633817088738,
+      "eval_loss": 0.7320939898490906,
+      "eval_mean_token_accuracy": 0.8006051541044471,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 89.8948,
+      "eval_samples_per_second": 15.796,
+      "eval_steps_per_second": 1.98,
+      "step": 700
+    },
+    {
+      "entropy": 0.723249051719904,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.8333877325057983,
+      "learning_rate": 0.00035879299840749777,
+      "loss": 0.7043714046478271,
+      "mean_token_accuracy": 0.8028345607221127,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.672436946898364,
+      "eval_loss": 0.713233470916748,
+      "eval_mean_token_accuracy": 0.8047760577684038,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 89.6394,
+      "eval_samples_per_second": 15.841,
+      "eval_steps_per_second": 1.986,
+      "step": 720
+    },
+    {
+      "entropy": 0.730014206469059,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.4580037593841553,
+      "learning_rate": 0.00035799216077976137,
+      "loss": 0.6985176563262939,
+      "mean_token_accuracy": 0.8015547141432762,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.6908873795123582,
+      "eval_loss": 0.7198202013969421,
+      "eval_mean_token_accuracy": 0.8024131682481659,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 89.3771,
+      "eval_samples_per_second": 15.888,
+      "eval_steps_per_second": 1.992,
+      "step": 740
+    },
+    {
+      "entropy": 0.7219605796039105,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.51682710647583,
+      "learning_rate": 0.000357141621647403,
+      "loss": 0.6844424724578857,
+      "mean_token_accuracy": 0.8032813094556331,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7113401109582922,
+      "eval_loss": 0.7195309400558472,
+      "eval_mean_token_accuracy": 0.8019187035185568,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 89.4658,
+      "eval_samples_per_second": 15.872,
+      "eval_steps_per_second": 1.99,
+      "step": 760
+    },
+    {
+      "entropy": 0.729826743900776,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 1.0838000774383545,
+      "learning_rate": 0.0003562416217029361,
+      "loss": 0.685858678817749,
+      "mean_token_accuracy": 0.8050567395985126,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.649370267484965,
+      "eval_loss": 0.7177029252052307,
+      "eval_mean_token_accuracy": 0.8050174043419656,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 89.2287,
+      "eval_samples_per_second": 15.914,
+      "eval_steps_per_second": 1.995,
+      "step": 780
+    },
+    {
+      "entropy": 0.6799364153295755,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 1.1364924907684326,
+      "learning_rate": 0.00035529241563569903,
+      "loss": 0.6714428901672364,
+      "mean_token_accuracy": 0.8087341658771038,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6603799047095052,
+      "eval_loss": 0.7201809287071228,
+      "eval_mean_token_accuracy": 0.802863868769635,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 89.2718,
+      "eval_samples_per_second": 15.906,
+      "eval_steps_per_second": 1.994,
+      "step": 800
+    },
+    {
+      "entropy": 0.7078610375523567,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 1.2191942930221558,
+      "learning_rate": 0.0003542942720597808,
+      "loss": 0.6996590614318847,
+      "mean_token_accuracy": 0.8049512051045895,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.7240545264120852,
+      "eval_loss": 0.7301309108734131,
+      "eval_mean_token_accuracy": 0.8015973166133581,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 89.3438,
+      "eval_samples_per_second": 15.894,
+      "eval_steps_per_second": 1.992,
+      "step": 820
+    },
+    {
+      "entropy": 0.6557391556826505,
+      "epoch": 2.024140012070006,
+      "grad_norm": 1.36203134059906,
+      "learning_rate": 0.0003532474734380064,
+      "loss": 0.6327113628387451,
+      "mean_token_accuracy": 0.8142829847026181,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6573808648613062,
+      "eval_loss": 0.7334365844726562,
+      "eval_mean_token_accuracy": 0.8043633893634496,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 89.3359,
+      "eval_samples_per_second": 15.895,
+      "eval_steps_per_second": 1.992,
+      "step": 840
+    },
+    {
+      "entropy": 0.6409128502011299,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 1.5804955959320068,
+      "learning_rate": 0.0003521523160020035,
+      "loss": 0.6023559093475341,
+      "mean_token_accuracy": 0.8222251623868942,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6983234788594621,
+      "eval_loss": 0.7232135534286499,
+      "eval_mean_token_accuracy": 0.8010654268639811,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 89.2615,
+      "eval_samples_per_second": 15.908,
+      "eval_steps_per_second": 1.994,
+      "step": 860
+    },
+    {
+      "entropy": 0.6151149723678827,
+      "epoch": 2.12070006035003,
+      "grad_norm": 1.1975436210632324,
+      "learning_rate": 0.00035100910966837193,
+      "loss": 0.5803564548492431,
+      "mean_token_accuracy": 0.8256325736641884,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6636555573243773,
+      "eval_loss": 0.7205662727355957,
+      "eval_mean_token_accuracy": 0.8036768369460374,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 89.2957,
+      "eval_samples_per_second": 15.902,
+      "eval_steps_per_second": 1.993,
+      "step": 880
+    },
+    {
+      "entropy": 0.6201727617532015,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9907131195068359,
+      "learning_rate": 0.0003498181779509813,
+      "loss": 0.6101872444152832,
+      "mean_token_accuracy": 0.8196512959897518,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.6473788086617931,
+      "eval_loss": 0.7062397003173828,
+      "eval_mean_token_accuracy": 0.8078362184963869,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 88.6198,
+      "eval_samples_per_second": 16.024,
+      "eval_steps_per_second": 2.009,
+      "step": 900
+    },
+    {
+      "entropy": 0.6097332935780286,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.9055633544921875,
+      "learning_rate": 0.00034857985786942026,
+      "loss": 0.5956094264984131,
+      "mean_token_accuracy": 0.8240575887262821,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.612629868341296,
+      "eval_loss": 0.7106770277023315,
+      "eval_mean_token_accuracy": 0.805324277181304,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 89.4383,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 920
+    },
+    {
+      "entropy": 0.6182701248675585,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.9513736367225647,
+      "learning_rate": 0.00034729449985362404,
+      "loss": 0.6019269466400147,
+      "mean_token_accuracy": 0.8187106013298034,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6437647456533453,
+      "eval_loss": 0.7173502445220947,
+      "eval_mean_token_accuracy": 0.8056439573175451,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.9568,
+      "eval_samples_per_second": 15.963,
+      "eval_steps_per_second": 2.001,
+      "step": 940
+    },
+    {
+      "entropy": 0.6508570514619351,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.9606207013130188,
+      "learning_rate": 0.0003459624676447067,
+      "loss": 0.6296409606933594,
+      "mean_token_accuracy": 0.8167732007801533,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6288247135248077,
+      "eval_loss": 0.7160675525665283,
+      "eval_mean_token_accuracy": 0.8066010689467527,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 89.0044,
+      "eval_samples_per_second": 15.954,
+      "eval_steps_per_second": 2.0,
+      "step": 960
+    },
+    {
+      "entropy": 0.6380081418901682,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.9677181839942932,
+      "learning_rate": 0.000344584138192027,
+      "loss": 0.6210546493530273,
+      "mean_token_accuracy": 0.8186135642230511,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6407805031604981,
+      "eval_loss": 0.7055043578147888,
+      "eval_mean_token_accuracy": 0.8070103157772107,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 89.5606,
+      "eval_samples_per_second": 15.855,
+      "eval_steps_per_second": 1.987,
+      "step": 980
+    },
+    {
+      "entropy": 0.6310251638293266,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 2.1046407222747803,
+      "learning_rate": 0.000343159901546516,
+      "loss": 0.6165830135345459,
+      "mean_token_accuracy": 0.8207855716347694,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6461507470420237,
+      "eval_loss": 0.7292136549949646,
+      "eval_mean_token_accuracy": 0.8030418261383356,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 89.315,
+      "eval_samples_per_second": 15.899,
+      "eval_steps_per_second": 1.993,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6256617344915867,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 1.1768349409103394,
+      "learning_rate": 0.0003416901607502972,
+      "loss": 0.6142709255218506,
+      "mean_token_accuracy": 0.8220669947564602,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6150523993406403,
+      "eval_loss": 0.7019563317298889,
+      "eval_mean_token_accuracy": 0.8094323099998946,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 89.419,
+      "eval_samples_per_second": 15.88,
+      "eval_steps_per_second": 1.991,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6126735735684633,
+      "epoch": 2.506940253470127,
+      "grad_norm": 1.8065496683120728,
+      "learning_rate": 0.00034017533172263055,
+      "loss": 0.6056458473205566,
+      "mean_token_accuracy": 0.8203762136399746,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6544345456562685,
+      "eval_loss": 0.7086517214775085,
+      "eval_mean_token_accuracy": 0.806709805901131,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 88.981,
+      "eval_samples_per_second": 15.958,
+      "eval_steps_per_second": 2.0,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6311754353344441,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 1.0576369762420654,
+      "learning_rate": 0.00033861584314221236,
+      "loss": 0.6176392555236816,
+      "mean_token_accuracy": 0.8177410490810871,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6273799200406235,
+      "eval_loss": 0.708163321018219,
+      "eval_mean_token_accuracy": 0.8032017099053672,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 89.7668,
+      "eval_samples_per_second": 15.819,
+      "eval_steps_per_second": 1.983,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6200249589979648,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.997065007686615,
+      "learning_rate": 0.0003370121363258637,
+      "loss": 0.6203888893127442,
+      "mean_token_accuracy": 0.8196275025606156,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6457576811983344,
+      "eval_loss": 0.6972290277481079,
+      "eval_mean_token_accuracy": 0.8095980849158898,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 89.2027,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 1080
+    },
+    {
+      "entropy": 0.626810473203659,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.865475058555603,
+      "learning_rate": 0.0003353646651036438,
+      "loss": 0.6102585792541504,
+      "mean_token_accuracy": 0.819708751142025,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.6412022431914726,
+      "eval_loss": 0.7036139965057373,
+      "eval_mean_token_accuracy": 0.8070317135098275,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 89.4733,
+      "eval_samples_per_second": 15.871,
+      "eval_steps_per_second": 1.989,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6581619590520859,
+      "epoch": 2.700060350030175,
+      "grad_norm": 1.00307035446167,
+      "learning_rate": 0.00033367389569042064,
+      "loss": 0.6397720336914062,
+      "mean_token_accuracy": 0.81413309648633,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6288736402318719,
+      "eval_loss": 0.6966825723648071,
+      "eval_mean_token_accuracy": 0.8094039669867312,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 88.7341,
+      "eval_samples_per_second": 16.003,
+      "eval_steps_per_second": 2.006,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6399412982165813,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.8589329719543457,
+      "learning_rate": 0.0003319403065539384,
+      "loss": 0.6347338199615479,
+      "mean_token_accuracy": 0.8139534957706929,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.683491913455256,
+      "eval_loss": 0.6984149217605591,
+      "eval_mean_token_accuracy": 0.8083989469522841,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 89.3929,
+      "eval_samples_per_second": 15.885,
+      "eval_steps_per_second": 1.991,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6264835461974144,
+      "epoch": 2.796620398310199,
+      "grad_norm": 0.7807853817939758,
+      "learning_rate": 0.0003301643882794163,
+      "loss": 0.6162629127502441,
+      "mean_token_accuracy": 0.8216292977333068,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6578792159476977,
+      "eval_loss": 0.6921441555023193,
+      "eval_mean_token_accuracy": 0.8094299177775223,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 89.2356,
+      "eval_samples_per_second": 15.913,
+      "eval_steps_per_second": 1.995,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6269129011780024,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 1.1675593852996826,
+      "learning_rate": 0.0003283466434307189,
+      "loss": 0.6235532760620117,
+      "mean_token_accuracy": 0.8188897788524627,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6533248447970058,
+      "eval_loss": 0.696849524974823,
+      "eval_mean_token_accuracy": 0.8102703713968898,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2602,
+      "eval_samples_per_second": 16.089,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.6259873781353236,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.9596486687660217,
+      "learning_rate": 0.00032648758640813655,
+      "loss": 0.6118728160858155,
+      "mean_token_accuracy": 0.8171859130263328,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6212078754821521,
+      "eval_loss": 0.6942870020866394,
+      "eval_mean_token_accuracy": 0.810484343365337,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 89.4375,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 1200
+    },
+    {
+      "entropy": 0.6142537571489811,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 0.9625058770179749,
+      "learning_rate": 0.00032458774330281615,
+      "loss": 0.6060261249542236,
+      "mean_token_accuracy": 0.8223113484680653,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6553665667437436,
+      "eval_loss": 0.6835269331932068,
+      "eval_mean_token_accuracy": 0.8134723968720168,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 89.5092,
+      "eval_samples_per_second": 15.864,
+      "eval_steps_per_second": 1.989,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6301972426474094,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.9657333493232727,
+      "learning_rate": 0.0003226476517478835,
+      "loss": 0.6277759075164795,
+      "mean_token_accuracy": 0.8164266042411328,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6594253118788258,
+      "eval_loss": 0.6820935606956482,
+      "eval_mean_token_accuracy": 0.8132368198941263,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 89.6211,
+      "eval_samples_per_second": 15.844,
+      "eval_steps_per_second": 1.986,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5346100655469027,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8541185259819031,
+      "learning_rate": 0.0003206678607662996,
+      "loss": 0.5261692047119141,
+      "mean_token_accuracy": 0.8382289425119177,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5626645662476507,
+      "eval_loss": 0.7028542160987854,
+      "eval_mean_token_accuracy": 0.8133395572056931,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 89.1267,
+      "eval_samples_per_second": 15.932,
+      "eval_steps_per_second": 1.997,
+      "step": 1260
+    },
+    {
+      "entropy": 0.5001101832836866,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.8208815455436707,
+      "learning_rate": 0.0003186489306154935,
+      "loss": 0.4850132942199707,
+      "mean_token_accuracy": 0.8468983843922615,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5889462046743779,
+      "eval_loss": 0.691015899181366,
+      "eval_mean_token_accuracy": 0.8135226466012805,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.9677,
+      "eval_samples_per_second": 15.961,
+      "eval_steps_per_second": 2.001,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5242766339331866,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 0.7893072962760925,
+      "learning_rate": 0.0003165914326288163,
+      "loss": 0.5061577320098877,
+      "mean_token_accuracy": 0.8425608821213245,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5543691962957382,
+      "eval_loss": 0.6913734078407288,
+      "eval_mean_token_accuracy": 0.8154889680026622,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 89.6963,
+      "eval_samples_per_second": 15.831,
+      "eval_steps_per_second": 1.984,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5327721010893584,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.8992080092430115,
+      "learning_rate": 0.0003144959490538604,
+      "loss": 0.5068036556243897,
+      "mean_token_accuracy": 0.8420542575418949,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5734540803378887,
+      "eval_loss": 0.6948716044425964,
+      "eval_mean_token_accuracy": 0.8137217944257715,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 89.4597,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5247377116233111,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.9291845560073853,
+      "learning_rate": 0.0003123630728876902,
+      "loss": 0.5023272037506104,
+      "mean_token_accuracy": 0.8435182586312294,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5492362935891312,
+      "eval_loss": 0.6980633735656738,
+      "eval_mean_token_accuracy": 0.8155082919624415,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 89.2124,
+      "eval_samples_per_second": 15.917,
+      "eval_steps_per_second": 1.995,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5380499072372913,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 1.0477524995803833,
+      "learning_rate": 0.00031019340770903136,
+      "loss": 0.5264591693878173,
+      "mean_token_accuracy": 0.8391959741711617,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5896503307511297,
+      "eval_loss": 0.6997144222259521,
+      "eval_mean_token_accuracy": 0.8124373062942805,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 89.1147,
+      "eval_samples_per_second": 15.935,
+      "eval_steps_per_second": 1.997,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5460653610527515,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 1.2473069429397583,
+      "learning_rate": 0.00030798756750746476,
+      "loss": 0.5319677829742432,
+      "mean_token_accuracy": 0.8364320226013661,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5659052182114526,
+      "eval_loss": 0.6924049258232117,
+      "eval_mean_token_accuracy": 0.8153053585732921,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 89.6468,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5367407951503992,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 1.2679712772369385,
+      "learning_rate": 0.00030574617650967485,
+      "loss": 0.5190927028656006,
+      "mean_token_accuracy": 0.8399469174444676,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5581759144081159,
+      "eval_loss": 0.6933317184448242,
+      "eval_mean_token_accuracy": 0.8139757862251796,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 89.0749,
+      "eval_samples_per_second": 15.942,
+      "eval_steps_per_second": 1.998,
+      "step": 1400
+    },
+    {
+      "entropy": 0.530736118927598,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.9717612862586975,
+      "learning_rate": 0.0003034698690028009,
+      "loss": 0.512363052368164,
+      "mean_token_accuracy": 0.8419127956032753,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5233335836549823,
+      "eval_loss": 0.6958550214767456,
+      "eval_mean_token_accuracy": 0.8158778501360604,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 88.8664,
+      "eval_samples_per_second": 15.979,
+      "eval_steps_per_second": 2.003,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5327631575986743,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7649208903312683,
+      "learning_rate": 0.00030115928915494116,
+      "loss": 0.5202179431915284,
+      "mean_token_accuracy": 0.8384683802723885,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.5960972657364406,
+      "eval_loss": 0.6984783411026001,
+      "eval_mean_token_accuracy": 0.8095930597085631,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 89.5168,
+      "eval_samples_per_second": 15.863,
+      "eval_steps_per_second": 1.988,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5428155034780502,
+      "epoch": 3.51901025950513,
+      "grad_norm": 1.2790874242782593,
+      "learning_rate": 0.00029881509083286116,
+      "loss": 0.524717378616333,
+      "mean_token_accuracy": 0.8346644505858422,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.550318562917495,
+      "eval_loss": 0.6988471746444702,
+      "eval_mean_token_accuracy": 0.8132969897784544,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 89.2598,
+      "eval_samples_per_second": 15.909,
+      "eval_steps_per_second": 1.994,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5489993002265692,
+      "epoch": 3.567290283645142,
+      "grad_norm": 1.055821418762207,
+      "learning_rate": 0.00029643793741695687,
+      "loss": 0.5421150207519532,
+      "mean_token_accuracy": 0.8345710933208466,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5816821718818685,
+      "eval_loss": 0.6853668689727783,
+      "eval_mean_token_accuracy": 0.8142332360985574,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 89.0424,
+      "eval_samples_per_second": 15.947,
+      "eval_steps_per_second": 1.999,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5451365381479263,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.9027532935142517,
+      "learning_rate": 0.00029402850161352583,
+      "loss": 0.5463609218597412,
+      "mean_token_accuracy": 0.8369908526539802,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5344233209832331,
+      "eval_loss": 0.6794531345367432,
+      "eval_mean_token_accuracy": 0.8177759824843889,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 89.123,
+      "eval_samples_per_second": 15.933,
+      "eval_steps_per_second": 1.997,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5521467242389917,
+      "epoch": 3.663850331925166,
+      "grad_norm": 1.4606311321258545,
+      "learning_rate": 0.0002915874652643997,
+      "loss": 0.5421902179718018,
+      "mean_token_accuracy": 0.8339455373585224,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5807709094513668,
+      "eval_loss": 0.6949691772460938,
+      "eval_mean_token_accuracy": 0.8121141439743256,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.5437,
+      "eval_samples_per_second": 16.037,
+      "eval_steps_per_second": 2.01,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5455610822886229,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7269843220710754,
+      "learning_rate": 0.0002891155191539905,
+      "loss": 0.5407682895660401,
+      "mean_token_accuracy": 0.8393648222088814,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.6032205542151847,
+      "eval_loss": 0.6790974140167236,
+      "eval_mean_token_accuracy": 0.8165463112043531,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 89.517,
+      "eval_samples_per_second": 15.863,
+      "eval_steps_per_second": 1.988,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5474163968116045,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.8148866295814514,
+      "learning_rate": 0.00028661336281380717,
+      "loss": 0.5394441604614257,
+      "mean_token_accuracy": 0.8336028568446636,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5797032238392348,
+      "eval_loss": 0.670870304107666,
+      "eval_mean_token_accuracy": 0.8182691593518417,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 89.4468,
+      "eval_samples_per_second": 15.875,
+      "eval_steps_per_second": 1.99,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5511750865727663,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.3441656827926636,
+      "learning_rate": 0.0002840817043244964,
+      "loss": 0.5378933906555176,
+      "mean_token_accuracy": 0.8349151819944381,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5297858750217417,
+      "eval_loss": 0.680833637714386,
+      "eval_mean_token_accuracy": 0.8180924275617921,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 89.5417,
+      "eval_samples_per_second": 15.859,
+      "eval_steps_per_second": 1.988,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5379613988101483,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9813932776451111,
+      "learning_rate": 0.00028152126011546396,
+      "loss": 0.5340342044830322,
+      "mean_token_accuracy": 0.8346010789275169,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5811898510777549,
+      "eval_loss": 0.6764267683029175,
+      "eval_mean_token_accuracy": 0.8180528236239144,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 88.7831,
+      "eval_samples_per_second": 15.994,
+      "eval_steps_per_second": 2.005,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5607876226305961,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.9617928266525269,
+      "learning_rate": 0.00027893275476213383,
+      "loss": 0.5434338569641113,
+      "mean_token_accuracy": 0.8332898400723934,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5474644318390428,
+      "eval_loss": 0.6740226745605469,
+      "eval_mean_token_accuracy": 0.8185851306058047,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 89.3246,
+      "eval_samples_per_second": 15.897,
+      "eval_steps_per_second": 1.993,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5354843523353339,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.5320990085601807,
+      "learning_rate": 0.0002763169207809021,
+      "loss": 0.5251852989196777,
+      "mean_token_accuracy": 0.8416872084140777,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.5436856410141742,
+      "eval_loss": 0.6675190925598145,
+      "eval_mean_token_accuracy": 0.8202396507343549,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 89.6328,
+      "eval_samples_per_second": 15.842,
+      "eval_steps_per_second": 1.986,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5418571938167919,
+      "epoch": 4.0,
+      "grad_norm": 4.690354347229004,
+      "learning_rate": 0.000273674498421843,
+      "loss": 0.548275089263916,
+      "mean_token_accuracy": 0.836557479647847,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5497228983747825,
+      "eval_loss": 0.6621812582015991,
+      "eval_mean_token_accuracy": 0.8205679802412398,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 89.6272,
+      "eval_samples_per_second": 15.843,
+      "eval_steps_per_second": 1.986,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4195931971073151,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6678048968315125,
+      "learning_rate": 0.0002710062354592273,
+      "loss": 0.3970795154571533,
+      "mean_token_accuracy": 0.8719995342195034,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.46495268753405367,
+      "eval_loss": 0.7061581611633301,
+      "eval_mean_token_accuracy": 0.8190107995204712,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 89.6387,
+      "eval_samples_per_second": 15.841,
+      "eval_steps_per_second": 1.986,
+      "step": 1680
+    },
+    {
+      "entropy": 0.4263830740004778,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.7026669383049011,
+      "learning_rate": 0.000268312886979911,
+      "loss": 0.398905086517334,
+      "mean_token_accuracy": 0.8691012300550938,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.45755320670229666,
+      "eval_loss": 0.7102291584014893,
+      "eval_mean_token_accuracy": 0.819212955370378,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 89.2777,
+      "eval_samples_per_second": 15.905,
+      "eval_steps_per_second": 1.994,
+      "step": 1700
+    },
+    {
+      "entropy": 0.41640330031514167,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.752216100692749,
+      "learning_rate": 0.00026559521516965437,
+      "loss": 0.39748263359069824,
+      "mean_token_accuracy": 0.8684553548693656,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.469178682297803,
+      "eval_loss": 0.7004125714302063,
+      "eval_mean_token_accuracy": 0.8194386949030201,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 89.8428,
+      "eval_samples_per_second": 15.805,
+      "eval_steps_per_second": 1.981,
+      "step": 1720
+    },
+    {
+      "entropy": 0.42079071439802646,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.8349477648735046,
+      "learning_rate": 0.0002628539890974329,
+      "loss": 0.40129623413085935,
+      "mean_token_accuracy": 0.8686790131032467,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48358210871058904,
+      "eval_loss": 0.70196533203125,
+      "eval_mean_token_accuracy": 0.818096743875675,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 90.444,
+      "eval_samples_per_second": 15.7,
+      "eval_steps_per_second": 1.968,
+      "step": 1740
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.7427670988903424e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1760/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1760/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1760/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1760/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1760/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..7ce4b072fcbe5b636867e13937c56fedee3b68b4
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1760/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.02728394274790723,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1760/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1760/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1760/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1760/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1760/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1760/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1760/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1760/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..5fc7697618ec4d88bdaad248c9b2778e9253e34f
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1760/trainer_state.json
@@ -0,0 +1,1882 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 4.24140012070006,
+  "eval_steps": 20,
+  "global_step": 1760,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.024795526266098,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.945375919342041,
+      "learning_rate": 1.6698127428345936e-05,
+      "loss": 1.7825420379638672,
+      "mean_token_accuracy": 0.6315708436071873,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.29921497588747,
+      "eval_loss": 1.169131875038147,
+      "eval_mean_token_accuracy": 0.7185706342204233,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 91.7898,
+      "eval_samples_per_second": 15.47,
+      "eval_steps_per_second": 1.939,
+      "step": 20
+    },
+    {
+      "entropy": 1.0346344470977784,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.7916979789733887,
+      "learning_rate": 3.427510366871008e-05,
+      "loss": 0.9480395317077637,
+      "mean_token_accuracy": 0.7543429024517536,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 0.9673695430326997,
+      "eval_loss": 0.8588430285453796,
+      "eval_mean_token_accuracy": 0.7737233002534073,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 89.8219,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 40
+    },
+    {
+      "entropy": 0.8977566614747048,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7972187995910645,
+      "learning_rate": 5.185207990907422e-05,
+      "loss": 0.8029604911804199,
+      "mean_token_accuracy": 0.7818016350269318,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8436836982041263,
+      "eval_loss": 0.7893033623695374,
+      "eval_mean_token_accuracy": 0.7858914984076211,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 89.9349,
+      "eval_samples_per_second": 15.789,
+      "eval_steps_per_second": 1.979,
+      "step": 60
+    },
+    {
+      "entropy": 0.8650321319699288,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.2128729820251465,
+      "learning_rate": 6.942905614943836e-05,
+      "loss": 0.77417893409729,
+      "mean_token_accuracy": 0.7875479467213153,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8209857180547179,
+      "eval_loss": 0.7572280168533325,
+      "eval_mean_token_accuracy": 0.7925970383574453,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 89.9459,
+      "eval_samples_per_second": 15.787,
+      "eval_steps_per_second": 1.979,
+      "step": 80
+    },
+    {
+      "entropy": 0.8096660938113928,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.1254287958145142,
+      "learning_rate": 8.700603238980251e-05,
+      "loss": 0.7170331001281738,
+      "mean_token_accuracy": 0.7994555257260799,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.792780208788561,
+      "eval_loss": 0.7442984580993652,
+      "eval_mean_token_accuracy": 0.7954704573984896,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 90.294,
+      "eval_samples_per_second": 15.726,
+      "eval_steps_per_second": 1.971,
+      "step": 100
+    },
+    {
+      "entropy": 0.7979829199612141,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2140507698059082,
+      "learning_rate": 0.00010458300863016666,
+      "loss": 0.7141861915588379,
+      "mean_token_accuracy": 0.8000014744699001,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.7938692486018277,
+      "eval_loss": 0.7279797196388245,
+      "eval_mean_token_accuracy": 0.7989798734027348,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 89.7189,
+      "eval_samples_per_second": 15.827,
+      "eval_steps_per_second": 1.984,
+      "step": 120
+    },
+    {
+      "entropy": 0.7916347607970238,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.0506458282470703,
+      "learning_rate": 0.0001221599848705308,
+      "loss": 0.7117055416107178,
+      "mean_token_accuracy": 0.7970046654343605,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.8052781063519167,
+      "eval_loss": 0.722576916217804,
+      "eval_mean_token_accuracy": 0.7982530523551984,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 90.0376,
+      "eval_samples_per_second": 15.771,
+      "eval_steps_per_second": 1.977,
+      "step": 140
+    },
+    {
+      "entropy": 0.7880666613578796,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8409207463264465,
+      "learning_rate": 0.00013973696111089492,
+      "loss": 0.7052523136138916,
+      "mean_token_accuracy": 0.8000123649835587,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7787443134891853,
+      "eval_loss": 0.7137772440910339,
+      "eval_mean_token_accuracy": 0.8014209287220173,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 90.2562,
+      "eval_samples_per_second": 15.733,
+      "eval_steps_per_second": 1.972,
+      "step": 160
+    },
+    {
+      "entropy": 0.7696007996797561,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.0182477235794067,
+      "learning_rate": 0.00015731393735125907,
+      "loss": 0.7008797645568847,
+      "mean_token_accuracy": 0.8007259473204613,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.752259682068664,
+      "eval_loss": 0.7171286344528198,
+      "eval_mean_token_accuracy": 0.80052122574174,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 90.0797,
+      "eval_samples_per_second": 15.764,
+      "eval_steps_per_second": 1.976,
+      "step": 180
+    },
+    {
+      "entropy": 0.7828126326203346,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9553632736206055,
+      "learning_rate": 0.0001748909135916232,
+      "loss": 0.7046853542327881,
+      "mean_token_accuracy": 0.7988562889397144,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7389638169427936,
+      "eval_loss": 0.7187935709953308,
+      "eval_mean_token_accuracy": 0.7995899999409579,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 89.9223,
+      "eval_samples_per_second": 15.791,
+      "eval_steps_per_second": 1.979,
+      "step": 200
+    },
+    {
+      "entropy": 0.7909286297857762,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 1.134459376335144,
+      "learning_rate": 0.00019246788983198735,
+      "loss": 0.6989312171936035,
+      "mean_token_accuracy": 0.7991565138101577,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7496965199374082,
+      "eval_loss": 0.7106321454048157,
+      "eval_mean_token_accuracy": 0.8046183442131857,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 89.6899,
+      "eval_samples_per_second": 15.832,
+      "eval_steps_per_second": 1.985,
+      "step": 220
+    },
+    {
+      "entropy": 0.7651963323354721,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.9557585716247559,
+      "learning_rate": 0.0002100448660723515,
+      "loss": 0.6984559059143066,
+      "mean_token_accuracy": 0.8019823372364044,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7615563079212488,
+      "eval_loss": 0.7098539471626282,
+      "eval_mean_token_accuracy": 0.8047233508544022,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 90.3752,
+      "eval_samples_per_second": 15.712,
+      "eval_steps_per_second": 1.97,
+      "step": 240
+    },
+    {
+      "entropy": 0.787507726252079,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.9279311299324036,
+      "learning_rate": 0.00022762184231271568,
+      "loss": 0.708138370513916,
+      "mean_token_accuracy": 0.8013308703899383,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7856640896100676,
+      "eval_loss": 0.7191570401191711,
+      "eval_mean_token_accuracy": 0.8024677249153008,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 93.7336,
+      "eval_samples_per_second": 15.149,
+      "eval_steps_per_second": 1.899,
+      "step": 260
+    },
+    {
+      "entropy": 0.768489234894514,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.7992776036262512,
+      "learning_rate": 0.0002451988185530798,
+      "loss": 0.7035849571228028,
+      "mean_token_accuracy": 0.8031105332076549,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7682498740346244,
+      "eval_loss": 0.7149022221565247,
+      "eval_mean_token_accuracy": 0.7973497995499814,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 89.7309,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 280
+    },
+    {
+      "entropy": 0.7720955617725849,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.1931514739990234,
+      "learning_rate": 0.0002627757947934439,
+      "loss": 0.6965402603149414,
+      "mean_token_accuracy": 0.8040348663926125,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7214708304807042,
+      "eval_loss": 0.7350823879241943,
+      "eval_mean_token_accuracy": 0.7942911925610532,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 89.6451,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 300
+    },
+    {
+      "entropy": 0.7930095426738262,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8994467258453369,
+      "learning_rate": 0.0002803527710338081,
+      "loss": 0.7331492900848389,
+      "mean_token_accuracy": 0.796185302734375,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.8142535488926963,
+      "eval_loss": 0.7237834334373474,
+      "eval_mean_token_accuracy": 0.799495273761535,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 90.6759,
+      "eval_samples_per_second": 15.66,
+      "eval_steps_per_second": 1.963,
+      "step": 320
+    },
+    {
+      "entropy": 0.7825102441012859,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 1.8224239349365234,
+      "learning_rate": 0.00029792974727417223,
+      "loss": 0.7331175804138184,
+      "mean_token_accuracy": 0.7977667152881622,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7327135443017724,
+      "eval_loss": 0.7388784289360046,
+      "eval_mean_token_accuracy": 0.8007041252730938,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 89.7692,
+      "eval_samples_per_second": 15.818,
+      "eval_steps_per_second": 1.983,
+      "step": 340
+    },
+    {
+      "entropy": 0.819331557303667,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 1.0407779216766357,
+      "learning_rate": 0.0003155067235145364,
+      "loss": 0.7456695556640625,
+      "mean_token_accuracy": 0.7936980701982975,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.8065585420372781,
+      "eval_loss": 0.735519528388977,
+      "eval_mean_token_accuracy": 0.7974206296245704,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 89.4336,
+      "eval_samples_per_second": 15.878,
+      "eval_steps_per_second": 1.99,
+      "step": 360
+    },
+    {
+      "entropy": 0.8114653021097183,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.1209490299224854,
+      "learning_rate": 0.0003330836997549005,
+      "loss": 0.7340899467468261,
+      "mean_token_accuracy": 0.7914272703230381,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.8025527285056168,
+      "eval_loss": 0.7295576930046082,
+      "eval_mean_token_accuracy": 0.7990576628218876,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 89.9579,
+      "eval_samples_per_second": 15.785,
+      "eval_steps_per_second": 1.979,
+      "step": 380
+    },
+    {
+      "entropy": 0.8042497783899307,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 1.2437798976898193,
+      "learning_rate": 0.0003506606759952647,
+      "loss": 0.7381744384765625,
+      "mean_token_accuracy": 0.7967362694442273,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7717011336530193,
+      "eval_loss": 0.7436173558235168,
+      "eval_mean_token_accuracy": 0.7975849482450592,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 89.8411,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 400
+    },
+    {
+      "entropy": 0.7987750010056929,
+      "epoch": 1.012070006035003,
+      "grad_norm": 1.2390918731689453,
+      "learning_rate": 0.000364721224843344,
+      "loss": 0.7317790031433106,
+      "mean_token_accuracy": 0.7986087551364651,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7333170604170038,
+      "eval_loss": 0.7456948757171631,
+      "eval_mean_token_accuracy": 0.7964897577682238,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 89.7288,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 420
+    },
+    {
+      "entropy": 0.7372976846992969,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.1552740335464478,
+      "learning_rate": 0.00036468510102269575,
+      "loss": 0.6734249114990234,
+      "mean_token_accuracy": 0.805834824591875,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.7130355526891987,
+      "eval_loss": 0.7555333375930786,
+      "eval_mean_token_accuracy": 0.7961652781186479,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 89.8713,
+      "eval_samples_per_second": 15.8,
+      "eval_steps_per_second": 1.981,
+      "step": 440
+    },
+    {
+      "entropy": 0.7465668715536594,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 1.1991469860076904,
+      "learning_rate": 0.0003645973816745066,
+      "loss": 0.6871879577636719,
+      "mean_token_accuracy": 0.8042450234293937,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7772979542110743,
+      "eval_loss": 0.7561826705932617,
+      "eval_mean_token_accuracy": 0.7953876427720102,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 89.8395,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 460
+    },
+    {
+      "entropy": 0.7477855801582336,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 1.325486660003662,
+      "learning_rate": 0.00036445809162231435,
+      "loss": 0.6836830139160156,
+      "mean_token_accuracy": 0.8021619468927383,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.7098110676481483,
+      "eval_loss": 0.758007287979126,
+      "eval_mean_token_accuracy": 0.7909953189030122,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 89.6101,
+      "eval_samples_per_second": 15.846,
+      "eval_steps_per_second": 1.986,
+      "step": 480
+    },
+    {
+      "entropy": 0.7632691070437432,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 1.0122262239456177,
+      "learning_rate": 0.0003642672702835562,
+      "loss": 0.7071213245391845,
+      "mean_token_accuracy": 0.8000680930912495,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7645382720432924,
+      "eval_loss": 0.7492606043815613,
+      "eval_mean_token_accuracy": 0.7976858676149604,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 89.6092,
+      "eval_samples_per_second": 15.847,
+      "eval_steps_per_second": 1.986,
+      "step": 500
+    },
+    {
+      "entropy": 0.7623726457357407,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.367774486541748,
+      "learning_rate": 0.00036402497165841384,
+      "loss": 0.6965175151824952,
+      "mean_token_accuracy": 0.8016501650214195,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7868083715438843,
+      "eval_loss": 0.7481449842453003,
+      "eval_mean_token_accuracy": 0.799330582779445,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 89.8241,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 520
+    },
+    {
+      "entropy": 0.7535207405686378,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 1.213614821434021,
+      "learning_rate": 0.00036373126431453207,
+      "loss": 0.7078310489654541,
+      "mean_token_accuracy": 0.7998907208442688,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7829091950748743,
+      "eval_loss": 0.7416072487831116,
+      "eval_mean_token_accuracy": 0.7969146250339036,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 89.4607,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 540
+    },
+    {
+      "entropy": 0.7527363449335098,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 1.0929982662200928,
+      "learning_rate": 0.00036338623136761495,
+      "loss": 0.7058369636535644,
+      "mean_token_accuracy": 0.7997087433934211,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.728419938114252,
+      "eval_loss": 0.744873046875,
+      "eval_mean_token_accuracy": 0.7966634296299366,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 89.6597,
+      "eval_samples_per_second": 15.838,
+      "eval_steps_per_second": 1.985,
+      "step": 560
+    },
+    {
+      "entropy": 0.7406099259853363,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.409364104270935,
+      "learning_rate": 0.00036298997045790515,
+      "loss": 0.6902852058410645,
+      "mean_token_accuracy": 0.805386807769537,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.6991632758231645,
+      "eval_loss": 0.7480319738388062,
+      "eval_mean_token_accuracy": 0.7970068006033308,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 89.8329,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 580
+    },
+    {
+      "entropy": 0.7459449704736472,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.7053111791610718,
+      "learning_rate": 0.00036254259372255277,
+      "loss": 0.7044640064239502,
+      "mean_token_accuracy": 0.8005018755793571,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7454566660891758,
+      "eval_loss": 0.7551484704017639,
+      "eval_mean_token_accuracy": 0.7897093731365846,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 89.8313,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 600
+    },
+    {
+      "entropy": 0.7621391348540782,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 1.174443244934082,
+      "learning_rate": 0.000362044227763882,
+      "loss": 0.7111573696136475,
+      "mean_token_accuracy": 0.798752411454916,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7664387949397055,
+      "eval_loss": 0.7365804314613342,
+      "eval_mean_token_accuracy": 0.7963101200843126,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 89.9623,
+      "eval_samples_per_second": 15.784,
+      "eval_steps_per_second": 1.979,
+      "step": 620
+    },
+    {
+      "entropy": 0.7424944408237935,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.2898184061050415,
+      "learning_rate": 0.000361495013613564,
+      "loss": 0.6913110733032226,
+      "mean_token_accuracy": 0.8009015507996082,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7255479529332579,
+      "eval_loss": 0.7345991730690002,
+      "eval_mean_token_accuracy": 0.8013592945056015,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 89.817,
+      "eval_samples_per_second": 15.81,
+      "eval_steps_per_second": 1.982,
+      "step": 640
+    },
+    {
+      "entropy": 0.7303367637097835,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.1310476064682007,
+      "learning_rate": 0.00036089510669270666,
+      "loss": 0.6875617027282714,
+      "mean_token_accuracy": 0.8035397171974182,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7101308248016271,
+      "eval_loss": 0.7394917011260986,
+      "eval_mean_token_accuracy": 0.7990792942850777,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 89.7431,
+      "eval_samples_per_second": 15.823,
+      "eval_steps_per_second": 1.983,
+      "step": 660
+    },
+    {
+      "entropy": 0.7387678384780884,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.9645389914512634,
+      "learning_rate": 0.0003602446767678725,
+      "loss": 0.6911204338073731,
+      "mean_token_accuracy": 0.8056333385407924,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7281073737010527,
+      "eval_loss": 0.7376708984375,
+      "eval_mean_token_accuracy": 0.7997356415464637,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 90.0338,
+      "eval_samples_per_second": 15.772,
+      "eval_steps_per_second": 1.977,
+      "step": 680
+    },
+    {
+      "entropy": 0.7449554048478604,
+      "epoch": 1.687990343995172,
+      "grad_norm": 1.9069629907608032,
+      "learning_rate": 0.0003595439079030364,
+      "loss": 0.7042286396026611,
+      "mean_token_accuracy": 0.8018639378249646,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7022633817088738,
+      "eval_loss": 0.7320939898490906,
+      "eval_mean_token_accuracy": 0.8006051541044471,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 89.8948,
+      "eval_samples_per_second": 15.796,
+      "eval_steps_per_second": 1.98,
+      "step": 700
+    },
+    {
+      "entropy": 0.723249051719904,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.8333877325057983,
+      "learning_rate": 0.00035879299840749777,
+      "loss": 0.7043714046478271,
+      "mean_token_accuracy": 0.8028345607221127,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.672436946898364,
+      "eval_loss": 0.713233470916748,
+      "eval_mean_token_accuracy": 0.8047760577684038,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 89.6394,
+      "eval_samples_per_second": 15.841,
+      "eval_steps_per_second": 1.986,
+      "step": 720
+    },
+    {
+      "entropy": 0.730014206469059,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.4580037593841553,
+      "learning_rate": 0.00035799216077976137,
+      "loss": 0.6985176563262939,
+      "mean_token_accuracy": 0.8015547141432762,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.6908873795123582,
+      "eval_loss": 0.7198202013969421,
+      "eval_mean_token_accuracy": 0.8024131682481659,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 89.3771,
+      "eval_samples_per_second": 15.888,
+      "eval_steps_per_second": 1.992,
+      "step": 740
+    },
+    {
+      "entropy": 0.7219605796039105,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.51682710647583,
+      "learning_rate": 0.000357141621647403,
+      "loss": 0.6844424724578857,
+      "mean_token_accuracy": 0.8032813094556331,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7113401109582922,
+      "eval_loss": 0.7195309400558472,
+      "eval_mean_token_accuracy": 0.8019187035185568,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 89.4658,
+      "eval_samples_per_second": 15.872,
+      "eval_steps_per_second": 1.99,
+      "step": 760
+    },
+    {
+      "entropy": 0.729826743900776,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 1.0838000774383545,
+      "learning_rate": 0.0003562416217029361,
+      "loss": 0.685858678817749,
+      "mean_token_accuracy": 0.8050567395985126,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.649370267484965,
+      "eval_loss": 0.7177029252052307,
+      "eval_mean_token_accuracy": 0.8050174043419656,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 89.2287,
+      "eval_samples_per_second": 15.914,
+      "eval_steps_per_second": 1.995,
+      "step": 780
+    },
+    {
+      "entropy": 0.6799364153295755,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 1.1364924907684326,
+      "learning_rate": 0.00035529241563569903,
+      "loss": 0.6714428901672364,
+      "mean_token_accuracy": 0.8087341658771038,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6603799047095052,
+      "eval_loss": 0.7201809287071228,
+      "eval_mean_token_accuracy": 0.802863868769635,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 89.2718,
+      "eval_samples_per_second": 15.906,
+      "eval_steps_per_second": 1.994,
+      "step": 800
+    },
+    {
+      "entropy": 0.7078610375523567,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 1.2191942930221558,
+      "learning_rate": 0.0003542942720597808,
+      "loss": 0.6996590614318847,
+      "mean_token_accuracy": 0.8049512051045895,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.7240545264120852,
+      "eval_loss": 0.7301309108734131,
+      "eval_mean_token_accuracy": 0.8015973166133581,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 89.3438,
+      "eval_samples_per_second": 15.894,
+      "eval_steps_per_second": 1.992,
+      "step": 820
+    },
+    {
+      "entropy": 0.6557391556826505,
+      "epoch": 2.024140012070006,
+      "grad_norm": 1.36203134059906,
+      "learning_rate": 0.0003532474734380064,
+      "loss": 0.6327113628387451,
+      "mean_token_accuracy": 0.8142829847026181,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6573808648613062,
+      "eval_loss": 0.7334365844726562,
+      "eval_mean_token_accuracy": 0.8043633893634496,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 89.3359,
+      "eval_samples_per_second": 15.895,
+      "eval_steps_per_second": 1.992,
+      "step": 840
+    },
+    {
+      "entropy": 0.6409128502011299,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 1.5804955959320068,
+      "learning_rate": 0.0003521523160020035,
+      "loss": 0.6023559093475341,
+      "mean_token_accuracy": 0.8222251623868942,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6983234788594621,
+      "eval_loss": 0.7232135534286499,
+      "eval_mean_token_accuracy": 0.8010654268639811,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 89.2615,
+      "eval_samples_per_second": 15.908,
+      "eval_steps_per_second": 1.994,
+      "step": 860
+    },
+    {
+      "entropy": 0.6151149723678827,
+      "epoch": 2.12070006035003,
+      "grad_norm": 1.1975436210632324,
+      "learning_rate": 0.00035100910966837193,
+      "loss": 0.5803564548492431,
+      "mean_token_accuracy": 0.8256325736641884,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6636555573243773,
+      "eval_loss": 0.7205662727355957,
+      "eval_mean_token_accuracy": 0.8036768369460374,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 89.2957,
+      "eval_samples_per_second": 15.902,
+      "eval_steps_per_second": 1.993,
+      "step": 880
+    },
+    {
+      "entropy": 0.6201727617532015,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9907131195068359,
+      "learning_rate": 0.0003498181779509813,
+      "loss": 0.6101872444152832,
+      "mean_token_accuracy": 0.8196512959897518,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.6473788086617931,
+      "eval_loss": 0.7062397003173828,
+      "eval_mean_token_accuracy": 0.8078362184963869,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 88.6198,
+      "eval_samples_per_second": 16.024,
+      "eval_steps_per_second": 2.009,
+      "step": 900
+    },
+    {
+      "entropy": 0.6097332935780286,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.9055633544921875,
+      "learning_rate": 0.00034857985786942026,
+      "loss": 0.5956094264984131,
+      "mean_token_accuracy": 0.8240575887262821,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.612629868341296,
+      "eval_loss": 0.7106770277023315,
+      "eval_mean_token_accuracy": 0.805324277181304,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 89.4383,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 920
+    },
+    {
+      "entropy": 0.6182701248675585,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.9513736367225647,
+      "learning_rate": 0.00034729449985362404,
+      "loss": 0.6019269466400147,
+      "mean_token_accuracy": 0.8187106013298034,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6437647456533453,
+      "eval_loss": 0.7173502445220947,
+      "eval_mean_token_accuracy": 0.8056439573175451,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.9568,
+      "eval_samples_per_second": 15.963,
+      "eval_steps_per_second": 2.001,
+      "step": 940
+    },
+    {
+      "entropy": 0.6508570514619351,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.9606207013130188,
+      "learning_rate": 0.0003459624676447067,
+      "loss": 0.6296409606933594,
+      "mean_token_accuracy": 0.8167732007801533,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6288247135248077,
+      "eval_loss": 0.7160675525665283,
+      "eval_mean_token_accuracy": 0.8066010689467527,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 89.0044,
+      "eval_samples_per_second": 15.954,
+      "eval_steps_per_second": 2.0,
+      "step": 960
+    },
+    {
+      "entropy": 0.6380081418901682,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.9677181839942932,
+      "learning_rate": 0.000344584138192027,
+      "loss": 0.6210546493530273,
+      "mean_token_accuracy": 0.8186135642230511,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6407805031604981,
+      "eval_loss": 0.7055043578147888,
+      "eval_mean_token_accuracy": 0.8070103157772107,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 89.5606,
+      "eval_samples_per_second": 15.855,
+      "eval_steps_per_second": 1.987,
+      "step": 980
+    },
+    {
+      "entropy": 0.6310251638293266,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 2.1046407222747803,
+      "learning_rate": 0.000343159901546516,
+      "loss": 0.6165830135345459,
+      "mean_token_accuracy": 0.8207855716347694,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6461507470420237,
+      "eval_loss": 0.7292136549949646,
+      "eval_mean_token_accuracy": 0.8030418261383356,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 89.315,
+      "eval_samples_per_second": 15.899,
+      "eval_steps_per_second": 1.993,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6256617344915867,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 1.1768349409103394,
+      "learning_rate": 0.0003416901607502972,
+      "loss": 0.6142709255218506,
+      "mean_token_accuracy": 0.8220669947564602,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6150523993406403,
+      "eval_loss": 0.7019563317298889,
+      "eval_mean_token_accuracy": 0.8094323099998946,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 89.419,
+      "eval_samples_per_second": 15.88,
+      "eval_steps_per_second": 1.991,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6126735735684633,
+      "epoch": 2.506940253470127,
+      "grad_norm": 1.8065496683120728,
+      "learning_rate": 0.00034017533172263055,
+      "loss": 0.6056458473205566,
+      "mean_token_accuracy": 0.8203762136399746,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6544345456562685,
+      "eval_loss": 0.7086517214775085,
+      "eval_mean_token_accuracy": 0.806709805901131,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 88.981,
+      "eval_samples_per_second": 15.958,
+      "eval_steps_per_second": 2.0,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6311754353344441,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 1.0576369762420654,
+      "learning_rate": 0.00033861584314221236,
+      "loss": 0.6176392555236816,
+      "mean_token_accuracy": 0.8177410490810871,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6273799200406235,
+      "eval_loss": 0.708163321018219,
+      "eval_mean_token_accuracy": 0.8032017099053672,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 89.7668,
+      "eval_samples_per_second": 15.819,
+      "eval_steps_per_second": 1.983,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6200249589979648,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.997065007686615,
+      "learning_rate": 0.0003370121363258637,
+      "loss": 0.6203888893127442,
+      "mean_token_accuracy": 0.8196275025606156,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6457576811983344,
+      "eval_loss": 0.6972290277481079,
+      "eval_mean_token_accuracy": 0.8095980849158898,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 89.2027,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 1080
+    },
+    {
+      "entropy": 0.626810473203659,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.865475058555603,
+      "learning_rate": 0.0003353646651036438,
+      "loss": 0.6102585792541504,
+      "mean_token_accuracy": 0.819708751142025,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.6412022431914726,
+      "eval_loss": 0.7036139965057373,
+      "eval_mean_token_accuracy": 0.8070317135098275,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 89.4733,
+      "eval_samples_per_second": 15.871,
+      "eval_steps_per_second": 1.989,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6581619590520859,
+      "epoch": 2.700060350030175,
+      "grad_norm": 1.00307035446167,
+      "learning_rate": 0.00033367389569042064,
+      "loss": 0.6397720336914062,
+      "mean_token_accuracy": 0.81413309648633,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6288736402318719,
+      "eval_loss": 0.6966825723648071,
+      "eval_mean_token_accuracy": 0.8094039669867312,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 88.7341,
+      "eval_samples_per_second": 16.003,
+      "eval_steps_per_second": 2.006,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6399412982165813,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.8589329719543457,
+      "learning_rate": 0.0003319403065539384,
+      "loss": 0.6347338199615479,
+      "mean_token_accuracy": 0.8139534957706929,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.683491913455256,
+      "eval_loss": 0.6984149217605591,
+      "eval_mean_token_accuracy": 0.8083989469522841,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 89.3929,
+      "eval_samples_per_second": 15.885,
+      "eval_steps_per_second": 1.991,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6264835461974144,
+      "epoch": 2.796620398310199,
+      "grad_norm": 0.7807853817939758,
+      "learning_rate": 0.0003301643882794163,
+      "loss": 0.6162629127502441,
+      "mean_token_accuracy": 0.8216292977333068,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6578792159476977,
+      "eval_loss": 0.6921441555023193,
+      "eval_mean_token_accuracy": 0.8094299177775223,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 89.2356,
+      "eval_samples_per_second": 15.913,
+      "eval_steps_per_second": 1.995,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6269129011780024,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 1.1675593852996826,
+      "learning_rate": 0.0003283466434307189,
+      "loss": 0.6235532760620117,
+      "mean_token_accuracy": 0.8188897788524627,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6533248447970058,
+      "eval_loss": 0.696849524974823,
+      "eval_mean_token_accuracy": 0.8102703713968898,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2602,
+      "eval_samples_per_second": 16.089,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.6259873781353236,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.9596486687660217,
+      "learning_rate": 0.00032648758640813655,
+      "loss": 0.6118728160858155,
+      "mean_token_accuracy": 0.8171859130263328,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6212078754821521,
+      "eval_loss": 0.6942870020866394,
+      "eval_mean_token_accuracy": 0.810484343365337,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 89.4375,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 1200
+    },
+    {
+      "entropy": 0.6142537571489811,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 0.9625058770179749,
+      "learning_rate": 0.00032458774330281615,
+      "loss": 0.6060261249542236,
+      "mean_token_accuracy": 0.8223113484680653,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6553665667437436,
+      "eval_loss": 0.6835269331932068,
+      "eval_mean_token_accuracy": 0.8134723968720168,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 89.5092,
+      "eval_samples_per_second": 15.864,
+      "eval_steps_per_second": 1.989,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6301972426474094,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.9657333493232727,
+      "learning_rate": 0.0003226476517478835,
+      "loss": 0.6277759075164795,
+      "mean_token_accuracy": 0.8164266042411328,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6594253118788258,
+      "eval_loss": 0.6820935606956482,
+      "eval_mean_token_accuracy": 0.8132368198941263,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 89.6211,
+      "eval_samples_per_second": 15.844,
+      "eval_steps_per_second": 1.986,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5346100655469027,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8541185259819031,
+      "learning_rate": 0.0003206678607662996,
+      "loss": 0.5261692047119141,
+      "mean_token_accuracy": 0.8382289425119177,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5626645662476507,
+      "eval_loss": 0.7028542160987854,
+      "eval_mean_token_accuracy": 0.8133395572056931,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 89.1267,
+      "eval_samples_per_second": 15.932,
+      "eval_steps_per_second": 1.997,
+      "step": 1260
+    },
+    {
+      "entropy": 0.5001101832836866,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.8208815455436707,
+      "learning_rate": 0.0003186489306154935,
+      "loss": 0.4850132942199707,
+      "mean_token_accuracy": 0.8468983843922615,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5889462046743779,
+      "eval_loss": 0.691015899181366,
+      "eval_mean_token_accuracy": 0.8135226466012805,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.9677,
+      "eval_samples_per_second": 15.961,
+      "eval_steps_per_second": 2.001,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5242766339331866,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 0.7893072962760925,
+      "learning_rate": 0.0003165914326288163,
+      "loss": 0.5061577320098877,
+      "mean_token_accuracy": 0.8425608821213245,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5543691962957382,
+      "eval_loss": 0.6913734078407288,
+      "eval_mean_token_accuracy": 0.8154889680026622,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 89.6963,
+      "eval_samples_per_second": 15.831,
+      "eval_steps_per_second": 1.984,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5327721010893584,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.8992080092430115,
+      "learning_rate": 0.0003144959490538604,
+      "loss": 0.5068036556243897,
+      "mean_token_accuracy": 0.8420542575418949,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5734540803378887,
+      "eval_loss": 0.6948716044425964,
+      "eval_mean_token_accuracy": 0.8137217944257715,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 89.4597,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5247377116233111,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.9291845560073853,
+      "learning_rate": 0.0003123630728876902,
+      "loss": 0.5023272037506104,
+      "mean_token_accuracy": 0.8435182586312294,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5492362935891312,
+      "eval_loss": 0.6980633735656738,
+      "eval_mean_token_accuracy": 0.8155082919624415,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 89.2124,
+      "eval_samples_per_second": 15.917,
+      "eval_steps_per_second": 1.995,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5380499072372913,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 1.0477524995803833,
+      "learning_rate": 0.00031019340770903136,
+      "loss": 0.5264591693878173,
+      "mean_token_accuracy": 0.8391959741711617,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5896503307511297,
+      "eval_loss": 0.6997144222259521,
+      "eval_mean_token_accuracy": 0.8124373062942805,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 89.1147,
+      "eval_samples_per_second": 15.935,
+      "eval_steps_per_second": 1.997,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5460653610527515,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 1.2473069429397583,
+      "learning_rate": 0.00030798756750746476,
+      "loss": 0.5319677829742432,
+      "mean_token_accuracy": 0.8364320226013661,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5659052182114526,
+      "eval_loss": 0.6924049258232117,
+      "eval_mean_token_accuracy": 0.8153053585732921,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 89.6468,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5367407951503992,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 1.2679712772369385,
+      "learning_rate": 0.00030574617650967485,
+      "loss": 0.5190927028656006,
+      "mean_token_accuracy": 0.8399469174444676,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5581759144081159,
+      "eval_loss": 0.6933317184448242,
+      "eval_mean_token_accuracy": 0.8139757862251796,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 89.0749,
+      "eval_samples_per_second": 15.942,
+      "eval_steps_per_second": 1.998,
+      "step": 1400
+    },
+    {
+      "entropy": 0.530736118927598,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.9717612862586975,
+      "learning_rate": 0.0003034698690028009,
+      "loss": 0.512363052368164,
+      "mean_token_accuracy": 0.8419127956032753,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5233335836549823,
+      "eval_loss": 0.6958550214767456,
+      "eval_mean_token_accuracy": 0.8158778501360604,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 88.8664,
+      "eval_samples_per_second": 15.979,
+      "eval_steps_per_second": 2.003,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5327631575986743,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7649208903312683,
+      "learning_rate": 0.00030115928915494116,
+      "loss": 0.5202179431915284,
+      "mean_token_accuracy": 0.8384683802723885,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.5960972657364406,
+      "eval_loss": 0.6984783411026001,
+      "eval_mean_token_accuracy": 0.8095930597085631,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 89.5168,
+      "eval_samples_per_second": 15.863,
+      "eval_steps_per_second": 1.988,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5428155034780502,
+      "epoch": 3.51901025950513,
+      "grad_norm": 1.2790874242782593,
+      "learning_rate": 0.00029881509083286116,
+      "loss": 0.524717378616333,
+      "mean_token_accuracy": 0.8346644505858422,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.550318562917495,
+      "eval_loss": 0.6988471746444702,
+      "eval_mean_token_accuracy": 0.8132969897784544,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 89.2598,
+      "eval_samples_per_second": 15.909,
+      "eval_steps_per_second": 1.994,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5489993002265692,
+      "epoch": 3.567290283645142,
+      "grad_norm": 1.055821418762207,
+      "learning_rate": 0.00029643793741695687,
+      "loss": 0.5421150207519532,
+      "mean_token_accuracy": 0.8345710933208466,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5816821718818685,
+      "eval_loss": 0.6853668689727783,
+      "eval_mean_token_accuracy": 0.8142332360985574,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 89.0424,
+      "eval_samples_per_second": 15.947,
+      "eval_steps_per_second": 1.999,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5451365381479263,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.9027532935142517,
+      "learning_rate": 0.00029402850161352583,
+      "loss": 0.5463609218597412,
+      "mean_token_accuracy": 0.8369908526539802,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5344233209832331,
+      "eval_loss": 0.6794531345367432,
+      "eval_mean_token_accuracy": 0.8177759824843889,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 89.123,
+      "eval_samples_per_second": 15.933,
+      "eval_steps_per_second": 1.997,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5521467242389917,
+      "epoch": 3.663850331925166,
+      "grad_norm": 1.4606311321258545,
+      "learning_rate": 0.0002915874652643997,
+      "loss": 0.5421902179718018,
+      "mean_token_accuracy": 0.8339455373585224,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5807709094513668,
+      "eval_loss": 0.6949691772460938,
+      "eval_mean_token_accuracy": 0.8121141439743256,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.5437,
+      "eval_samples_per_second": 16.037,
+      "eval_steps_per_second": 2.01,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5455610822886229,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7269843220710754,
+      "learning_rate": 0.0002891155191539905,
+      "loss": 0.5407682895660401,
+      "mean_token_accuracy": 0.8393648222088814,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.6032205542151847,
+      "eval_loss": 0.6790974140167236,
+      "eval_mean_token_accuracy": 0.8165463112043531,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 89.517,
+      "eval_samples_per_second": 15.863,
+      "eval_steps_per_second": 1.988,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5474163968116045,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.8148866295814514,
+      "learning_rate": 0.00028661336281380717,
+      "loss": 0.5394441604614257,
+      "mean_token_accuracy": 0.8336028568446636,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5797032238392348,
+      "eval_loss": 0.670870304107666,
+      "eval_mean_token_accuracy": 0.8182691593518417,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 89.4468,
+      "eval_samples_per_second": 15.875,
+      "eval_steps_per_second": 1.99,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5511750865727663,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.3441656827926636,
+      "learning_rate": 0.0002840817043244964,
+      "loss": 0.5378933906555176,
+      "mean_token_accuracy": 0.8349151819944381,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5297858750217417,
+      "eval_loss": 0.680833637714386,
+      "eval_mean_token_accuracy": 0.8180924275617921,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 89.5417,
+      "eval_samples_per_second": 15.859,
+      "eval_steps_per_second": 1.988,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5379613988101483,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9813932776451111,
+      "learning_rate": 0.00028152126011546396,
+      "loss": 0.5340342044830322,
+      "mean_token_accuracy": 0.8346010789275169,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5811898510777549,
+      "eval_loss": 0.6764267683029175,
+      "eval_mean_token_accuracy": 0.8180528236239144,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 88.7831,
+      "eval_samples_per_second": 15.994,
+      "eval_steps_per_second": 2.005,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5607876226305961,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.9617928266525269,
+      "learning_rate": 0.00027893275476213383,
+      "loss": 0.5434338569641113,
+      "mean_token_accuracy": 0.8332898400723934,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5474644318390428,
+      "eval_loss": 0.6740226745605469,
+      "eval_mean_token_accuracy": 0.8185851306058047,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 89.3246,
+      "eval_samples_per_second": 15.897,
+      "eval_steps_per_second": 1.993,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5354843523353339,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.5320990085601807,
+      "learning_rate": 0.0002763169207809021,
+      "loss": 0.5251852989196777,
+      "mean_token_accuracy": 0.8416872084140777,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.5436856410141742,
+      "eval_loss": 0.6675190925598145,
+      "eval_mean_token_accuracy": 0.8202396507343549,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 89.6328,
+      "eval_samples_per_second": 15.842,
+      "eval_steps_per_second": 1.986,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5418571938167919,
+      "epoch": 4.0,
+      "grad_norm": 4.690354347229004,
+      "learning_rate": 0.000273674498421843,
+      "loss": 0.548275089263916,
+      "mean_token_accuracy": 0.836557479647847,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5497228983747825,
+      "eval_loss": 0.6621812582015991,
+      "eval_mean_token_accuracy": 0.8205679802412398,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 89.6272,
+      "eval_samples_per_second": 15.843,
+      "eval_steps_per_second": 1.986,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4195931971073151,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6678048968315125,
+      "learning_rate": 0.0002710062354592273,
+      "loss": 0.3970795154571533,
+      "mean_token_accuracy": 0.8719995342195034,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.46495268753405367,
+      "eval_loss": 0.7061581611633301,
+      "eval_mean_token_accuracy": 0.8190107995204712,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 89.6387,
+      "eval_samples_per_second": 15.841,
+      "eval_steps_per_second": 1.986,
+      "step": 1680
+    },
+    {
+      "entropy": 0.4263830740004778,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.7026669383049011,
+      "learning_rate": 0.000268312886979911,
+      "loss": 0.398905086517334,
+      "mean_token_accuracy": 0.8691012300550938,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.45755320670229666,
+      "eval_loss": 0.7102291584014893,
+      "eval_mean_token_accuracy": 0.819212955370378,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 89.2777,
+      "eval_samples_per_second": 15.905,
+      "eval_steps_per_second": 1.994,
+      "step": 1700
+    },
+    {
+      "entropy": 0.41640330031514167,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.752216100692749,
+      "learning_rate": 0.00026559521516965437,
+      "loss": 0.39748263359069824,
+      "mean_token_accuracy": 0.8684553548693656,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.469178682297803,
+      "eval_loss": 0.7004125714302063,
+      "eval_mean_token_accuracy": 0.8194386949030201,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 89.8428,
+      "eval_samples_per_second": 15.805,
+      "eval_steps_per_second": 1.981,
+      "step": 1720
+    },
+    {
+      "entropy": 0.42079071439802646,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.8349477648735046,
+      "learning_rate": 0.0002628539890974329,
+      "loss": 0.40129623413085935,
+      "mean_token_accuracy": 0.8686790131032467,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48358210871058904,
+      "eval_loss": 0.70196533203125,
+      "eval_mean_token_accuracy": 0.818096743875675,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 90.444,
+      "eval_samples_per_second": 15.7,
+      "eval_steps_per_second": 1.968,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4285690750926733,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.781974732875824,
+      "learning_rate": 0.00026008998449779933,
+      "loss": 0.40457854270935056,
+      "mean_token_accuracy": 0.8679066866636276,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.4653420230645812,
+      "eval_loss": 0.7041526436805725,
+      "eval_mean_token_accuracy": 0.8190121017815022,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 89.5248,
+      "eval_samples_per_second": 15.862,
+      "eval_steps_per_second": 1.988,
+      "step": 1760
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.762383734659584e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1780/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1780/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1780/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1780/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1780/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..7ce4b072fcbe5b636867e13937c56fedee3b68b4
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1780/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.02728394274790723,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1780/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1780/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1780/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1780/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1780/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1780/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1780/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1780/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..e38fe58ce735f34392ef88999f08a9043898ee91
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1780/trainer_state.json
@@ -0,0 +1,1903 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 4.289680144840072,
+  "eval_steps": 20,
+  "global_step": 1780,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.024795526266098,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.945375919342041,
+      "learning_rate": 1.6698127428345936e-05,
+      "loss": 1.7825420379638672,
+      "mean_token_accuracy": 0.6315708436071873,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.29921497588747,
+      "eval_loss": 1.169131875038147,
+      "eval_mean_token_accuracy": 0.7185706342204233,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 91.7898,
+      "eval_samples_per_second": 15.47,
+      "eval_steps_per_second": 1.939,
+      "step": 20
+    },
+    {
+      "entropy": 1.0346344470977784,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.7916979789733887,
+      "learning_rate": 3.427510366871008e-05,
+      "loss": 0.9480395317077637,
+      "mean_token_accuracy": 0.7543429024517536,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 0.9673695430326997,
+      "eval_loss": 0.8588430285453796,
+      "eval_mean_token_accuracy": 0.7737233002534073,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 89.8219,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 40
+    },
+    {
+      "entropy": 0.8977566614747048,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7972187995910645,
+      "learning_rate": 5.185207990907422e-05,
+      "loss": 0.8029604911804199,
+      "mean_token_accuracy": 0.7818016350269318,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8436836982041263,
+      "eval_loss": 0.7893033623695374,
+      "eval_mean_token_accuracy": 0.7858914984076211,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 89.9349,
+      "eval_samples_per_second": 15.789,
+      "eval_steps_per_second": 1.979,
+      "step": 60
+    },
+    {
+      "entropy": 0.8650321319699288,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.2128729820251465,
+      "learning_rate": 6.942905614943836e-05,
+      "loss": 0.77417893409729,
+      "mean_token_accuracy": 0.7875479467213153,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8209857180547179,
+      "eval_loss": 0.7572280168533325,
+      "eval_mean_token_accuracy": 0.7925970383574453,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 89.9459,
+      "eval_samples_per_second": 15.787,
+      "eval_steps_per_second": 1.979,
+      "step": 80
+    },
+    {
+      "entropy": 0.8096660938113928,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.1254287958145142,
+      "learning_rate": 8.700603238980251e-05,
+      "loss": 0.7170331001281738,
+      "mean_token_accuracy": 0.7994555257260799,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.792780208788561,
+      "eval_loss": 0.7442984580993652,
+      "eval_mean_token_accuracy": 0.7954704573984896,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 90.294,
+      "eval_samples_per_second": 15.726,
+      "eval_steps_per_second": 1.971,
+      "step": 100
+    },
+    {
+      "entropy": 0.7979829199612141,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2140507698059082,
+      "learning_rate": 0.00010458300863016666,
+      "loss": 0.7141861915588379,
+      "mean_token_accuracy": 0.8000014744699001,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.7938692486018277,
+      "eval_loss": 0.7279797196388245,
+      "eval_mean_token_accuracy": 0.7989798734027348,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 89.7189,
+      "eval_samples_per_second": 15.827,
+      "eval_steps_per_second": 1.984,
+      "step": 120
+    },
+    {
+      "entropy": 0.7916347607970238,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.0506458282470703,
+      "learning_rate": 0.0001221599848705308,
+      "loss": 0.7117055416107178,
+      "mean_token_accuracy": 0.7970046654343605,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.8052781063519167,
+      "eval_loss": 0.722576916217804,
+      "eval_mean_token_accuracy": 0.7982530523551984,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 90.0376,
+      "eval_samples_per_second": 15.771,
+      "eval_steps_per_second": 1.977,
+      "step": 140
+    },
+    {
+      "entropy": 0.7880666613578796,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8409207463264465,
+      "learning_rate": 0.00013973696111089492,
+      "loss": 0.7052523136138916,
+      "mean_token_accuracy": 0.8000123649835587,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7787443134891853,
+      "eval_loss": 0.7137772440910339,
+      "eval_mean_token_accuracy": 0.8014209287220173,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 90.2562,
+      "eval_samples_per_second": 15.733,
+      "eval_steps_per_second": 1.972,
+      "step": 160
+    },
+    {
+      "entropy": 0.7696007996797561,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.0182477235794067,
+      "learning_rate": 0.00015731393735125907,
+      "loss": 0.7008797645568847,
+      "mean_token_accuracy": 0.8007259473204613,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.752259682068664,
+      "eval_loss": 0.7171286344528198,
+      "eval_mean_token_accuracy": 0.80052122574174,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 90.0797,
+      "eval_samples_per_second": 15.764,
+      "eval_steps_per_second": 1.976,
+      "step": 180
+    },
+    {
+      "entropy": 0.7828126326203346,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9553632736206055,
+      "learning_rate": 0.0001748909135916232,
+      "loss": 0.7046853542327881,
+      "mean_token_accuracy": 0.7988562889397144,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7389638169427936,
+      "eval_loss": 0.7187935709953308,
+      "eval_mean_token_accuracy": 0.7995899999409579,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 89.9223,
+      "eval_samples_per_second": 15.791,
+      "eval_steps_per_second": 1.979,
+      "step": 200
+    },
+    {
+      "entropy": 0.7909286297857762,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 1.134459376335144,
+      "learning_rate": 0.00019246788983198735,
+      "loss": 0.6989312171936035,
+      "mean_token_accuracy": 0.7991565138101577,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7496965199374082,
+      "eval_loss": 0.7106321454048157,
+      "eval_mean_token_accuracy": 0.8046183442131857,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 89.6899,
+      "eval_samples_per_second": 15.832,
+      "eval_steps_per_second": 1.985,
+      "step": 220
+    },
+    {
+      "entropy": 0.7651963323354721,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.9557585716247559,
+      "learning_rate": 0.0002100448660723515,
+      "loss": 0.6984559059143066,
+      "mean_token_accuracy": 0.8019823372364044,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7615563079212488,
+      "eval_loss": 0.7098539471626282,
+      "eval_mean_token_accuracy": 0.8047233508544022,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 90.3752,
+      "eval_samples_per_second": 15.712,
+      "eval_steps_per_second": 1.97,
+      "step": 240
+    },
+    {
+      "entropy": 0.787507726252079,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.9279311299324036,
+      "learning_rate": 0.00022762184231271568,
+      "loss": 0.708138370513916,
+      "mean_token_accuracy": 0.8013308703899383,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7856640896100676,
+      "eval_loss": 0.7191570401191711,
+      "eval_mean_token_accuracy": 0.8024677249153008,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 93.7336,
+      "eval_samples_per_second": 15.149,
+      "eval_steps_per_second": 1.899,
+      "step": 260
+    },
+    {
+      "entropy": 0.768489234894514,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.7992776036262512,
+      "learning_rate": 0.0002451988185530798,
+      "loss": 0.7035849571228028,
+      "mean_token_accuracy": 0.8031105332076549,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7682498740346244,
+      "eval_loss": 0.7149022221565247,
+      "eval_mean_token_accuracy": 0.7973497995499814,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 89.7309,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 280
+    },
+    {
+      "entropy": 0.7720955617725849,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.1931514739990234,
+      "learning_rate": 0.0002627757947934439,
+      "loss": 0.6965402603149414,
+      "mean_token_accuracy": 0.8040348663926125,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7214708304807042,
+      "eval_loss": 0.7350823879241943,
+      "eval_mean_token_accuracy": 0.7942911925610532,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 89.6451,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 300
+    },
+    {
+      "entropy": 0.7930095426738262,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8994467258453369,
+      "learning_rate": 0.0002803527710338081,
+      "loss": 0.7331492900848389,
+      "mean_token_accuracy": 0.796185302734375,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.8142535488926963,
+      "eval_loss": 0.7237834334373474,
+      "eval_mean_token_accuracy": 0.799495273761535,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 90.6759,
+      "eval_samples_per_second": 15.66,
+      "eval_steps_per_second": 1.963,
+      "step": 320
+    },
+    {
+      "entropy": 0.7825102441012859,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 1.8224239349365234,
+      "learning_rate": 0.00029792974727417223,
+      "loss": 0.7331175804138184,
+      "mean_token_accuracy": 0.7977667152881622,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7327135443017724,
+      "eval_loss": 0.7388784289360046,
+      "eval_mean_token_accuracy": 0.8007041252730938,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 89.7692,
+      "eval_samples_per_second": 15.818,
+      "eval_steps_per_second": 1.983,
+      "step": 340
+    },
+    {
+      "entropy": 0.819331557303667,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 1.0407779216766357,
+      "learning_rate": 0.0003155067235145364,
+      "loss": 0.7456695556640625,
+      "mean_token_accuracy": 0.7936980701982975,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.8065585420372781,
+      "eval_loss": 0.735519528388977,
+      "eval_mean_token_accuracy": 0.7974206296245704,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 89.4336,
+      "eval_samples_per_second": 15.878,
+      "eval_steps_per_second": 1.99,
+      "step": 360
+    },
+    {
+      "entropy": 0.8114653021097183,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.1209490299224854,
+      "learning_rate": 0.0003330836997549005,
+      "loss": 0.7340899467468261,
+      "mean_token_accuracy": 0.7914272703230381,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.8025527285056168,
+      "eval_loss": 0.7295576930046082,
+      "eval_mean_token_accuracy": 0.7990576628218876,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 89.9579,
+      "eval_samples_per_second": 15.785,
+      "eval_steps_per_second": 1.979,
+      "step": 380
+    },
+    {
+      "entropy": 0.8042497783899307,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 1.2437798976898193,
+      "learning_rate": 0.0003506606759952647,
+      "loss": 0.7381744384765625,
+      "mean_token_accuracy": 0.7967362694442273,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7717011336530193,
+      "eval_loss": 0.7436173558235168,
+      "eval_mean_token_accuracy": 0.7975849482450592,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 89.8411,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 400
+    },
+    {
+      "entropy": 0.7987750010056929,
+      "epoch": 1.012070006035003,
+      "grad_norm": 1.2390918731689453,
+      "learning_rate": 0.000364721224843344,
+      "loss": 0.7317790031433106,
+      "mean_token_accuracy": 0.7986087551364651,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7333170604170038,
+      "eval_loss": 0.7456948757171631,
+      "eval_mean_token_accuracy": 0.7964897577682238,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 89.7288,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 420
+    },
+    {
+      "entropy": 0.7372976846992969,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.1552740335464478,
+      "learning_rate": 0.00036468510102269575,
+      "loss": 0.6734249114990234,
+      "mean_token_accuracy": 0.805834824591875,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.7130355526891987,
+      "eval_loss": 0.7555333375930786,
+      "eval_mean_token_accuracy": 0.7961652781186479,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 89.8713,
+      "eval_samples_per_second": 15.8,
+      "eval_steps_per_second": 1.981,
+      "step": 440
+    },
+    {
+      "entropy": 0.7465668715536594,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 1.1991469860076904,
+      "learning_rate": 0.0003645973816745066,
+      "loss": 0.6871879577636719,
+      "mean_token_accuracy": 0.8042450234293937,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7772979542110743,
+      "eval_loss": 0.7561826705932617,
+      "eval_mean_token_accuracy": 0.7953876427720102,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 89.8395,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 460
+    },
+    {
+      "entropy": 0.7477855801582336,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 1.325486660003662,
+      "learning_rate": 0.00036445809162231435,
+      "loss": 0.6836830139160156,
+      "mean_token_accuracy": 0.8021619468927383,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.7098110676481483,
+      "eval_loss": 0.758007287979126,
+      "eval_mean_token_accuracy": 0.7909953189030122,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 89.6101,
+      "eval_samples_per_second": 15.846,
+      "eval_steps_per_second": 1.986,
+      "step": 480
+    },
+    {
+      "entropy": 0.7632691070437432,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 1.0122262239456177,
+      "learning_rate": 0.0003642672702835562,
+      "loss": 0.7071213245391845,
+      "mean_token_accuracy": 0.8000680930912495,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7645382720432924,
+      "eval_loss": 0.7492606043815613,
+      "eval_mean_token_accuracy": 0.7976858676149604,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 89.6092,
+      "eval_samples_per_second": 15.847,
+      "eval_steps_per_second": 1.986,
+      "step": 500
+    },
+    {
+      "entropy": 0.7623726457357407,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.367774486541748,
+      "learning_rate": 0.00036402497165841384,
+      "loss": 0.6965175151824952,
+      "mean_token_accuracy": 0.8016501650214195,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7868083715438843,
+      "eval_loss": 0.7481449842453003,
+      "eval_mean_token_accuracy": 0.799330582779445,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 89.8241,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 520
+    },
+    {
+      "entropy": 0.7535207405686378,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 1.213614821434021,
+      "learning_rate": 0.00036373126431453207,
+      "loss": 0.7078310489654541,
+      "mean_token_accuracy": 0.7998907208442688,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7829091950748743,
+      "eval_loss": 0.7416072487831116,
+      "eval_mean_token_accuracy": 0.7969146250339036,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 89.4607,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 540
+    },
+    {
+      "entropy": 0.7527363449335098,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 1.0929982662200928,
+      "learning_rate": 0.00036338623136761495,
+      "loss": 0.7058369636535644,
+      "mean_token_accuracy": 0.7997087433934211,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.728419938114252,
+      "eval_loss": 0.744873046875,
+      "eval_mean_token_accuracy": 0.7966634296299366,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 89.6597,
+      "eval_samples_per_second": 15.838,
+      "eval_steps_per_second": 1.985,
+      "step": 560
+    },
+    {
+      "entropy": 0.7406099259853363,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.409364104270935,
+      "learning_rate": 0.00036298997045790515,
+      "loss": 0.6902852058410645,
+      "mean_token_accuracy": 0.805386807769537,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.6991632758231645,
+      "eval_loss": 0.7480319738388062,
+      "eval_mean_token_accuracy": 0.7970068006033308,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 89.8329,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 580
+    },
+    {
+      "entropy": 0.7459449704736472,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.7053111791610718,
+      "learning_rate": 0.00036254259372255277,
+      "loss": 0.7044640064239502,
+      "mean_token_accuracy": 0.8005018755793571,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7454566660891758,
+      "eval_loss": 0.7551484704017639,
+      "eval_mean_token_accuracy": 0.7897093731365846,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 89.8313,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 600
+    },
+    {
+      "entropy": 0.7621391348540782,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 1.174443244934082,
+      "learning_rate": 0.000362044227763882,
+      "loss": 0.7111573696136475,
+      "mean_token_accuracy": 0.798752411454916,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7664387949397055,
+      "eval_loss": 0.7365804314613342,
+      "eval_mean_token_accuracy": 0.7963101200843126,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 89.9623,
+      "eval_samples_per_second": 15.784,
+      "eval_steps_per_second": 1.979,
+      "step": 620
+    },
+    {
+      "entropy": 0.7424944408237935,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.2898184061050415,
+      "learning_rate": 0.000361495013613564,
+      "loss": 0.6913110733032226,
+      "mean_token_accuracy": 0.8009015507996082,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7255479529332579,
+      "eval_loss": 0.7345991730690002,
+      "eval_mean_token_accuracy": 0.8013592945056015,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 89.817,
+      "eval_samples_per_second": 15.81,
+      "eval_steps_per_second": 1.982,
+      "step": 640
+    },
+    {
+      "entropy": 0.7303367637097835,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.1310476064682007,
+      "learning_rate": 0.00036089510669270666,
+      "loss": 0.6875617027282714,
+      "mean_token_accuracy": 0.8035397171974182,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7101308248016271,
+      "eval_loss": 0.7394917011260986,
+      "eval_mean_token_accuracy": 0.7990792942850777,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 89.7431,
+      "eval_samples_per_second": 15.823,
+      "eval_steps_per_second": 1.983,
+      "step": 660
+    },
+    {
+      "entropy": 0.7387678384780884,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.9645389914512634,
+      "learning_rate": 0.0003602446767678725,
+      "loss": 0.6911204338073731,
+      "mean_token_accuracy": 0.8056333385407924,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7281073737010527,
+      "eval_loss": 0.7376708984375,
+      "eval_mean_token_accuracy": 0.7997356415464637,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 90.0338,
+      "eval_samples_per_second": 15.772,
+      "eval_steps_per_second": 1.977,
+      "step": 680
+    },
+    {
+      "entropy": 0.7449554048478604,
+      "epoch": 1.687990343995172,
+      "grad_norm": 1.9069629907608032,
+      "learning_rate": 0.0003595439079030364,
+      "loss": 0.7042286396026611,
+      "mean_token_accuracy": 0.8018639378249646,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7022633817088738,
+      "eval_loss": 0.7320939898490906,
+      "eval_mean_token_accuracy": 0.8006051541044471,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 89.8948,
+      "eval_samples_per_second": 15.796,
+      "eval_steps_per_second": 1.98,
+      "step": 700
+    },
+    {
+      "entropy": 0.723249051719904,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.8333877325057983,
+      "learning_rate": 0.00035879299840749777,
+      "loss": 0.7043714046478271,
+      "mean_token_accuracy": 0.8028345607221127,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.672436946898364,
+      "eval_loss": 0.713233470916748,
+      "eval_mean_token_accuracy": 0.8047760577684038,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 89.6394,
+      "eval_samples_per_second": 15.841,
+      "eval_steps_per_second": 1.986,
+      "step": 720
+    },
+    {
+      "entropy": 0.730014206469059,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.4580037593841553,
+      "learning_rate": 0.00035799216077976137,
+      "loss": 0.6985176563262939,
+      "mean_token_accuracy": 0.8015547141432762,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.6908873795123582,
+      "eval_loss": 0.7198202013969421,
+      "eval_mean_token_accuracy": 0.8024131682481659,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 89.3771,
+      "eval_samples_per_second": 15.888,
+      "eval_steps_per_second": 1.992,
+      "step": 740
+    },
+    {
+      "entropy": 0.7219605796039105,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.51682710647583,
+      "learning_rate": 0.000357141621647403,
+      "loss": 0.6844424724578857,
+      "mean_token_accuracy": 0.8032813094556331,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7113401109582922,
+      "eval_loss": 0.7195309400558472,
+      "eval_mean_token_accuracy": 0.8019187035185568,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 89.4658,
+      "eval_samples_per_second": 15.872,
+      "eval_steps_per_second": 1.99,
+      "step": 760
+    },
+    {
+      "entropy": 0.729826743900776,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 1.0838000774383545,
+      "learning_rate": 0.0003562416217029361,
+      "loss": 0.685858678817749,
+      "mean_token_accuracy": 0.8050567395985126,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.649370267484965,
+      "eval_loss": 0.7177029252052307,
+      "eval_mean_token_accuracy": 0.8050174043419656,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 89.2287,
+      "eval_samples_per_second": 15.914,
+      "eval_steps_per_second": 1.995,
+      "step": 780
+    },
+    {
+      "entropy": 0.6799364153295755,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 1.1364924907684326,
+      "learning_rate": 0.00035529241563569903,
+      "loss": 0.6714428901672364,
+      "mean_token_accuracy": 0.8087341658771038,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6603799047095052,
+      "eval_loss": 0.7201809287071228,
+      "eval_mean_token_accuracy": 0.802863868769635,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 89.2718,
+      "eval_samples_per_second": 15.906,
+      "eval_steps_per_second": 1.994,
+      "step": 800
+    },
+    {
+      "entropy": 0.7078610375523567,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 1.2191942930221558,
+      "learning_rate": 0.0003542942720597808,
+      "loss": 0.6996590614318847,
+      "mean_token_accuracy": 0.8049512051045895,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.7240545264120852,
+      "eval_loss": 0.7301309108734131,
+      "eval_mean_token_accuracy": 0.8015973166133581,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 89.3438,
+      "eval_samples_per_second": 15.894,
+      "eval_steps_per_second": 1.992,
+      "step": 820
+    },
+    {
+      "entropy": 0.6557391556826505,
+      "epoch": 2.024140012070006,
+      "grad_norm": 1.36203134059906,
+      "learning_rate": 0.0003532474734380064,
+      "loss": 0.6327113628387451,
+      "mean_token_accuracy": 0.8142829847026181,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6573808648613062,
+      "eval_loss": 0.7334365844726562,
+      "eval_mean_token_accuracy": 0.8043633893634496,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 89.3359,
+      "eval_samples_per_second": 15.895,
+      "eval_steps_per_second": 1.992,
+      "step": 840
+    },
+    {
+      "entropy": 0.6409128502011299,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 1.5804955959320068,
+      "learning_rate": 0.0003521523160020035,
+      "loss": 0.6023559093475341,
+      "mean_token_accuracy": 0.8222251623868942,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6983234788594621,
+      "eval_loss": 0.7232135534286499,
+      "eval_mean_token_accuracy": 0.8010654268639811,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 89.2615,
+      "eval_samples_per_second": 15.908,
+      "eval_steps_per_second": 1.994,
+      "step": 860
+    },
+    {
+      "entropy": 0.6151149723678827,
+      "epoch": 2.12070006035003,
+      "grad_norm": 1.1975436210632324,
+      "learning_rate": 0.00035100910966837193,
+      "loss": 0.5803564548492431,
+      "mean_token_accuracy": 0.8256325736641884,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6636555573243773,
+      "eval_loss": 0.7205662727355957,
+      "eval_mean_token_accuracy": 0.8036768369460374,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 89.2957,
+      "eval_samples_per_second": 15.902,
+      "eval_steps_per_second": 1.993,
+      "step": 880
+    },
+    {
+      "entropy": 0.6201727617532015,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9907131195068359,
+      "learning_rate": 0.0003498181779509813,
+      "loss": 0.6101872444152832,
+      "mean_token_accuracy": 0.8196512959897518,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.6473788086617931,
+      "eval_loss": 0.7062397003173828,
+      "eval_mean_token_accuracy": 0.8078362184963869,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 88.6198,
+      "eval_samples_per_second": 16.024,
+      "eval_steps_per_second": 2.009,
+      "step": 900
+    },
+    {
+      "entropy": 0.6097332935780286,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.9055633544921875,
+      "learning_rate": 0.00034857985786942026,
+      "loss": 0.5956094264984131,
+      "mean_token_accuracy": 0.8240575887262821,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.612629868341296,
+      "eval_loss": 0.7106770277023315,
+      "eval_mean_token_accuracy": 0.805324277181304,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 89.4383,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 920
+    },
+    {
+      "entropy": 0.6182701248675585,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.9513736367225647,
+      "learning_rate": 0.00034729449985362404,
+      "loss": 0.6019269466400147,
+      "mean_token_accuracy": 0.8187106013298034,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6437647456533453,
+      "eval_loss": 0.7173502445220947,
+      "eval_mean_token_accuracy": 0.8056439573175451,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.9568,
+      "eval_samples_per_second": 15.963,
+      "eval_steps_per_second": 2.001,
+      "step": 940
+    },
+    {
+      "entropy": 0.6508570514619351,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.9606207013130188,
+      "learning_rate": 0.0003459624676447067,
+      "loss": 0.6296409606933594,
+      "mean_token_accuracy": 0.8167732007801533,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6288247135248077,
+      "eval_loss": 0.7160675525665283,
+      "eval_mean_token_accuracy": 0.8066010689467527,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 89.0044,
+      "eval_samples_per_second": 15.954,
+      "eval_steps_per_second": 2.0,
+      "step": 960
+    },
+    {
+      "entropy": 0.6380081418901682,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.9677181839942932,
+      "learning_rate": 0.000344584138192027,
+      "loss": 0.6210546493530273,
+      "mean_token_accuracy": 0.8186135642230511,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6407805031604981,
+      "eval_loss": 0.7055043578147888,
+      "eval_mean_token_accuracy": 0.8070103157772107,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 89.5606,
+      "eval_samples_per_second": 15.855,
+      "eval_steps_per_second": 1.987,
+      "step": 980
+    },
+    {
+      "entropy": 0.6310251638293266,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 2.1046407222747803,
+      "learning_rate": 0.000343159901546516,
+      "loss": 0.6165830135345459,
+      "mean_token_accuracy": 0.8207855716347694,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6461507470420237,
+      "eval_loss": 0.7292136549949646,
+      "eval_mean_token_accuracy": 0.8030418261383356,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 89.315,
+      "eval_samples_per_second": 15.899,
+      "eval_steps_per_second": 1.993,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6256617344915867,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 1.1768349409103394,
+      "learning_rate": 0.0003416901607502972,
+      "loss": 0.6142709255218506,
+      "mean_token_accuracy": 0.8220669947564602,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6150523993406403,
+      "eval_loss": 0.7019563317298889,
+      "eval_mean_token_accuracy": 0.8094323099998946,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 89.419,
+      "eval_samples_per_second": 15.88,
+      "eval_steps_per_second": 1.991,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6126735735684633,
+      "epoch": 2.506940253470127,
+      "grad_norm": 1.8065496683120728,
+      "learning_rate": 0.00034017533172263055,
+      "loss": 0.6056458473205566,
+      "mean_token_accuracy": 0.8203762136399746,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6544345456562685,
+      "eval_loss": 0.7086517214775085,
+      "eval_mean_token_accuracy": 0.806709805901131,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 88.981,
+      "eval_samples_per_second": 15.958,
+      "eval_steps_per_second": 2.0,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6311754353344441,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 1.0576369762420654,
+      "learning_rate": 0.00033861584314221236,
+      "loss": 0.6176392555236816,
+      "mean_token_accuracy": 0.8177410490810871,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6273799200406235,
+      "eval_loss": 0.708163321018219,
+      "eval_mean_token_accuracy": 0.8032017099053672,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 89.7668,
+      "eval_samples_per_second": 15.819,
+      "eval_steps_per_second": 1.983,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6200249589979648,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.997065007686615,
+      "learning_rate": 0.0003370121363258637,
+      "loss": 0.6203888893127442,
+      "mean_token_accuracy": 0.8196275025606156,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6457576811983344,
+      "eval_loss": 0.6972290277481079,
+      "eval_mean_token_accuracy": 0.8095980849158898,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 89.2027,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 1080
+    },
+    {
+      "entropy": 0.626810473203659,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.865475058555603,
+      "learning_rate": 0.0003353646651036438,
+      "loss": 0.6102585792541504,
+      "mean_token_accuracy": 0.819708751142025,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.6412022431914726,
+      "eval_loss": 0.7036139965057373,
+      "eval_mean_token_accuracy": 0.8070317135098275,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 89.4733,
+      "eval_samples_per_second": 15.871,
+      "eval_steps_per_second": 1.989,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6581619590520859,
+      "epoch": 2.700060350030175,
+      "grad_norm": 1.00307035446167,
+      "learning_rate": 0.00033367389569042064,
+      "loss": 0.6397720336914062,
+      "mean_token_accuracy": 0.81413309648633,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6288736402318719,
+      "eval_loss": 0.6966825723648071,
+      "eval_mean_token_accuracy": 0.8094039669867312,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 88.7341,
+      "eval_samples_per_second": 16.003,
+      "eval_steps_per_second": 2.006,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6399412982165813,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.8589329719543457,
+      "learning_rate": 0.0003319403065539384,
+      "loss": 0.6347338199615479,
+      "mean_token_accuracy": 0.8139534957706929,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.683491913455256,
+      "eval_loss": 0.6984149217605591,
+      "eval_mean_token_accuracy": 0.8083989469522841,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 89.3929,
+      "eval_samples_per_second": 15.885,
+      "eval_steps_per_second": 1.991,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6264835461974144,
+      "epoch": 2.796620398310199,
+      "grad_norm": 0.7807853817939758,
+      "learning_rate": 0.0003301643882794163,
+      "loss": 0.6162629127502441,
+      "mean_token_accuracy": 0.8216292977333068,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6578792159476977,
+      "eval_loss": 0.6921441555023193,
+      "eval_mean_token_accuracy": 0.8094299177775223,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 89.2356,
+      "eval_samples_per_second": 15.913,
+      "eval_steps_per_second": 1.995,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6269129011780024,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 1.1675593852996826,
+      "learning_rate": 0.0003283466434307189,
+      "loss": 0.6235532760620117,
+      "mean_token_accuracy": 0.8188897788524627,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6533248447970058,
+      "eval_loss": 0.696849524974823,
+      "eval_mean_token_accuracy": 0.8102703713968898,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2602,
+      "eval_samples_per_second": 16.089,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.6259873781353236,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.9596486687660217,
+      "learning_rate": 0.00032648758640813655,
+      "loss": 0.6118728160858155,
+      "mean_token_accuracy": 0.8171859130263328,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6212078754821521,
+      "eval_loss": 0.6942870020866394,
+      "eval_mean_token_accuracy": 0.810484343365337,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 89.4375,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 1200
+    },
+    {
+      "entropy": 0.6142537571489811,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 0.9625058770179749,
+      "learning_rate": 0.00032458774330281615,
+      "loss": 0.6060261249542236,
+      "mean_token_accuracy": 0.8223113484680653,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6553665667437436,
+      "eval_loss": 0.6835269331932068,
+      "eval_mean_token_accuracy": 0.8134723968720168,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 89.5092,
+      "eval_samples_per_second": 15.864,
+      "eval_steps_per_second": 1.989,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6301972426474094,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.9657333493232727,
+      "learning_rate": 0.0003226476517478835,
+      "loss": 0.6277759075164795,
+      "mean_token_accuracy": 0.8164266042411328,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6594253118788258,
+      "eval_loss": 0.6820935606956482,
+      "eval_mean_token_accuracy": 0.8132368198941263,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 89.6211,
+      "eval_samples_per_second": 15.844,
+      "eval_steps_per_second": 1.986,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5346100655469027,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8541185259819031,
+      "learning_rate": 0.0003206678607662996,
+      "loss": 0.5261692047119141,
+      "mean_token_accuracy": 0.8382289425119177,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5626645662476507,
+      "eval_loss": 0.7028542160987854,
+      "eval_mean_token_accuracy": 0.8133395572056931,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 89.1267,
+      "eval_samples_per_second": 15.932,
+      "eval_steps_per_second": 1.997,
+      "step": 1260
+    },
+    {
+      "entropy": 0.5001101832836866,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.8208815455436707,
+      "learning_rate": 0.0003186489306154935,
+      "loss": 0.4850132942199707,
+      "mean_token_accuracy": 0.8468983843922615,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5889462046743779,
+      "eval_loss": 0.691015899181366,
+      "eval_mean_token_accuracy": 0.8135226466012805,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.9677,
+      "eval_samples_per_second": 15.961,
+      "eval_steps_per_second": 2.001,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5242766339331866,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 0.7893072962760925,
+      "learning_rate": 0.0003165914326288163,
+      "loss": 0.5061577320098877,
+      "mean_token_accuracy": 0.8425608821213245,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5543691962957382,
+      "eval_loss": 0.6913734078407288,
+      "eval_mean_token_accuracy": 0.8154889680026622,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 89.6963,
+      "eval_samples_per_second": 15.831,
+      "eval_steps_per_second": 1.984,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5327721010893584,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.8992080092430115,
+      "learning_rate": 0.0003144959490538604,
+      "loss": 0.5068036556243897,
+      "mean_token_accuracy": 0.8420542575418949,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5734540803378887,
+      "eval_loss": 0.6948716044425964,
+      "eval_mean_token_accuracy": 0.8137217944257715,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 89.4597,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5247377116233111,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.9291845560073853,
+      "learning_rate": 0.0003123630728876902,
+      "loss": 0.5023272037506104,
+      "mean_token_accuracy": 0.8435182586312294,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5492362935891312,
+      "eval_loss": 0.6980633735656738,
+      "eval_mean_token_accuracy": 0.8155082919624415,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 89.2124,
+      "eval_samples_per_second": 15.917,
+      "eval_steps_per_second": 1.995,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5380499072372913,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 1.0477524995803833,
+      "learning_rate": 0.00031019340770903136,
+      "loss": 0.5264591693878173,
+      "mean_token_accuracy": 0.8391959741711617,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5896503307511297,
+      "eval_loss": 0.6997144222259521,
+      "eval_mean_token_accuracy": 0.8124373062942805,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 89.1147,
+      "eval_samples_per_second": 15.935,
+      "eval_steps_per_second": 1.997,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5460653610527515,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 1.2473069429397583,
+      "learning_rate": 0.00030798756750746476,
+      "loss": 0.5319677829742432,
+      "mean_token_accuracy": 0.8364320226013661,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5659052182114526,
+      "eval_loss": 0.6924049258232117,
+      "eval_mean_token_accuracy": 0.8153053585732921,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 89.6468,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5367407951503992,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 1.2679712772369385,
+      "learning_rate": 0.00030574617650967485,
+      "loss": 0.5190927028656006,
+      "mean_token_accuracy": 0.8399469174444676,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5581759144081159,
+      "eval_loss": 0.6933317184448242,
+      "eval_mean_token_accuracy": 0.8139757862251796,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 89.0749,
+      "eval_samples_per_second": 15.942,
+      "eval_steps_per_second": 1.998,
+      "step": 1400
+    },
+    {
+      "entropy": 0.530736118927598,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.9717612862586975,
+      "learning_rate": 0.0003034698690028009,
+      "loss": 0.512363052368164,
+      "mean_token_accuracy": 0.8419127956032753,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5233335836549823,
+      "eval_loss": 0.6958550214767456,
+      "eval_mean_token_accuracy": 0.8158778501360604,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 88.8664,
+      "eval_samples_per_second": 15.979,
+      "eval_steps_per_second": 2.003,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5327631575986743,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7649208903312683,
+      "learning_rate": 0.00030115928915494116,
+      "loss": 0.5202179431915284,
+      "mean_token_accuracy": 0.8384683802723885,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.5960972657364406,
+      "eval_loss": 0.6984783411026001,
+      "eval_mean_token_accuracy": 0.8095930597085631,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 89.5168,
+      "eval_samples_per_second": 15.863,
+      "eval_steps_per_second": 1.988,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5428155034780502,
+      "epoch": 3.51901025950513,
+      "grad_norm": 1.2790874242782593,
+      "learning_rate": 0.00029881509083286116,
+      "loss": 0.524717378616333,
+      "mean_token_accuracy": 0.8346644505858422,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.550318562917495,
+      "eval_loss": 0.6988471746444702,
+      "eval_mean_token_accuracy": 0.8132969897784544,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 89.2598,
+      "eval_samples_per_second": 15.909,
+      "eval_steps_per_second": 1.994,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5489993002265692,
+      "epoch": 3.567290283645142,
+      "grad_norm": 1.055821418762207,
+      "learning_rate": 0.00029643793741695687,
+      "loss": 0.5421150207519532,
+      "mean_token_accuracy": 0.8345710933208466,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5816821718818685,
+      "eval_loss": 0.6853668689727783,
+      "eval_mean_token_accuracy": 0.8142332360985574,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 89.0424,
+      "eval_samples_per_second": 15.947,
+      "eval_steps_per_second": 1.999,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5451365381479263,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.9027532935142517,
+      "learning_rate": 0.00029402850161352583,
+      "loss": 0.5463609218597412,
+      "mean_token_accuracy": 0.8369908526539802,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5344233209832331,
+      "eval_loss": 0.6794531345367432,
+      "eval_mean_token_accuracy": 0.8177759824843889,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 89.123,
+      "eval_samples_per_second": 15.933,
+      "eval_steps_per_second": 1.997,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5521467242389917,
+      "epoch": 3.663850331925166,
+      "grad_norm": 1.4606311321258545,
+      "learning_rate": 0.0002915874652643997,
+      "loss": 0.5421902179718018,
+      "mean_token_accuracy": 0.8339455373585224,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5807709094513668,
+      "eval_loss": 0.6949691772460938,
+      "eval_mean_token_accuracy": 0.8121141439743256,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.5437,
+      "eval_samples_per_second": 16.037,
+      "eval_steps_per_second": 2.01,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5455610822886229,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7269843220710754,
+      "learning_rate": 0.0002891155191539905,
+      "loss": 0.5407682895660401,
+      "mean_token_accuracy": 0.8393648222088814,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.6032205542151847,
+      "eval_loss": 0.6790974140167236,
+      "eval_mean_token_accuracy": 0.8165463112043531,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 89.517,
+      "eval_samples_per_second": 15.863,
+      "eval_steps_per_second": 1.988,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5474163968116045,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.8148866295814514,
+      "learning_rate": 0.00028661336281380717,
+      "loss": 0.5394441604614257,
+      "mean_token_accuracy": 0.8336028568446636,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5797032238392348,
+      "eval_loss": 0.670870304107666,
+      "eval_mean_token_accuracy": 0.8182691593518417,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 89.4468,
+      "eval_samples_per_second": 15.875,
+      "eval_steps_per_second": 1.99,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5511750865727663,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.3441656827926636,
+      "learning_rate": 0.0002840817043244964,
+      "loss": 0.5378933906555176,
+      "mean_token_accuracy": 0.8349151819944381,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5297858750217417,
+      "eval_loss": 0.680833637714386,
+      "eval_mean_token_accuracy": 0.8180924275617921,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 89.5417,
+      "eval_samples_per_second": 15.859,
+      "eval_steps_per_second": 1.988,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5379613988101483,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9813932776451111,
+      "learning_rate": 0.00028152126011546396,
+      "loss": 0.5340342044830322,
+      "mean_token_accuracy": 0.8346010789275169,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5811898510777549,
+      "eval_loss": 0.6764267683029175,
+      "eval_mean_token_accuracy": 0.8180528236239144,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 88.7831,
+      "eval_samples_per_second": 15.994,
+      "eval_steps_per_second": 2.005,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5607876226305961,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.9617928266525269,
+      "learning_rate": 0.00027893275476213383,
+      "loss": 0.5434338569641113,
+      "mean_token_accuracy": 0.8332898400723934,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5474644318390428,
+      "eval_loss": 0.6740226745605469,
+      "eval_mean_token_accuracy": 0.8185851306058047,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 89.3246,
+      "eval_samples_per_second": 15.897,
+      "eval_steps_per_second": 1.993,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5354843523353339,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.5320990085601807,
+      "learning_rate": 0.0002763169207809021,
+      "loss": 0.5251852989196777,
+      "mean_token_accuracy": 0.8416872084140777,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.5436856410141742,
+      "eval_loss": 0.6675190925598145,
+      "eval_mean_token_accuracy": 0.8202396507343549,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 89.6328,
+      "eval_samples_per_second": 15.842,
+      "eval_steps_per_second": 1.986,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5418571938167919,
+      "epoch": 4.0,
+      "grad_norm": 4.690354347229004,
+      "learning_rate": 0.000273674498421843,
+      "loss": 0.548275089263916,
+      "mean_token_accuracy": 0.836557479647847,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5497228983747825,
+      "eval_loss": 0.6621812582015991,
+      "eval_mean_token_accuracy": 0.8205679802412398,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 89.6272,
+      "eval_samples_per_second": 15.843,
+      "eval_steps_per_second": 1.986,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4195931971073151,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6678048968315125,
+      "learning_rate": 0.0002710062354592273,
+      "loss": 0.3970795154571533,
+      "mean_token_accuracy": 0.8719995342195034,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.46495268753405367,
+      "eval_loss": 0.7061581611633301,
+      "eval_mean_token_accuracy": 0.8190107995204712,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 89.6387,
+      "eval_samples_per_second": 15.841,
+      "eval_steps_per_second": 1.986,
+      "step": 1680
+    },
+    {
+      "entropy": 0.4263830740004778,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.7026669383049011,
+      "learning_rate": 0.000268312886979911,
+      "loss": 0.398905086517334,
+      "mean_token_accuracy": 0.8691012300550938,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.45755320670229666,
+      "eval_loss": 0.7102291584014893,
+      "eval_mean_token_accuracy": 0.819212955370378,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 89.2777,
+      "eval_samples_per_second": 15.905,
+      "eval_steps_per_second": 1.994,
+      "step": 1700
+    },
+    {
+      "entropy": 0.41640330031514167,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.752216100692749,
+      "learning_rate": 0.00026559521516965437,
+      "loss": 0.39748263359069824,
+      "mean_token_accuracy": 0.8684553548693656,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.469178682297803,
+      "eval_loss": 0.7004125714302063,
+      "eval_mean_token_accuracy": 0.8194386949030201,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 89.8428,
+      "eval_samples_per_second": 15.805,
+      "eval_steps_per_second": 1.981,
+      "step": 1720
+    },
+    {
+      "entropy": 0.42079071439802646,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.8349477648735046,
+      "learning_rate": 0.0002628539890974329,
+      "loss": 0.40129623413085935,
+      "mean_token_accuracy": 0.8686790131032467,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48358210871058904,
+      "eval_loss": 0.70196533203125,
+      "eval_mean_token_accuracy": 0.818096743875675,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 90.444,
+      "eval_samples_per_second": 15.7,
+      "eval_steps_per_second": 1.968,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4285690750926733,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.781974732875824,
+      "learning_rate": 0.00026008998449779933,
+      "loss": 0.40457854270935056,
+      "mean_token_accuracy": 0.8679066866636276,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.4653420230645812,
+      "eval_loss": 0.7041526436805725,
+      "eval_mean_token_accuracy": 0.8190121017815022,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 89.5248,
+      "eval_samples_per_second": 15.862,
+      "eval_steps_per_second": 1.988,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4335357129573822,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.8383703827857971,
+      "learning_rate": 0.0002573039835513604,
+      "loss": 0.41876921653747556,
+      "mean_token_accuracy": 0.8663770146667957,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.5015502416350869,
+      "eval_loss": 0.6904259324073792,
+      "eval_mean_token_accuracy": 0.8200626497188311,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 89.7014,
+      "eval_samples_per_second": 15.83,
+      "eval_steps_per_second": 1.984,
+      "step": 1780
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.7824312093750272e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-180/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-180/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-180/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-180/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-180/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..7ce4b072fcbe5b636867e13937c56fedee3b68b4
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-180/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.02728394274790723,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-180/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-180/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-180/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-180/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-180/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-180/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-180/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-180/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..682e10525879325be7ac4fbf6d19b21dbab6235f
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-180/trainer_state.json
@@ -0,0 +1,223 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.43452021726010864,
+  "eval_steps": 20,
+  "global_step": 180,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.024795526266098,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.945375919342041,
+      "learning_rate": 1.6698127428345936e-05,
+      "loss": 1.7825420379638672,
+      "mean_token_accuracy": 0.6315708436071873,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.29921497588747,
+      "eval_loss": 1.169131875038147,
+      "eval_mean_token_accuracy": 0.7185706342204233,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 91.7898,
+      "eval_samples_per_second": 15.47,
+      "eval_steps_per_second": 1.939,
+      "step": 20
+    },
+    {
+      "entropy": 1.0346344470977784,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.7916979789733887,
+      "learning_rate": 3.427510366871008e-05,
+      "loss": 0.9480395317077637,
+      "mean_token_accuracy": 0.7543429024517536,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 0.9673695430326997,
+      "eval_loss": 0.8588430285453796,
+      "eval_mean_token_accuracy": 0.7737233002534073,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 89.8219,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 40
+    },
+    {
+      "entropy": 0.8977566614747048,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7972187995910645,
+      "learning_rate": 5.185207990907422e-05,
+      "loss": 0.8029604911804199,
+      "mean_token_accuracy": 0.7818016350269318,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8436836982041263,
+      "eval_loss": 0.7893033623695374,
+      "eval_mean_token_accuracy": 0.7858914984076211,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 89.9349,
+      "eval_samples_per_second": 15.789,
+      "eval_steps_per_second": 1.979,
+      "step": 60
+    },
+    {
+      "entropy": 0.8650321319699288,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.2128729820251465,
+      "learning_rate": 6.942905614943836e-05,
+      "loss": 0.77417893409729,
+      "mean_token_accuracy": 0.7875479467213153,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8209857180547179,
+      "eval_loss": 0.7572280168533325,
+      "eval_mean_token_accuracy": 0.7925970383574453,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 89.9459,
+      "eval_samples_per_second": 15.787,
+      "eval_steps_per_second": 1.979,
+      "step": 80
+    },
+    {
+      "entropy": 0.8096660938113928,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.1254287958145142,
+      "learning_rate": 8.700603238980251e-05,
+      "loss": 0.7170331001281738,
+      "mean_token_accuracy": 0.7994555257260799,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.792780208788561,
+      "eval_loss": 0.7442984580993652,
+      "eval_mean_token_accuracy": 0.7954704573984896,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 90.294,
+      "eval_samples_per_second": 15.726,
+      "eval_steps_per_second": 1.971,
+      "step": 100
+    },
+    {
+      "entropy": 0.7979829199612141,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2140507698059082,
+      "learning_rate": 0.00010458300863016666,
+      "loss": 0.7141861915588379,
+      "mean_token_accuracy": 0.8000014744699001,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.7938692486018277,
+      "eval_loss": 0.7279797196388245,
+      "eval_mean_token_accuracy": 0.7989798734027348,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 89.7189,
+      "eval_samples_per_second": 15.827,
+      "eval_steps_per_second": 1.984,
+      "step": 120
+    },
+    {
+      "entropy": 0.7916347607970238,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.0506458282470703,
+      "learning_rate": 0.0001221599848705308,
+      "loss": 0.7117055416107178,
+      "mean_token_accuracy": 0.7970046654343605,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.8052781063519167,
+      "eval_loss": 0.722576916217804,
+      "eval_mean_token_accuracy": 0.7982530523551984,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 90.0376,
+      "eval_samples_per_second": 15.771,
+      "eval_steps_per_second": 1.977,
+      "step": 140
+    },
+    {
+      "entropy": 0.7880666613578796,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8409207463264465,
+      "learning_rate": 0.00013973696111089492,
+      "loss": 0.7052523136138916,
+      "mean_token_accuracy": 0.8000123649835587,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7787443134891853,
+      "eval_loss": 0.7137772440910339,
+      "eval_mean_token_accuracy": 0.8014209287220173,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 90.2562,
+      "eval_samples_per_second": 15.733,
+      "eval_steps_per_second": 1.972,
+      "step": 160
+    },
+    {
+      "entropy": 0.7696007996797561,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.0182477235794067,
+      "learning_rate": 0.00015731393735125907,
+      "loss": 0.7008797645568847,
+      "mean_token_accuracy": 0.8007259473204613,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.752259682068664,
+      "eval_loss": 0.7171286344528198,
+      "eval_mean_token_accuracy": 0.80052122574174,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 90.0797,
+      "eval_samples_per_second": 15.764,
+      "eval_steps_per_second": 1.976,
+      "step": 180
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.842464801808384e+16,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1800/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1800/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1800/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1800/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1800/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..7ce4b072fcbe5b636867e13937c56fedee3b68b4
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1800/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.02728394274790723,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1800/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1800/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1800/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1800/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1800/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1800/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1800/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1800/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..4f3aa1833d1bf90351e208e3b574def2714cbdc5
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test1/checkpoint-1800/trainer_state.json
@@ -0,0 +1,1924 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 4.337960168980085,
+  "eval_steps": 20,
+  "global_step": 1800,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.024795526266098,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.945375919342041,
+      "learning_rate": 1.6698127428345936e-05,
+      "loss": 1.7825420379638672,
+      "mean_token_accuracy": 0.6315708436071873,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.29921497588747,
+      "eval_loss": 1.169131875038147,
+      "eval_mean_token_accuracy": 0.7185706342204233,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 91.7898,
+      "eval_samples_per_second": 15.47,
+      "eval_steps_per_second": 1.939,
+      "step": 20
+    },
+    {
+      "entropy": 1.0346344470977784,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.7916979789733887,
+      "learning_rate": 3.427510366871008e-05,
+      "loss": 0.9480395317077637,
+      "mean_token_accuracy": 0.7543429024517536,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 0.9673695430326997,
+      "eval_loss": 0.8588430285453796,
+      "eval_mean_token_accuracy": 0.7737233002534073,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 89.8219,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 40
+    },
+    {
+      "entropy": 0.8977566614747048,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7972187995910645,
+      "learning_rate": 5.185207990907422e-05,
+      "loss": 0.8029604911804199,
+      "mean_token_accuracy": 0.7818016350269318,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8436836982041263,
+      "eval_loss": 0.7893033623695374,
+      "eval_mean_token_accuracy": 0.7858914984076211,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 89.9349,
+      "eval_samples_per_second": 15.789,
+      "eval_steps_per_second": 1.979,
+      "step": 60
+    },
+    {
+      "entropy": 0.8650321319699288,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.2128729820251465,
+      "learning_rate": 6.942905614943836e-05,
+      "loss": 0.77417893409729,
+      "mean_token_accuracy": 0.7875479467213153,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8209857180547179,
+      "eval_loss": 0.7572280168533325,
+      "eval_mean_token_accuracy": 0.7925970383574453,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 89.9459,
+      "eval_samples_per_second": 15.787,
+      "eval_steps_per_second": 1.979,
+      "step": 80
+    },
+    {
+      "entropy": 0.8096660938113928,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.1254287958145142,
+      "learning_rate": 8.700603238980251e-05,
+      "loss": 0.7170331001281738,
+      "mean_token_accuracy": 0.7994555257260799,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.792780208788561,
+      "eval_loss": 0.7442984580993652,
+      "eval_mean_token_accuracy": 0.7954704573984896,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 90.294,
+      "eval_samples_per_second": 15.726,
+      "eval_steps_per_second": 1.971,
+      "step": 100
+    },
+    {
+      "entropy": 0.7979829199612141,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2140507698059082,
+      "learning_rate": 0.00010458300863016666,
+      "loss": 0.7141861915588379,
+      "mean_token_accuracy": 0.8000014744699001,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.7938692486018277,
+      "eval_loss": 0.7279797196388245,
+      "eval_mean_token_accuracy": 0.7989798734027348,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 89.7189,
+      "eval_samples_per_second": 15.827,
+      "eval_steps_per_second": 1.984,
+      "step": 120
+    },
+    {
+      "entropy": 0.7916347607970238,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.0506458282470703,
+      "learning_rate": 0.0001221599848705308,
+      "loss": 0.7117055416107178,
+      "mean_token_accuracy": 0.7970046654343605,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.8052781063519167,
+      "eval_loss": 0.722576916217804,
+      "eval_mean_token_accuracy": 0.7982530523551984,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 90.0376,
+      "eval_samples_per_second": 15.771,
+      "eval_steps_per_second": 1.977,
+      "step": 140
+    },
+    {
+      "entropy": 0.7880666613578796,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8409207463264465,
+      "learning_rate": 0.00013973696111089492,
+      "loss": 0.7052523136138916,
+      "mean_token_accuracy": 0.8000123649835587,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7787443134891853,
+      "eval_loss": 0.7137772440910339,
+      "eval_mean_token_accuracy": 0.8014209287220173,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 90.2562,
+      "eval_samples_per_second": 15.733,
+      "eval_steps_per_second": 1.972,
+      "step": 160
+    },
+    {
+      "entropy": 0.7696007996797561,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.0182477235794067,
+      "learning_rate": 0.00015731393735125907,
+      "loss": 0.7008797645568847,
+      "mean_token_accuracy": 0.8007259473204613,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.752259682068664,
+      "eval_loss": 0.7171286344528198,
+      "eval_mean_token_accuracy": 0.80052122574174,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 90.0797,
+      "eval_samples_per_second": 15.764,
+      "eval_steps_per_second": 1.976,
+      "step": 180
+    },
+    {
+      "entropy": 0.7828126326203346,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9553632736206055,
+      "learning_rate": 0.0001748909135916232,
+      "loss": 0.7046853542327881,
+      "mean_token_accuracy": 0.7988562889397144,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7389638169427936,
+      "eval_loss": 0.7187935709953308,
+      "eval_mean_token_accuracy": 0.7995899999409579,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 89.9223,
+      "eval_samples_per_second": 15.791,
+      "eval_steps_per_second": 1.979,
+      "step": 200
+    },
+    {
+      "entropy": 0.7909286297857762,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 1.134459376335144,
+      "learning_rate": 0.00019246788983198735,
+      "loss": 0.6989312171936035,
+      "mean_token_accuracy": 0.7991565138101577,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7496965199374082,
+      "eval_loss": 0.7106321454048157,
+      "eval_mean_token_accuracy": 0.8046183442131857,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 89.6899,
+      "eval_samples_per_second": 15.832,
+      "eval_steps_per_second": 1.985,
+      "step": 220
+    },
+    {
+      "entropy": 0.7651963323354721,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.9557585716247559,
+      "learning_rate": 0.0002100448660723515,
+      "loss": 0.6984559059143066,
+      "mean_token_accuracy": 0.8019823372364044,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7615563079212488,
+      "eval_loss": 0.7098539471626282,
+      "eval_mean_token_accuracy": 0.8047233508544022,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 90.3752,
+      "eval_samples_per_second": 15.712,
+      "eval_steps_per_second": 1.97,
+      "step": 240
+    },
+    {
+      "entropy": 0.787507726252079,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.9279311299324036,
+      "learning_rate": 0.00022762184231271568,
+      "loss": 0.708138370513916,
+      "mean_token_accuracy": 0.8013308703899383,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7856640896100676,
+      "eval_loss": 0.7191570401191711,
+      "eval_mean_token_accuracy": 0.8024677249153008,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 93.7336,
+      "eval_samples_per_second": 15.149,
+      "eval_steps_per_second": 1.899,
+      "step": 260
+    },
+    {
+      "entropy": 0.768489234894514,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.7992776036262512,
+      "learning_rate": 0.0002451988185530798,
+      "loss": 0.7035849571228028,
+      "mean_token_accuracy": 0.8031105332076549,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7682498740346244,
+      "eval_loss": 0.7149022221565247,
+      "eval_mean_token_accuracy": 0.7973497995499814,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 89.7309,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 280
+    },
+    {
+      "entropy": 0.7720955617725849,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.1931514739990234,
+      "learning_rate": 0.0002627757947934439,
+      "loss": 0.6965402603149414,
+      "mean_token_accuracy": 0.8040348663926125,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7214708304807042,
+      "eval_loss": 0.7350823879241943,
+      "eval_mean_token_accuracy": 0.7942911925610532,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 89.6451,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 300
+    },
+    {
+      "entropy": 0.7930095426738262,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8994467258453369,
+      "learning_rate": 0.0002803527710338081,
+      "loss": 0.7331492900848389,
+      "mean_token_accuracy": 0.796185302734375,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.8142535488926963,
+      "eval_loss": 0.7237834334373474,
+      "eval_mean_token_accuracy": 0.799495273761535,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 90.6759,
+      "eval_samples_per_second": 15.66,
+      "eval_steps_per_second": 1.963,
+      "step": 320
+    },
+    {
+      "entropy": 0.7825102441012859,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 1.8224239349365234,
+      "learning_rate": 0.00029792974727417223,
+      "loss": 0.7331175804138184,
+      "mean_token_accuracy": 0.7977667152881622,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7327135443017724,
+      "eval_loss": 0.7388784289360046,
+      "eval_mean_token_accuracy": 0.8007041252730938,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 89.7692,
+      "eval_samples_per_second": 15.818,
+      "eval_steps_per_second": 1.983,
+      "step": 340
+    },
+    {
+      "entropy": 0.819331557303667,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 1.0407779216766357,
+      "learning_rate": 0.0003155067235145364,
+      "loss": 0.7456695556640625,
+      "mean_token_accuracy": 0.7936980701982975,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.8065585420372781,
+      "eval_loss": 0.735519528388977,
+      "eval_mean_token_accuracy": 0.7974206296245704,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 89.4336,
+      "eval_samples_per_second": 15.878,
+      "eval_steps_per_second": 1.99,
+      "step": 360
+    },
+    {
+      "entropy": 0.8114653021097183,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.1209490299224854,
+      "learning_rate": 0.0003330836997549005,
+      "loss": 0.7340899467468261,
+      "mean_token_accuracy": 0.7914272703230381,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.8025527285056168,
+      "eval_loss": 0.7295576930046082,
+      "eval_mean_token_accuracy": 0.7990576628218876,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 89.9579,
+      "eval_samples_per_second": 15.785,
+      "eval_steps_per_second": 1.979,
+      "step": 380
+    },
+    {
+      "entropy": 0.8042497783899307,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 1.2437798976898193,
+      "learning_rate": 0.0003506606759952647,
+      "loss": 0.7381744384765625,
+      "mean_token_accuracy": 0.7967362694442273,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7717011336530193,
+      "eval_loss": 0.7436173558235168,
+      "eval_mean_token_accuracy": 0.7975849482450592,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 89.8411,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 400
+    },
+    {
+      "entropy": 0.7987750010056929,
+      "epoch": 1.012070006035003,
+      "grad_norm": 1.2390918731689453,
+      "learning_rate": 0.000364721224843344,
+      "loss": 0.7317790031433106,
+      "mean_token_accuracy": 0.7986087551364651,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7333170604170038,
+      "eval_loss": 0.7456948757171631,
+      "eval_mean_token_accuracy": 0.7964897577682238,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 89.7288,
+      "eval_samples_per_second": 15.825,
+      "eval_steps_per_second": 1.984,
+      "step": 420
+    },
+    {
+      "entropy": 0.7372976846992969,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.1552740335464478,
+      "learning_rate": 0.00036468510102269575,
+      "loss": 0.6734249114990234,
+      "mean_token_accuracy": 0.805834824591875,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.7130355526891987,
+      "eval_loss": 0.7555333375930786,
+      "eval_mean_token_accuracy": 0.7961652781186479,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 89.8713,
+      "eval_samples_per_second": 15.8,
+      "eval_steps_per_second": 1.981,
+      "step": 440
+    },
+    {
+      "entropy": 0.7465668715536594,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 1.1991469860076904,
+      "learning_rate": 0.0003645973816745066,
+      "loss": 0.6871879577636719,
+      "mean_token_accuracy": 0.8042450234293937,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7772979542110743,
+      "eval_loss": 0.7561826705932617,
+      "eval_mean_token_accuracy": 0.7953876427720102,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 89.8395,
+      "eval_samples_per_second": 15.806,
+      "eval_steps_per_second": 1.981,
+      "step": 460
+    },
+    {
+      "entropy": 0.7477855801582336,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 1.325486660003662,
+      "learning_rate": 0.00036445809162231435,
+      "loss": 0.6836830139160156,
+      "mean_token_accuracy": 0.8021619468927383,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.7098110676481483,
+      "eval_loss": 0.758007287979126,
+      "eval_mean_token_accuracy": 0.7909953189030122,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 89.6101,
+      "eval_samples_per_second": 15.846,
+      "eval_steps_per_second": 1.986,
+      "step": 480
+    },
+    {
+      "entropy": 0.7632691070437432,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 1.0122262239456177,
+      "learning_rate": 0.0003642672702835562,
+      "loss": 0.7071213245391845,
+      "mean_token_accuracy": 0.8000680930912495,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7645382720432924,
+      "eval_loss": 0.7492606043815613,
+      "eval_mean_token_accuracy": 0.7976858676149604,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 89.6092,
+      "eval_samples_per_second": 15.847,
+      "eval_steps_per_second": 1.986,
+      "step": 500
+    },
+    {
+      "entropy": 0.7623726457357407,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.367774486541748,
+      "learning_rate": 0.00036402497165841384,
+      "loss": 0.6965175151824952,
+      "mean_token_accuracy": 0.8016501650214195,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7868083715438843,
+      "eval_loss": 0.7481449842453003,
+      "eval_mean_token_accuracy": 0.799330582779445,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 89.8241,
+      "eval_samples_per_second": 15.809,
+      "eval_steps_per_second": 1.982,
+      "step": 520
+    },
+    {
+      "entropy": 0.7535207405686378,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 1.213614821434021,
+      "learning_rate": 0.00036373126431453207,
+      "loss": 0.7078310489654541,
+      "mean_token_accuracy": 0.7998907208442688,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7829091950748743,
+      "eval_loss": 0.7416072487831116,
+      "eval_mean_token_accuracy": 0.7969146250339036,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 89.4607,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 540
+    },
+    {
+      "entropy": 0.7527363449335098,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 1.0929982662200928,
+      "learning_rate": 0.00036338623136761495,
+      "loss": 0.7058369636535644,
+      "mean_token_accuracy": 0.7997087433934211,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.728419938114252,
+      "eval_loss": 0.744873046875,
+      "eval_mean_token_accuracy": 0.7966634296299366,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 89.6597,
+      "eval_samples_per_second": 15.838,
+      "eval_steps_per_second": 1.985,
+      "step": 560
+    },
+    {
+      "entropy": 0.7406099259853363,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.409364104270935,
+      "learning_rate": 0.00036298997045790515,
+      "loss": 0.6902852058410645,
+      "mean_token_accuracy": 0.805386807769537,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.6991632758231645,
+      "eval_loss": 0.7480319738388062,
+      "eval_mean_token_accuracy": 0.7970068006033308,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 89.8329,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 580
+    },
+    {
+      "entropy": 0.7459449704736472,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.7053111791610718,
+      "learning_rate": 0.00036254259372255277,
+      "loss": 0.7044640064239502,
+      "mean_token_accuracy": 0.8005018755793571,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7454566660891758,
+      "eval_loss": 0.7551484704017639,
+      "eval_mean_token_accuracy": 0.7897093731365846,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 89.8313,
+      "eval_samples_per_second": 15.807,
+      "eval_steps_per_second": 1.981,
+      "step": 600
+    },
+    {
+      "entropy": 0.7621391348540782,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 1.174443244934082,
+      "learning_rate": 0.000362044227763882,
+      "loss": 0.7111573696136475,
+      "mean_token_accuracy": 0.798752411454916,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7664387949397055,
+      "eval_loss": 0.7365804314613342,
+      "eval_mean_token_accuracy": 0.7963101200843126,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 89.9623,
+      "eval_samples_per_second": 15.784,
+      "eval_steps_per_second": 1.979,
+      "step": 620
+    },
+    {
+      "entropy": 0.7424944408237935,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.2898184061050415,
+      "learning_rate": 0.000361495013613564,
+      "loss": 0.6913110733032226,
+      "mean_token_accuracy": 0.8009015507996082,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7255479529332579,
+      "eval_loss": 0.7345991730690002,
+      "eval_mean_token_accuracy": 0.8013592945056015,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 89.817,
+      "eval_samples_per_second": 15.81,
+      "eval_steps_per_second": 1.982,
+      "step": 640
+    },
+    {
+      "entropy": 0.7303367637097835,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.1310476064682007,
+      "learning_rate": 0.00036089510669270666,
+      "loss": 0.6875617027282714,
+      "mean_token_accuracy": 0.8035397171974182,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7101308248016271,
+      "eval_loss": 0.7394917011260986,
+      "eval_mean_token_accuracy": 0.7990792942850777,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 89.7431,
+      "eval_samples_per_second": 15.823,
+      "eval_steps_per_second": 1.983,
+      "step": 660
+    },
+    {
+      "entropy": 0.7387678384780884,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.9645389914512634,
+      "learning_rate": 0.0003602446767678725,
+      "loss": 0.6911204338073731,
+      "mean_token_accuracy": 0.8056333385407924,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7281073737010527,
+      "eval_loss": 0.7376708984375,
+      "eval_mean_token_accuracy": 0.7997356415464637,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 90.0338,
+      "eval_samples_per_second": 15.772,
+      "eval_steps_per_second": 1.977,
+      "step": 680
+    },
+    {
+      "entropy": 0.7449554048478604,
+      "epoch": 1.687990343995172,
+      "grad_norm": 1.9069629907608032,
+      "learning_rate": 0.0003595439079030364,
+      "loss": 0.7042286396026611,
+      "mean_token_accuracy": 0.8018639378249646,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7022633817088738,
+      "eval_loss": 0.7320939898490906,
+      "eval_mean_token_accuracy": 0.8006051541044471,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 89.8948,
+      "eval_samples_per_second": 15.796,
+      "eval_steps_per_second": 1.98,
+      "step": 700
+    },
+    {
+      "entropy": 0.723249051719904,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.8333877325057983,
+      "learning_rate": 0.00035879299840749777,
+      "loss": 0.7043714046478271,
+      "mean_token_accuracy": 0.8028345607221127,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.672436946898364,
+      "eval_loss": 0.713233470916748,
+      "eval_mean_token_accuracy": 0.8047760577684038,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 89.6394,
+      "eval_samples_per_second": 15.841,
+      "eval_steps_per_second": 1.986,
+      "step": 720
+    },
+    {
+      "entropy": 0.730014206469059,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.4580037593841553,
+      "learning_rate": 0.00035799216077976137,
+      "loss": 0.6985176563262939,
+      "mean_token_accuracy": 0.8015547141432762,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.6908873795123582,
+      "eval_loss": 0.7198202013969421,
+      "eval_mean_token_accuracy": 0.8024131682481659,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 89.3771,
+      "eval_samples_per_second": 15.888,
+      "eval_steps_per_second": 1.992,
+      "step": 740
+    },
+    {
+      "entropy": 0.7219605796039105,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.51682710647583,
+      "learning_rate": 0.000357141621647403,
+      "loss": 0.6844424724578857,
+      "mean_token_accuracy": 0.8032813094556331,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7113401109582922,
+      "eval_loss": 0.7195309400558472,
+      "eval_mean_token_accuracy": 0.8019187035185568,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 89.4658,
+      "eval_samples_per_second": 15.872,
+      "eval_steps_per_second": 1.99,
+      "step": 760
+    },
+    {
+      "entropy": 0.729826743900776,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 1.0838000774383545,
+      "learning_rate": 0.0003562416217029361,
+      "loss": 0.685858678817749,
+      "mean_token_accuracy": 0.8050567395985126,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.649370267484965,
+      "eval_loss": 0.7177029252052307,
+      "eval_mean_token_accuracy": 0.8050174043419656,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 89.2287,
+      "eval_samples_per_second": 15.914,
+      "eval_steps_per_second": 1.995,
+      "step": 780
+    },
+    {
+      "entropy": 0.6799364153295755,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 1.1364924907684326,
+      "learning_rate": 0.00035529241563569903,
+      "loss": 0.6714428901672364,
+      "mean_token_accuracy": 0.8087341658771038,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6603799047095052,
+      "eval_loss": 0.7201809287071228,
+      "eval_mean_token_accuracy": 0.802863868769635,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 89.2718,
+      "eval_samples_per_second": 15.906,
+      "eval_steps_per_second": 1.994,
+      "step": 800
+    },
+    {
+      "entropy": 0.7078610375523567,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 1.2191942930221558,
+      "learning_rate": 0.0003542942720597808,
+      "loss": 0.6996590614318847,
+      "mean_token_accuracy": 0.8049512051045895,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.7240545264120852,
+      "eval_loss": 0.7301309108734131,
+      "eval_mean_token_accuracy": 0.8015973166133581,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 89.3438,
+      "eval_samples_per_second": 15.894,
+      "eval_steps_per_second": 1.992,
+      "step": 820
+    },
+    {
+      "entropy": 0.6557391556826505,
+      "epoch": 2.024140012070006,
+      "grad_norm": 1.36203134059906,
+      "learning_rate": 0.0003532474734380064,
+      "loss": 0.6327113628387451,
+      "mean_token_accuracy": 0.8142829847026181,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6573808648613062,
+      "eval_loss": 0.7334365844726562,
+      "eval_mean_token_accuracy": 0.8043633893634496,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 89.3359,
+      "eval_samples_per_second": 15.895,
+      "eval_steps_per_second": 1.992,
+      "step": 840
+    },
+    {
+      "entropy": 0.6409128502011299,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 1.5804955959320068,
+      "learning_rate": 0.0003521523160020035,
+      "loss": 0.6023559093475341,
+      "mean_token_accuracy": 0.8222251623868942,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6983234788594621,
+      "eval_loss": 0.7232135534286499,
+      "eval_mean_token_accuracy": 0.8010654268639811,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 89.2615,
+      "eval_samples_per_second": 15.908,
+      "eval_steps_per_second": 1.994,
+      "step": 860
+    },
+    {
+      "entropy": 0.6151149723678827,
+      "epoch": 2.12070006035003,
+      "grad_norm": 1.1975436210632324,
+      "learning_rate": 0.00035100910966837193,
+      "loss": 0.5803564548492431,
+      "mean_token_accuracy": 0.8256325736641884,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6636555573243773,
+      "eval_loss": 0.7205662727355957,
+      "eval_mean_token_accuracy": 0.8036768369460374,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 89.2957,
+      "eval_samples_per_second": 15.902,
+      "eval_steps_per_second": 1.993,
+      "step": 880
+    },
+    {
+      "entropy": 0.6201727617532015,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9907131195068359,
+      "learning_rate": 0.0003498181779509813,
+      "loss": 0.6101872444152832,
+      "mean_token_accuracy": 0.8196512959897518,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.6473788086617931,
+      "eval_loss": 0.7062397003173828,
+      "eval_mean_token_accuracy": 0.8078362184963869,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 88.6198,
+      "eval_samples_per_second": 16.024,
+      "eval_steps_per_second": 2.009,
+      "step": 900
+    },
+    {
+      "entropy": 0.6097332935780286,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.9055633544921875,
+      "learning_rate": 0.00034857985786942026,
+      "loss": 0.5956094264984131,
+      "mean_token_accuracy": 0.8240575887262821,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.612629868341296,
+      "eval_loss": 0.7106770277023315,
+      "eval_mean_token_accuracy": 0.805324277181304,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 89.4383,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 920
+    },
+    {
+      "entropy": 0.6182701248675585,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.9513736367225647,
+      "learning_rate": 0.00034729449985362404,
+      "loss": 0.6019269466400147,
+      "mean_token_accuracy": 0.8187106013298034,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6437647456533453,
+      "eval_loss": 0.7173502445220947,
+      "eval_mean_token_accuracy": 0.8056439573175451,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.9568,
+      "eval_samples_per_second": 15.963,
+      "eval_steps_per_second": 2.001,
+      "step": 940
+    },
+    {
+      "entropy": 0.6508570514619351,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.9606207013130188,
+      "learning_rate": 0.0003459624676447067,
+      "loss": 0.6296409606933594,
+      "mean_token_accuracy": 0.8167732007801533,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6288247135248077,
+      "eval_loss": 0.7160675525665283,
+      "eval_mean_token_accuracy": 0.8066010689467527,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 89.0044,
+      "eval_samples_per_second": 15.954,
+      "eval_steps_per_second": 2.0,
+      "step": 960
+    },
+    {
+      "entropy": 0.6380081418901682,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.9677181839942932,
+      "learning_rate": 0.000344584138192027,
+      "loss": 0.6210546493530273,
+      "mean_token_accuracy": 0.8186135642230511,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6407805031604981,
+      "eval_loss": 0.7055043578147888,
+      "eval_mean_token_accuracy": 0.8070103157772107,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 89.5606,
+      "eval_samples_per_second": 15.855,
+      "eval_steps_per_second": 1.987,
+      "step": 980
+    },
+    {
+      "entropy": 0.6310251638293266,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 2.1046407222747803,
+      "learning_rate": 0.000343159901546516,
+      "loss": 0.6165830135345459,
+      "mean_token_accuracy": 0.8207855716347694,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6461507470420237,
+      "eval_loss": 0.7292136549949646,
+      "eval_mean_token_accuracy": 0.8030418261383356,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 89.315,
+      "eval_samples_per_second": 15.899,
+      "eval_steps_per_second": 1.993,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6256617344915867,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 1.1768349409103394,
+      "learning_rate": 0.0003416901607502972,
+      "loss": 0.6142709255218506,
+      "mean_token_accuracy": 0.8220669947564602,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6150523993406403,
+      "eval_loss": 0.7019563317298889,
+      "eval_mean_token_accuracy": 0.8094323099998946,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 89.419,
+      "eval_samples_per_second": 15.88,
+      "eval_steps_per_second": 1.991,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6126735735684633,
+      "epoch": 2.506940253470127,
+      "grad_norm": 1.8065496683120728,
+      "learning_rate": 0.00034017533172263055,
+      "loss": 0.6056458473205566,
+      "mean_token_accuracy": 0.8203762136399746,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6544345456562685,
+      "eval_loss": 0.7086517214775085,
+      "eval_mean_token_accuracy": 0.806709805901131,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 88.981,
+      "eval_samples_per_second": 15.958,
+      "eval_steps_per_second": 2.0,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6311754353344441,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 1.0576369762420654,
+      "learning_rate": 0.00033861584314221236,
+      "loss": 0.6176392555236816,
+      "mean_token_accuracy": 0.8177410490810871,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6273799200406235,
+      "eval_loss": 0.708163321018219,
+      "eval_mean_token_accuracy": 0.8032017099053672,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 89.7668,
+      "eval_samples_per_second": 15.819,
+      "eval_steps_per_second": 1.983,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6200249589979648,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.997065007686615,
+      "learning_rate": 0.0003370121363258637,
+      "loss": 0.6203888893127442,
+      "mean_token_accuracy": 0.8196275025606156,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6457576811983344,
+      "eval_loss": 0.6972290277481079,
+      "eval_mean_token_accuracy": 0.8095980849158898,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 89.2027,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 1080
+    },
+    {
+      "entropy": 0.626810473203659,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.865475058555603,
+      "learning_rate": 0.0003353646651036438,
+      "loss": 0.6102585792541504,
+      "mean_token_accuracy": 0.819708751142025,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.6412022431914726,
+      "eval_loss": 0.7036139965057373,
+      "eval_mean_token_accuracy": 0.8070317135098275,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 89.4733,
+      "eval_samples_per_second": 15.871,
+      "eval_steps_per_second": 1.989,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6581619590520859,
+      "epoch": 2.700060350030175,
+      "grad_norm": 1.00307035446167,
+      "learning_rate": 0.00033367389569042064,
+      "loss": 0.6397720336914062,
+      "mean_token_accuracy": 0.81413309648633,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6288736402318719,
+      "eval_loss": 0.6966825723648071,
+      "eval_mean_token_accuracy": 0.8094039669867312,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 88.7341,
+      "eval_samples_per_second": 16.003,
+      "eval_steps_per_second": 2.006,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6399412982165813,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.8589329719543457,
+      "learning_rate": 0.0003319403065539384,
+      "loss": 0.6347338199615479,
+      "mean_token_accuracy": 0.8139534957706929,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.683491913455256,
+      "eval_loss": 0.6984149217605591,
+      "eval_mean_token_accuracy": 0.8083989469522841,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 89.3929,
+      "eval_samples_per_second": 15.885,
+      "eval_steps_per_second": 1.991,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6264835461974144,
+      "epoch": 2.796620398310199,
+      "grad_norm": 0.7807853817939758,
+      "learning_rate": 0.0003301643882794163,
+      "loss": 0.6162629127502441,
+      "mean_token_accuracy": 0.8216292977333068,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6578792159476977,
+      "eval_loss": 0.6921441555023193,
+      "eval_mean_token_accuracy": 0.8094299177775223,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 89.2356,
+      "eval_samples_per_second": 15.913,
+      "eval_steps_per_second": 1.995,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6269129011780024,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 1.1675593852996826,
+      "learning_rate": 0.0003283466434307189,
+      "loss": 0.6235532760620117,
+      "mean_token_accuracy": 0.8188897788524627,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6533248447970058,
+      "eval_loss": 0.696849524974823,
+      "eval_mean_token_accuracy": 0.8102703713968898,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2602,
+      "eval_samples_per_second": 16.089,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.6259873781353236,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.9596486687660217,
+      "learning_rate": 0.00032648758640813655,
+      "loss": 0.6118728160858155,
+      "mean_token_accuracy": 0.8171859130263328,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6212078754821521,
+      "eval_loss": 0.6942870020866394,
+      "eval_mean_token_accuracy": 0.810484343365337,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 89.4375,
+      "eval_samples_per_second": 15.877,
+      "eval_steps_per_second": 1.99,
+      "step": 1200
+    },
+    {
+      "entropy": 0.6142537571489811,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 0.9625058770179749,
+      "learning_rate": 0.00032458774330281615,
+      "loss": 0.6060261249542236,
+      "mean_token_accuracy": 0.8223113484680653,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6553665667437436,
+      "eval_loss": 0.6835269331932068,
+      "eval_mean_token_accuracy": 0.8134723968720168,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 89.5092,
+      "eval_samples_per_second": 15.864,
+      "eval_steps_per_second": 1.989,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6301972426474094,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.9657333493232727,
+      "learning_rate": 0.0003226476517478835,
+      "loss": 0.6277759075164795,
+      "mean_token_accuracy": 0.8164266042411328,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6594253118788258,
+      "eval_loss": 0.6820935606956482,
+      "eval_mean_token_accuracy": 0.8132368198941263,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 89.6211,
+      "eval_samples_per_second": 15.844,
+      "eval_steps_per_second": 1.986,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5346100655469027,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8541185259819031,
+      "learning_rate": 0.0003206678607662996,
+      "loss": 0.5261692047119141,
+      "mean_token_accuracy": 0.8382289425119177,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5626645662476507,
+      "eval_loss": 0.7028542160987854,
+      "eval_mean_token_accuracy": 0.8133395572056931,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 89.1267,
+      "eval_samples_per_second": 15.932,
+      "eval_steps_per_second": 1.997,
+      "step": 1260
+    },
+    {
+      "entropy": 0.5001101832836866,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.8208815455436707,
+      "learning_rate": 0.0003186489306154935,
+      "loss": 0.4850132942199707,
+      "mean_token_accuracy": 0.8468983843922615,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5889462046743779,
+      "eval_loss": 0.691015899181366,
+      "eval_mean_token_accuracy": 0.8135226466012805,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.9677,
+      "eval_samples_per_second": 15.961,
+      "eval_steps_per_second": 2.001,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5242766339331866,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 0.7893072962760925,
+      "learning_rate": 0.0003165914326288163,
+      "loss": 0.5061577320098877,
+      "mean_token_accuracy": 0.8425608821213245,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5543691962957382,
+      "eval_loss": 0.6913734078407288,
+      "eval_mean_token_accuracy": 0.8154889680026622,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 89.6963,
+      "eval_samples_per_second": 15.831,
+      "eval_steps_per_second": 1.984,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5327721010893584,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.8992080092430115,
+      "learning_rate": 0.0003144959490538604,
+      "loss": 0.5068036556243897,
+      "mean_token_accuracy": 0.8420542575418949,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5734540803378887,
+      "eval_loss": 0.6948716044425964,
+      "eval_mean_token_accuracy": 0.8137217944257715,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 89.4597,
+      "eval_samples_per_second": 15.873,
+      "eval_steps_per_second": 1.99,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5247377116233111,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.9291845560073853,
+      "learning_rate": 0.0003123630728876902,
+      "loss": 0.5023272037506104,
+      "mean_token_accuracy": 0.8435182586312294,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5492362935891312,
+      "eval_loss": 0.6980633735656738,
+      "eval_mean_token_accuracy": 0.8155082919624415,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 89.2124,
+      "eval_samples_per_second": 15.917,
+      "eval_steps_per_second": 1.995,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5380499072372913,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 1.0477524995803833,
+      "learning_rate": 0.00031019340770903136,
+      "loss": 0.5264591693878173,
+      "mean_token_accuracy": 0.8391959741711617,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5896503307511297,
+      "eval_loss": 0.6997144222259521,
+      "eval_mean_token_accuracy": 0.8124373062942805,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 89.1147,
+      "eval_samples_per_second": 15.935,
+      "eval_steps_per_second": 1.997,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5460653610527515,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 1.2473069429397583,
+      "learning_rate": 0.00030798756750746476,
+      "loss": 0.5319677829742432,
+      "mean_token_accuracy": 0.8364320226013661,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5659052182114526,
+      "eval_loss": 0.6924049258232117,
+      "eval_mean_token_accuracy": 0.8153053585732921,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 89.6468,
+      "eval_samples_per_second": 15.84,
+      "eval_steps_per_second": 1.986,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5367407951503992,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 1.2679712772369385,
+      "learning_rate": 0.00030574617650967485,
+      "loss": 0.5190927028656006,
+      "mean_token_accuracy": 0.8399469174444676,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5581759144081159,
+      "eval_loss": 0.6933317184448242,
+      "eval_mean_token_accuracy": 0.8139757862251796,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 89.0749,
+      "eval_samples_per_second": 15.942,
+      "eval_steps_per_second": 1.998,
+      "step": 1400
+    },
+    {
+      "entropy": 0.530736118927598,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.9717612862586975,
+      "learning_rate": 0.0003034698690028009,
+      "loss": 0.512363052368164,
+      "mean_token_accuracy": 0.8419127956032753,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5233335836549823,
+      "eval_loss": 0.6958550214767456,
+      "eval_mean_token_accuracy": 0.8158778501360604,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 88.8664,
+      "eval_samples_per_second": 15.979,
+      "eval_steps_per_second": 2.003,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5327631575986743,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7649208903312683,
+      "learning_rate": 0.00030115928915494116,
+      "loss": 0.5202179431915284,
+      "mean_token_accuracy": 0.8384683802723885,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.5960972657364406,
+      "eval_loss": 0.6984783411026001,
+      "eval_mean_token_accuracy": 0.8095930597085631,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 89.5168,
+      "eval_samples_per_second": 15.863,
+      "eval_steps_per_second": 1.988,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5428155034780502,
+      "epoch": 3.51901025950513,
+      "grad_norm": 1.2790874242782593,
+      "learning_rate": 0.00029881509083286116,
+      "loss": 0.524717378616333,
+      "mean_token_accuracy": 0.8346644505858422,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.550318562917495,
+      "eval_loss": 0.6988471746444702,
+      "eval_mean_token_accuracy": 0.8132969897784544,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 89.2598,
+      "eval_samples_per_second": 15.909,
+      "eval_steps_per_second": 1.994,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5489993002265692,
+      "epoch": 3.567290283645142,
+      "grad_norm": 1.055821418762207,
+      "learning_rate": 0.00029643793741695687,
+      "loss": 0.5421150207519532,
+      "mean_token_accuracy": 0.8345710933208466,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5816821718818685,
+      "eval_loss": 0.6853668689727783,
+      "eval_mean_token_accuracy": 0.8142332360985574,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 89.0424,
+      "eval_samples_per_second": 15.947,
+      "eval_steps_per_second": 1.999,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5451365381479263,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.9027532935142517,
+      "learning_rate": 0.00029402850161352583,
+      "loss": 0.5463609218597412,
+      "mean_token_accuracy": 0.8369908526539802,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5344233209832331,
+      "eval_loss": 0.6794531345367432,
+      "eval_mean_token_accuracy": 0.8177759824843889,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 89.123,
+      "eval_samples_per_second": 15.933,
+      "eval_steps_per_second": 1.997,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5521467242389917,
+      "epoch": 3.663850331925166,
+      "grad_norm": 1.4606311321258545,
+      "learning_rate": 0.0002915874652643997,
+      "loss": 0.5421902179718018,
+      "mean_token_accuracy": 0.8339455373585224,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5807709094513668,
+      "eval_loss": 0.6949691772460938,
+      "eval_mean_token_accuracy": 0.8121141439743256,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.5437,
+      "eval_samples_per_second": 16.037,
+      "eval_steps_per_second": 2.01,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5455610822886229,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7269843220710754,
+      "learning_rate": 0.0002891155191539905,
+      "loss": 0.5407682895660401,
+      "mean_token_accuracy": 0.8393648222088814,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.6032205542151847,
+      "eval_loss": 0.6790974140167236,
+      "eval_mean_token_accuracy": 0.8165463112043531,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 89.517,
+      "eval_samples_per_second": 15.863,
+      "eval_steps_per_second": 1.988,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5474163968116045,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.8148866295814514,
+      "learning_rate": 0.00028661336281380717,
+      "loss": 0.5394441604614257,
+      "mean_token_accuracy": 0.8336028568446636,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5797032238392348,
+      "eval_loss": 0.670870304107666,
+      "eval_mean_token_accuracy": 0.8182691593518417,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 89.4468,
+      "eval_samples_per_second": 15.875,
+      "eval_steps_per_second": 1.99,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5511750865727663,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.3441656827926636,
+      "learning_rate": 0.0002840817043244964,
+      "loss": 0.5378933906555176,
+      "mean_token_accuracy": 0.8349151819944381,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5297858750217417,
+      "eval_loss": 0.680833637714386,
+      "eval_mean_token_accuracy": 0.8180924275617921,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 89.5417,
+      "eval_samples_per_second": 15.859,
+      "eval_steps_per_second": 1.988,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5379613988101483,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9813932776451111,
+      "learning_rate": 0.00028152126011546396,
+      "loss": 0.5340342044830322,
+      "mean_token_accuracy": 0.8346010789275169,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5811898510777549,
+      "eval_loss": 0.6764267683029175,
+      "eval_mean_token_accuracy": 0.8180528236239144,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 88.7831,
+      "eval_samples_per_second": 15.994,
+      "eval_steps_per_second": 2.005,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5607876226305961,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.9617928266525269,
+      "learning_rate": 0.00027893275476213383,
+      "loss": 0.5434338569641113,
+      "mean_token_accuracy": 0.8332898400723934,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5474644318390428,
+      "eval_loss": 0.6740226745605469,
+      "eval_mean_token_accuracy": 0.8185851306058047,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 89.3246,
+      "eval_samples_per_second": 15.897,
+      "eval_steps_per_second": 1.993,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5354843523353339,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.5320990085601807,
+      "learning_rate": 0.0002763169207809021,
+      "loss": 0.5251852989196777,
+      "mean_token_accuracy": 0.8416872084140777,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.5436856410141742,
+      "eval_loss": 0.6675190925598145,
+      "eval_mean_token_accuracy": 0.8202396507343549,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 89.6328,
+      "eval_samples_per_second": 15.842,
+      "eval_steps_per_second": 1.986,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5418571938167919,
+      "epoch": 4.0,
+      "grad_norm": 4.690354347229004,
+      "learning_rate": 0.000273674498421843,
+      "loss": 0.548275089263916,
+      "mean_token_accuracy": 0.836557479647847,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5497228983747825,
+      "eval_loss": 0.6621812582015991,
+      "eval_mean_token_accuracy": 0.8205679802412398,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 89.6272,
+      "eval_samples_per_second": 15.843,
+      "eval_steps_per_second": 1.986,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4195931971073151,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6678048968315125,
+      "learning_rate": 0.0002710062354592273,
+      "loss": 0.3970795154571533,
+      "mean_token_accuracy": 0.8719995342195034,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.46495268753405367,
+      "eval_loss": 0.7061581611633301,
+      "eval_mean_token_accuracy": 0.8190107995204712,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 89.6387,
+      "eval_samples_per_second": 15.841,
+      "eval_steps_per_second": 1.986,
+      "step": 1680
+    },
+    {
+      "entropy": 0.4263830740004778,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.7026669383049011,
+      "learning_rate": 0.000268312886979911,
+      "loss": 0.398905086517334,
+      "mean_token_accuracy": 0.8691012300550938,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.45755320670229666,
+      "eval_loss": 0.7102291584014893,
+      "eval_mean_token_accuracy": 0.819212955370378,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 89.2777,
+      "eval_samples_per_second": 15.905,
+      "eval_steps_per_second": 1.994,
+      "step": 1700
+    },
+    {
+      "entropy": 0.41640330031514167,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.752216100692749,
+      "learning_rate": 0.00026559521516965437,
+      "loss": 0.39748263359069824,
+      "mean_token_accuracy": 0.8684553548693656,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.469178682297803,
+      "eval_loss": 0.7004125714302063,
+      "eval_mean_token_accuracy": 0.8194386949030201,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 89.8428,
+      "eval_samples_per_second": 15.805,
+      "eval_steps_per_second": 1.981,
+      "step": 1720
+    },
+    {
+      "entropy": 0.42079071439802646,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.8349477648735046,
+      "learning_rate": 0.0002628539890974329,
+      "loss": 0.40129623413085935,
+      "mean_token_accuracy": 0.8686790131032467,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48358210871058904,
+      "eval_loss": 0.70196533203125,
+      "eval_mean_token_accuracy": 0.818096743875675,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 90.444,
+      "eval_samples_per_second": 15.7,
+      "eval_steps_per_second": 1.968,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4285690750926733,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.781974732875824,
+      "learning_rate": 0.00026008998449779933,
+      "loss": 0.40457854270935056,
+      "mean_token_accuracy": 0.8679066866636276,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.4653420230645812,
+      "eval_loss": 0.7041526436805725,
+      "eval_mean_token_accuracy": 0.8190121017815022,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 89.5248,
+      "eval_samples_per_second": 15.862,
+      "eval_steps_per_second": 1.988,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4335357129573822,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.8383703827857971,
+      "learning_rate": 0.0002573039835513604,
+      "loss": 0.41876921653747556,
+      "mean_token_accuracy": 0.8663770146667957,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.5015502416350869,
+      "eval_loss": 0.6904259324073792,
+      "eval_mean_token_accuracy": 0.8200626497188311,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 89.7014,
+      "eval_samples_per_second": 15.83,
+      "eval_steps_per_second": 1.984,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4258002445101738,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.7313240766525269,
+      "learning_rate": 0.00025449677466342867,
+      "loss": 0.40549774169921876,
+      "mean_token_accuracy": 0.8665009342133999,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.49225394293833313,
+      "eval_loss": 0.6914838552474976,
+      "eval_mean_token_accuracy": 0.8200667657878962,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 91.6837,
+      "eval_samples_per_second": 15.488,
+      "eval_steps_per_second": 1.941,
+      "step": 1800
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.8035854016335258e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-1980/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-1980/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-1980/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-1980/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-1980/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-1980/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-1980/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-1980/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-1980/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-1980/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-1980/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-1980/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-1980/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-1980/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..d5c23212dd31aedbbc7ce0ace97f36c8bf2c5fc0
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-1980/trainer_state.json
@@ -0,0 +1,2113 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 4.772480386240193,
+  "eval_steps": 20,
+  "global_step": 1980,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.9366227277457203e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-20/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-20/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-20/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-20/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-20/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-20/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-20/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-20/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-20/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-20/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-20/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-20/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-20/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-20/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..eab9ea30e4820a6d866de21279b332866b819921
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-20/trainer_state.json
@@ -0,0 +1,55 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.04828002414001207,
+  "eval_steps": 20,
+  "global_step": 20,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1980622700961792.0,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-200/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-200/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-200/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-200/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-200/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-200/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-200/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-200/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-200/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-200/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-200/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-200/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-200/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-200/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..e7faecad0e5485d520712f4c86398d7523e2de06
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-200/trainer_state.json
@@ -0,0 +1,244 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.4828002414001207,
+  "eval_steps": 20,
+  "global_step": 200,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.002938588074803e+16,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2000/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2000/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2000/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2000/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2000/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2000/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2000/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2000/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2000/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2000/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2000/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2000/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2000/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2000/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..e06986da34f4cc6d65c9433694a91294c02506eb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2000/trainer_state.json
@@ -0,0 +1,2134 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 4.820760410380205,
+  "eval_steps": 20,
+  "global_step": 2000,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.956564043065692e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2020/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2020/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2020/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2020/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2020/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2020/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2020/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2020/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2020/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2020/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2020/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2020/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2020/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2020/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..bdead565fe99ea51bad173a428f597c421f16a98
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2020/trainer_state.json
@@ -0,0 +1,2155 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 4.869040434520217,
+  "eval_steps": 20,
+  "global_step": 2020,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.9775676437352653e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2040/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2040/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2040/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2040/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2040/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2040/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2040/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2040/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2040/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2040/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2040/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2040/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2040/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2040/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..18b70d75d7f9bdc19ef082b899bc65a900ed4d30
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2040/trainer_state.json
@@ -0,0 +1,2176 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 4.917320458660229,
+  "eval_steps": 20,
+  "global_step": 2040,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.996087023009239e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2060/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2060/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2060/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2060/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2060/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2060/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2060/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2060/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2060/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2060/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2060/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2060/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2060/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2060/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..ae0000c47a490271cc79a5c820e49d114a36d2f4
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2060/trainer_state.json
@@ -0,0 +1,2197 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 4.965600482800241,
+  "eval_steps": 20,
+  "global_step": 2060,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.0141362247874355e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2080/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2080/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2080/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2080/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2080/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2080/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2080/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2080/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2080/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2080/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2080/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2080/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2080/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2080/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..b7f001f4ad835f76a021df2902e492496dbef859
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2080/trainer_state.json
@@ -0,0 +1,2218 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 5.012070006035003,
+  "eval_steps": 20,
+  "global_step": 2080,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.034164382592635e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2100/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2100/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2100/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2100/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2100/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2100/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2100/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2100/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2100/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2100/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2100/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2100/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2100/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2100/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..23122add48828c62f5dfd1b8913cc0ef3376325d
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2100/trainer_state.json
@@ -0,0 +1,2239 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 5.060350030175015,
+  "eval_steps": 20,
+  "global_step": 2100,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.0530469213502874e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2120/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2120/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2120/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2120/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2120/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2120/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2120/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2120/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2120/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2120/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2120/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2120/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2120/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2120/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..ba4b833ba7fad512f560f4e4c617f59c00606992
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2120/trainer_state.json
@@ -0,0 +1,2260 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 5.108630054315027,
+  "eval_steps": 20,
+  "global_step": 2120,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.0717110380996403e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2140/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2140/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2140/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2140/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2140/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2140/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2140/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2140/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2140/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2140/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2140/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2140/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2140/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2140/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..7c1a61156389705a460f54812d7f7df87005f47a
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2140/trainer_state.json
@@ -0,0 +1,2281 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 5.15691007845504,
+  "eval_steps": 20,
+  "global_step": 2140,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.090833051589284e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2160/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2160/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2160/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2160/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2160/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2160/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2160/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2160/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2160/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2160/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2160/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2160/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2160/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2160/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..b54988c810517ebf71a916cd105e89338d8523b3
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2160/trainer_state.json
@@ -0,0 +1,2302 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 5.2051901025950515,
+  "eval_steps": 20,
+  "global_step": 2160,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.1095296246209946e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2180/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2180/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2180/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2180/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2180/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2180/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2180/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2180/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2180/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2180/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2180/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2180/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2180/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2180/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..35c335e5ed20742fda69ad42952b2144c78acbca
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2180/trainer_state.json
@@ -0,0 +1,2323 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 5.253470126735063,
+  "eval_steps": 20,
+  "global_step": 2180,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.129812169504133e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-220/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-220/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-220/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-220/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-220/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-220/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-220/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-220/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-220/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-220/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-220/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-220/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-220/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-220/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..ec51bf7ad065cdf131531cea62b9ceb5714d3d87
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-220/trainer_state.json
@@ -0,0 +1,265 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.5310802655401328,
+  "eval_steps": 20,
+  "global_step": 220,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.18071129571328e+16,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2200/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2200/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2200/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2200/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2200/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2200/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2200/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2200/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2200/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2200/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2200/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2200/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2200/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2200/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..5c3fdd27c8a15e3c4297cf6c17c1eaa51c624b4e
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2200/trainer_state.json
@@ -0,0 +1,2344 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 5.301750150875075,
+  "eval_steps": 20,
+  "global_step": 2200,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.148549093589586e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2220/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2220/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2220/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2220/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2220/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2220/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2220/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2220/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2220/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2220/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2220/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2220/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2220/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2220/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..bede130010a92c3e190957528ee99ee9d7d8ccd4
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2220/trainer_state.json
@@ -0,0 +1,2365 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 5.350030175015087,
+  "eval_steps": 20,
+  "global_step": 2220,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.1683140923486413e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2240/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2240/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2240/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2240/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2240/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2240/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2240/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2240/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2240/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2240/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2240/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2240/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2240/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2240/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..4a0e1cd5bb5fe1d1f02099decc715eabb5528449
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2240/trainer_state.json
@@ -0,0 +1,2386 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 5.3983101991551,
+  "eval_steps": 20,
+  "global_step": 2240,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.187844002359808e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2260/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2260/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2260/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2260/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2260/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2260/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2260/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2260/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2260/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2260/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2260/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2260/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2260/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2260/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..90385fc1cfffc72e6015903d9795ab9751642b03
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2260/trainer_state.json
@@ -0,0 +1,2407 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 5.446590223295112,
+  "eval_steps": 20,
+  "global_step": 2260,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.2065818036420813e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2280/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2280/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2280/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2280/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2280/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2280/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2280/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2280/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2280/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2280/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2280/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2280/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2280/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2280/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..50eb2da858d6ef868950c637dcedb3aa44828272
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2280/trainer_state.json
@@ -0,0 +1,2428 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 5.494870247435124,
+  "eval_steps": 20,
+  "global_step": 2280,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.2258836424799232e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2300/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2300/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2300/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2300/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2300/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2300/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2300/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2300/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2300/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2300/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2300/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2300/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2300/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2300/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..821e06b5d757bad8f96e5c4bce613a90e3de2ef2
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2300/trainer_state.json
@@ -0,0 +1,2449 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 5.5431502715751355,
+  "eval_steps": 20,
+  "global_step": 2300,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.2464670658724864e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2320/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2320/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2320/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2320/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2320/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2320/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2320/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2320/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2320/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2320/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2320/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2320/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2320/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2320/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..6439d770fe64ebcb1a34ad597aa135d3352bd1f2
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2320/trainer_state.json
@@ -0,0 +1,2470 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 5.591430295715147,
+  "eval_steps": 20,
+  "global_step": 2320,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.2683092667024384e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2340/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2340/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2340/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2340/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2340/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2340/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2340/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2340/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2340/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2340/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2340/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2340/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2340/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2340/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..4556c146abe3b666205842349ce1a29d57b68fa3
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2340/trainer_state.json
@@ -0,0 +1,2491 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 5.63971031985516,
+  "eval_steps": 20,
+  "global_step": 2340,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.2901076076913664e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2360/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2360/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2360/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2360/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2360/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2360/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2360/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2360/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2360/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2360/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2360/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2360/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2360/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2360/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..157c6c9b89652a3e5de38736e229df11dd977773
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2360/trainer_state.json
@@ -0,0 +1,2512 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 5.687990343995172,
+  "eval_steps": 20,
+  "global_step": 2360,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.310536644443525e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2380/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2380/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2380/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2380/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2380/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2380/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2380/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2380/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2380/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2380/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2380/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2380/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2380/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2380/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..fdc762f0617759ecd83412579f007fe2f40fd706
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2380/trainer_state.json
@@ -0,0 +1,2533 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 5.736270368135184,
+  "eval_steps": 20,
+  "global_step": 2380,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.3297955006371635e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-240/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-240/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-240/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-240/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-240/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-240/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-240/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-240/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-240/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-240/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-240/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-240/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-240/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-240/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..28ee29eb84e32c864fe215caffb4de6dbf6b73b0
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-240/trainer_state.json
@@ -0,0 +1,286 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.5793602896801449,
+  "eval_steps": 20,
+  "global_step": 240,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.3722647654014976e+16,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2400/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2400/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2400/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2400/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2400/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2400/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2400/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2400/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2400/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2400/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2400/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2400/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2400/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2400/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..658f64f023d7d3289f20913d8b01c71f87f87040
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2400/trainer_state.json
@@ -0,0 +1,2554 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 5.784550392275197,
+  "eval_steps": 20,
+  "global_step": 2400,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.3500657647648154e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2420/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2420/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2420/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2420/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2420/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2420/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2420/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2420/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2420/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2420/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2420/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2420/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2420/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2420/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..0347929814804d1c54b132e8f575a7949fb009eb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2420/trainer_state.json
@@ -0,0 +1,2575 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 5.8328304164152085,
+  "eval_steps": 20,
+  "global_step": 2420,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.3693877791295283e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2440/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2440/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2440/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2440/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2440/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2440/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2440/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2440/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2440/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2440/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2440/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2440/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2440/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2440/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..38d1d28b4867fec18c718259a9a02eb71b7e3d7d
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2440/trainer_state.json
@@ -0,0 +1,2596 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 5.88111044055522,
+  "eval_steps": 20,
+  "global_step": 2440,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.3887703200748544e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2460/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2460/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2460/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2460/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2460/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2460/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2460/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2460/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2460/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2460/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2460/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2460/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2460/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2460/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..e33dc1a49f4e7849aea0e667f057ffd370152dab
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2460/trainer_state.json
@@ -0,0 +1,2617 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 5.929390464695232,
+  "eval_steps": 20,
+  "global_step": 2460,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.4083888269648896e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2480/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2480/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2480/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2480/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2480/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2480/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2480/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2480/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2480/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2480/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2480/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2480/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2480/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2480/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..4af553a6bee24b4082c5d5febd51b7af653e9d89
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2480/trainer_state.json
@@ -0,0 +1,2638 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 5.977670488835244,
+  "eval_steps": 20,
+  "global_step": 2480,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.4278713683477504e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2500/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2500/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2500/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2500/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2500/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2500/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2500/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2500/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2500/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2500/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2500/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2500/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2500/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2500/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..1ace1eb34540ce725bea5745e358df0fadc5f9f9
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2500/trainer_state.json
@@ -0,0 +1,2659 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 6.024140012070006,
+  "eval_steps": 20,
+  "global_step": 2500,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.4485023796678246e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2520/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2520/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2520/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2520/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2520/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2520/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2520/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2520/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2520/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2520/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2520/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2520/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2520/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2520/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..a0b8f1794fa4884b4d51a2123faf3aabb9fe53f7
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2520/trainer_state.json
@@ -0,0 +1,2680 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 6.072420036210018,
+  "eval_steps": 20,
+  "global_step": 2520,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.4673068479084544e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2540/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2540/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2540/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2540/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2540/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2540/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2540/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2540/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2540/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2540/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2540/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2540/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2540/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2540/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..fb40c6512296edce2c96ff535850961a0b38eabd
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2540/trainer_state.json
@@ -0,0 +1,2701 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 6.12070006035003,
+  "eval_steps": 20,
+  "global_step": 2540,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.486442019350405e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2560/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2560/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2560/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2560/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2560/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2560/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2560/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2560/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2560/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2560/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2560/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2560/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2560/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2560/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..23dc97df31f24e5c1b5e4b7aa2c7c1e440c56b32
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2560/trainer_state.json
@@ -0,0 +1,2722 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 6.168980084490042,
+  "eval_steps": 20,
+  "global_step": 2560,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.5044105190211174e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2580/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2580/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2580/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2580/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2580/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2580/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2580/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2580/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2580/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2580/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2580/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2580/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2580/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2580/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..533448a9cbe849fb25d8f5edefdf89bf4154b86d
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2580/trainer_state.json
@@ -0,0 +1,2743 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 6.217260108630055,
+  "eval_steps": 20,
+  "global_step": 2580,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.5242299039830426e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-260/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-260/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-260/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-260/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-260/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-260/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-260/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-260/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-260/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-260/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-260/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-260/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-260/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-260/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..40b26c99f97f7b5375db99d6e691bb2d6a722c0d
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-260/trainer_state.json
@@ -0,0 +1,307 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.627640313820157,
+  "eval_steps": 20,
+  "global_step": 260,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.560572606853939e+16,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2600/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2600/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2600/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2600/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2600/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2600/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2600/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2600/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2600/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2600/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2600/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2600/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2600/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2600/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..dc47727306f2833936a6a92ed5b0743598bac4bf
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2600/trainer_state.json
@@ -0,0 +1,2764 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 6.265540132770067,
+  "eval_steps": 20,
+  "global_step": 2600,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.5449440297018573e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2620/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2620/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2620/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2620/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2620/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2620/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2620/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2620/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2620/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2620/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2620/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2620/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2620/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2620/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..b233f0f90b67711aa2731c9af3065f360a8d82b8
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2620/trainer_state.json
@@ -0,0 +1,2785 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 6.3138201569100785,
+  "eval_steps": 20,
+  "global_step": 2620,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.5660011393774797e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2640/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2640/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2640/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2640/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2640/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2640/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2640/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2640/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2640/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2640/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2640/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2640/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2640/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2640/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..dccc2cc2014751816288ad918b68c267d2cd42ed
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2640/trainer_state.json
@@ -0,0 +1,2806 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 6.36210018105009,
+  "eval_steps": 20,
+  "global_step": 2640,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.5865819311795814e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2660/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2660/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2660/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2660/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2660/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2660/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2660/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2660/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2660/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2660/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2660/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2660/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2660/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2660/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..5dc056fb2ddd521095113ec7bd517d2c67a746c8
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2660/trainer_state.json
@@ -0,0 +1,2827 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 6.410380205190102,
+  "eval_steps": 20,
+  "global_step": 2660,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.6047513089221837e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2680/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2680/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2680/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2680/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2680/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2680/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2680/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2680/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2680/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2680/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2680/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2680/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2680/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2680/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..d82401e2a088e0ff250f5d6d44599a7b2e18cbef
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2680/trainer_state.json
@@ -0,0 +1,2848 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 6.458660229330115,
+  "eval_steps": 20,
+  "global_step": 2680,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.6248496424730214e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2700/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2700/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2700/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2700/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2700/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2700/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2700/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2700/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2700/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2700/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2700/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2700/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2700/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2700/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..a3f26bcedeca9fd56cbb640e3bde811d7268725b
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2700/trainer_state.json
@@ -0,0 +1,2869 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 6.506940253470127,
+  "eval_steps": 20,
+  "global_step": 2700,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.6441006038952755e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2720/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2720/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2720/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2720/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2720/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2720/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2720/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2720/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2720/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2720/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2720/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2720/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2720/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2720/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..836f2b4ceafffd600f4d0abce044a8a37ec9b69f
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2720/trainer_state.json
@@ -0,0 +1,2890 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 6.555220277610139,
+  "eval_steps": 20,
+  "global_step": 2720,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.661749803923333e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2740/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2740/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2740/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2740/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2740/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2740/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2740/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2740/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2740/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2740/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2740/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2740/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2740/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2740/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..43091e391c2a98f529473101ed677d1b7068092e
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2740/trainer_state.json
@@ -0,0 +1,2911 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 6.603500301750151,
+  "eval_steps": 20,
+  "global_step": 2740,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.680984098605998e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2760/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2760/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2760/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2760/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2760/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2760/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2760/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2760/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2760/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2760/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2760/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2760/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2760/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2760/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..8171cad0986f96f0e07b201f89d911322ee805dc
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2760/trainer_state.json
@@ -0,0 +1,2932 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 6.651780325890163,
+  "eval_steps": 20,
+  "global_step": 2760,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.700181551022203e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2780/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2780/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2780/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2780/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2780/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2780/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2780/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2780/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2780/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2780/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2780/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2780/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2780/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2780/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..f6198de356a225b47482c578a30b4baddc367b91
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2780/trainer_state.json
@@ -0,0 +1,2953 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 6.700060350030175,
+  "eval_steps": 20,
+  "global_step": 2780,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.718843913377915e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-280/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-280/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-280/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-280/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-280/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-280/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-280/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-280/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-280/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-280/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-280/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-280/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-280/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-280/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..7206c7b360a82d6e5e345ecba613257b021b5ff7
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-280/trainer_state.json
@@ -0,0 +1,328 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.6759203379601689,
+  "eval_steps": 20,
+  "global_step": 280,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.760327866813645e+16,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2800/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2800/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2800/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2800/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2800/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2800/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2800/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2800/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2800/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2800/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2800/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2800/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2800/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2800/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..1fc0258615ad8cbcdcf2c3c77b60ccfc04b7bbea
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2800/trainer_state.json
@@ -0,0 +1,2974 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 6.748340374170187,
+  "eval_steps": 20,
+  "global_step": 2800,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.737784347125719e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2820/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2820/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2820/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2820/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2820/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2820/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2820/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2820/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2820/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2820/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2820/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2820/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2820/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2820/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..9207b60ec65ed234952c48bafe27b126be401ce2
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2820/trainer_state.json
@@ -0,0 +1,2995 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 6.796620398310199,
+  "eval_steps": 20,
+  "global_step": 2820,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.758241454176133e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2840/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2840/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2840/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2840/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2840/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2840/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2840/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2840/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2840/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2840/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2840/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2840/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2840/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2840/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..699bf58eeba3578ce90d2b4e07d408cf5dd224e6
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2840/trainer_state.json
@@ -0,0 +1,3016 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 6.844900422450211,
+  "eval_steps": 20,
+  "global_step": 2840,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.778160839575593e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2860/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2860/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2860/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2860/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2860/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2860/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2860/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2860/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2860/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2860/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2860/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2860/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2860/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2860/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..e5849bb70f4695e15fe28f32e6d545d1781c7730
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2860/trainer_state.json
@@ -0,0 +1,3037 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 6.8931804465902236,
+  "eval_steps": 20,
+  "global_step": 2860,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.7973670639600026e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2880/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2880/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2880/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2880/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2880/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2880/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2880/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2880/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2880/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2880/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2880/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2880/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2880/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2880/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..29a7a061a67675c58c5923c417c54476b64e95d6
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2880/trainer_state.json
@@ -0,0 +1,3058 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 6.941460470730235,
+  "eval_steps": 20,
+  "global_step": 2880,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.8180978564184064e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2900/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2900/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2900/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2900/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2900/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2900/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2900/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2900/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2900/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2900/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2900/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2900/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2900/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2900/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..b436411be2d5b941b5aeb9cb48e88f6f89ccc12d
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2900/trainer_state.json
@@ -0,0 +1,3079 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 6.989740494870247,
+  "eval_steps": 20,
+  "global_step": 2900,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.837826890107822e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2920/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2920/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2920/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2920/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2920/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2920/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2920/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2920/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2920/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2920/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2920/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2920/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2920/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2920/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..1394feb251929998c89f62f76218e56db8e5f61b
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2920/trainer_state.json
@@ -0,0 +1,3100 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 7.036210018105009,
+  "eval_steps": 20,
+  "global_step": 2920,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.8573758791498342e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2940/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2940/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2940/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2940/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2940/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2940/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2940/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2940/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2940/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2940/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2940/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2940/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2940/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2940/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..01b999feaf411c5742fbb6e32f9c1c0f07fc71f2
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2940/trainer_state.json
@@ -0,0 +1,3121 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 7.084490042245021,
+  "eval_steps": 20,
+  "global_step": 2940,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.8770347370936115e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2960/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2960/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2960/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2960/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2960/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2960/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2960/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2960/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2960/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2960/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2960/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2960/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2960/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2960/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..378849b7d6d9ba4a4d0d9698df6ebf280bf30d46
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2960/trainer_state.json
@@ -0,0 +1,3142 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 7.132770066385033,
+  "eval_steps": 20,
+  "global_step": 2960,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.896958508477174e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2980/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2980/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2980/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2980/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2980/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2980/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2980/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2980/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2980/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2980/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2980/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2980/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2980/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2980/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..a99d74567808158f22169cf3a9c5fe2165b522d9
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-2980/trainer_state.json
@@ -0,0 +1,3163 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 7.181050090525045,
+  "eval_steps": 20,
+  "global_step": 2980,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    },
+    {
+      "entropy": 0.1326121488586068,
+      "epoch": 7.181050090525045,
+      "grad_norm": 0.4843117296695709,
+      "learning_rate": 8.411870386507724e-05,
+      "loss": 0.09537842869758606,
+      "mean_token_accuracy": 0.9645151920616627,
+      "num_tokens": 6898444.0,
+      "step": 2980
+    },
+    {
+      "epoch": 7.181050090525045,
+      "eval_entropy": 0.27999477783280813,
+      "eval_loss": 0.9031037092208862,
+      "eval_mean_token_accuracy": 0.8261882204018282,
+      "eval_num_tokens": 6898444.0,
+      "eval_runtime": 88.5602,
+      "eval_samples_per_second": 16.034,
+      "eval_steps_per_second": 2.01,
+      "step": 2980
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.915818240117494e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-300/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-300/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-300/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-300/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-300/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-300/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-300/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-300/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-300/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-300/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-300/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-300/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-300/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-300/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..027156dcbaeda5daeb9ced7e720bd84de71eec8f
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-300/trainer_state.json
@@ -0,0 +1,349 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.724200362100181,
+  "eval_steps": 20,
+  "global_step": 300,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.9639427927834624e+16,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3000/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3000/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3000/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3000/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3000/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3000/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3000/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3000/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3000/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3000/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3000/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3000/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3000/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3000/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..a5db819bd0bb4d2cbdd0965bacfe6d1c04ec5aea
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3000/trainer_state.json
@@ -0,0 +1,3184 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 7.229330114665057,
+  "eval_steps": 20,
+  "global_step": 3000,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    },
+    {
+      "entropy": 0.1326121488586068,
+      "epoch": 7.181050090525045,
+      "grad_norm": 0.4843117296695709,
+      "learning_rate": 8.411870386507724e-05,
+      "loss": 0.09537842869758606,
+      "mean_token_accuracy": 0.9645151920616627,
+      "num_tokens": 6898444.0,
+      "step": 2980
+    },
+    {
+      "epoch": 7.181050090525045,
+      "eval_entropy": 0.27999477783280813,
+      "eval_loss": 0.9031037092208862,
+      "eval_mean_token_accuracy": 0.8261882204018282,
+      "eval_num_tokens": 6898444.0,
+      "eval_runtime": 88.5602,
+      "eval_samples_per_second": 16.034,
+      "eval_steps_per_second": 2.01,
+      "step": 2980
+    },
+    {
+      "entropy": 0.1156751093454659,
+      "epoch": 7.229330114665057,
+      "grad_norm": 0.5297922492027283,
+      "learning_rate": 8.149629351430326e-05,
+      "loss": 0.09203402996063233,
+      "mean_token_accuracy": 0.9689733698964119,
+      "num_tokens": 6947106.0,
+      "step": 3000
+    },
+    {
+      "epoch": 7.229330114665057,
+      "eval_entropy": 0.2740225845508361,
+      "eval_loss": 0.8876440525054932,
+      "eval_mean_token_accuracy": 0.8265116857678703,
+      "eval_num_tokens": 6947106.0,
+      "eval_runtime": 87.7971,
+      "eval_samples_per_second": 16.174,
+      "eval_steps_per_second": 2.027,
+      "step": 3000
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.937224474127667e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3020/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3020/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3020/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3020/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3020/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3020/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3020/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3020/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3020/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3020/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3020/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3020/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3020/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3020/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..c77d20c645623dc039a24d854c6b2118e9ac3c53
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3020/trainer_state.json
@@ -0,0 +1,3205 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 7.27761013880507,
+  "eval_steps": 20,
+  "global_step": 3020,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    },
+    {
+      "entropy": 0.1326121488586068,
+      "epoch": 7.181050090525045,
+      "grad_norm": 0.4843117296695709,
+      "learning_rate": 8.411870386507724e-05,
+      "loss": 0.09537842869758606,
+      "mean_token_accuracy": 0.9645151920616627,
+      "num_tokens": 6898444.0,
+      "step": 2980
+    },
+    {
+      "epoch": 7.181050090525045,
+      "eval_entropy": 0.27999477783280813,
+      "eval_loss": 0.9031037092208862,
+      "eval_mean_token_accuracy": 0.8261882204018282,
+      "eval_num_tokens": 6898444.0,
+      "eval_runtime": 88.5602,
+      "eval_samples_per_second": 16.034,
+      "eval_steps_per_second": 2.01,
+      "step": 2980
+    },
+    {
+      "entropy": 0.1156751093454659,
+      "epoch": 7.229330114665057,
+      "grad_norm": 0.5297922492027283,
+      "learning_rate": 8.149629351430326e-05,
+      "loss": 0.09203402996063233,
+      "mean_token_accuracy": 0.9689733698964119,
+      "num_tokens": 6947106.0,
+      "step": 3000
+    },
+    {
+      "epoch": 7.229330114665057,
+      "eval_entropy": 0.2740225845508361,
+      "eval_loss": 0.8876440525054932,
+      "eval_mean_token_accuracy": 0.8265116857678703,
+      "eval_num_tokens": 6947106.0,
+      "eval_runtime": 87.7971,
+      "eval_samples_per_second": 16.174,
+      "eval_steps_per_second": 2.027,
+      "step": 3000
+    },
+    {
+      "entropy": 0.12116114404052496,
+      "epoch": 7.27761013880507,
+      "grad_norm": 0.5467398166656494,
+      "learning_rate": 7.890406302404618e-05,
+      "loss": 0.0933056652545929,
+      "mean_token_accuracy": 0.9676455594599247,
+      "num_tokens": 6992784.0,
+      "step": 3020
+    },
+    {
+      "epoch": 7.27761013880507,
+      "eval_entropy": 0.27238342793804876,
+      "eval_loss": 0.9166117310523987,
+      "eval_mean_token_accuracy": 0.825499015242866,
+      "eval_num_tokens": 6992784.0,
+      "eval_runtime": 88.3276,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 3020
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.956919297141084e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3040/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3040/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3040/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3040/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3040/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3040/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3040/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3040/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3040/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3040/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3040/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3040/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3040/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3040/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..bf34fbaf82e4bc333c3b02d1e92c511f797813c0
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3040/trainer_state.json
@@ -0,0 +1,3226 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 7.325890162945082,
+  "eval_steps": 20,
+  "global_step": 3040,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    },
+    {
+      "entropy": 0.1326121488586068,
+      "epoch": 7.181050090525045,
+      "grad_norm": 0.4843117296695709,
+      "learning_rate": 8.411870386507724e-05,
+      "loss": 0.09537842869758606,
+      "mean_token_accuracy": 0.9645151920616627,
+      "num_tokens": 6898444.0,
+      "step": 2980
+    },
+    {
+      "epoch": 7.181050090525045,
+      "eval_entropy": 0.27999477783280813,
+      "eval_loss": 0.9031037092208862,
+      "eval_mean_token_accuracy": 0.8261882204018282,
+      "eval_num_tokens": 6898444.0,
+      "eval_runtime": 88.5602,
+      "eval_samples_per_second": 16.034,
+      "eval_steps_per_second": 2.01,
+      "step": 2980
+    },
+    {
+      "entropy": 0.1156751093454659,
+      "epoch": 7.229330114665057,
+      "grad_norm": 0.5297922492027283,
+      "learning_rate": 8.149629351430326e-05,
+      "loss": 0.09203402996063233,
+      "mean_token_accuracy": 0.9689733698964119,
+      "num_tokens": 6947106.0,
+      "step": 3000
+    },
+    {
+      "epoch": 7.229330114665057,
+      "eval_entropy": 0.2740225845508361,
+      "eval_loss": 0.8876440525054932,
+      "eval_mean_token_accuracy": 0.8265116857678703,
+      "eval_num_tokens": 6947106.0,
+      "eval_runtime": 87.7971,
+      "eval_samples_per_second": 16.174,
+      "eval_steps_per_second": 2.027,
+      "step": 3000
+    },
+    {
+      "entropy": 0.12116114404052496,
+      "epoch": 7.27761013880507,
+      "grad_norm": 0.5467398166656494,
+      "learning_rate": 7.890406302404618e-05,
+      "loss": 0.0933056652545929,
+      "mean_token_accuracy": 0.9676455594599247,
+      "num_tokens": 6992784.0,
+      "step": 3020
+    },
+    {
+      "epoch": 7.27761013880507,
+      "eval_entropy": 0.27238342793804876,
+      "eval_loss": 0.9166117310523987,
+      "eval_mean_token_accuracy": 0.825499015242866,
+      "eval_num_tokens": 6992784.0,
+      "eval_runtime": 88.3276,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 3020
+    },
+    {
+      "entropy": 0.11715119928121567,
+      "epoch": 7.325890162945082,
+      "grad_norm": 0.4976702332496643,
+      "learning_rate": 7.634274596487176e-05,
+      "loss": 0.09449659585952759,
+      "mean_token_accuracy": 0.966506478190422,
+      "num_tokens": 7040482.0,
+      "step": 3040
+    },
+    {
+      "epoch": 7.325890162945082,
+      "eval_entropy": 0.2788975556077582,
+      "eval_loss": 0.8833065629005432,
+      "eval_mean_token_accuracy": 0.8269920596915684,
+      "eval_num_tokens": 7040482.0,
+      "eval_runtime": 88.9016,
+      "eval_samples_per_second": 15.973,
+      "eval_steps_per_second": 2.002,
+      "step": 3040
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.976495698583736e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3060/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3060/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3060/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3060/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3060/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3060/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3060/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3060/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3060/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3060/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3060/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3060/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3060/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3060/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..b5aef08d8b661f5ae26f8e38af21116c1bee1d71
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3060/trainer_state.json
@@ -0,0 +1,3247 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 7.3741701870850935,
+  "eval_steps": 20,
+  "global_step": 3060,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    },
+    {
+      "entropy": 0.1326121488586068,
+      "epoch": 7.181050090525045,
+      "grad_norm": 0.4843117296695709,
+      "learning_rate": 8.411870386507724e-05,
+      "loss": 0.09537842869758606,
+      "mean_token_accuracy": 0.9645151920616627,
+      "num_tokens": 6898444.0,
+      "step": 2980
+    },
+    {
+      "epoch": 7.181050090525045,
+      "eval_entropy": 0.27999477783280813,
+      "eval_loss": 0.9031037092208862,
+      "eval_mean_token_accuracy": 0.8261882204018282,
+      "eval_num_tokens": 6898444.0,
+      "eval_runtime": 88.5602,
+      "eval_samples_per_second": 16.034,
+      "eval_steps_per_second": 2.01,
+      "step": 2980
+    },
+    {
+      "entropy": 0.1156751093454659,
+      "epoch": 7.229330114665057,
+      "grad_norm": 0.5297922492027283,
+      "learning_rate": 8.149629351430326e-05,
+      "loss": 0.09203402996063233,
+      "mean_token_accuracy": 0.9689733698964119,
+      "num_tokens": 6947106.0,
+      "step": 3000
+    },
+    {
+      "epoch": 7.229330114665057,
+      "eval_entropy": 0.2740225845508361,
+      "eval_loss": 0.8876440525054932,
+      "eval_mean_token_accuracy": 0.8265116857678703,
+      "eval_num_tokens": 6947106.0,
+      "eval_runtime": 87.7971,
+      "eval_samples_per_second": 16.174,
+      "eval_steps_per_second": 2.027,
+      "step": 3000
+    },
+    {
+      "entropy": 0.12116114404052496,
+      "epoch": 7.27761013880507,
+      "grad_norm": 0.5467398166656494,
+      "learning_rate": 7.890406302404618e-05,
+      "loss": 0.0933056652545929,
+      "mean_token_accuracy": 0.9676455594599247,
+      "num_tokens": 6992784.0,
+      "step": 3020
+    },
+    {
+      "epoch": 7.27761013880507,
+      "eval_entropy": 0.27238342793804876,
+      "eval_loss": 0.9166117310523987,
+      "eval_mean_token_accuracy": 0.825499015242866,
+      "eval_num_tokens": 6992784.0,
+      "eval_runtime": 88.3276,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 3020
+    },
+    {
+      "entropy": 0.11715119928121567,
+      "epoch": 7.325890162945082,
+      "grad_norm": 0.4976702332496643,
+      "learning_rate": 7.634274596487176e-05,
+      "loss": 0.09449659585952759,
+      "mean_token_accuracy": 0.966506478190422,
+      "num_tokens": 7040482.0,
+      "step": 3040
+    },
+    {
+      "epoch": 7.325890162945082,
+      "eval_entropy": 0.2788975556077582,
+      "eval_loss": 0.8833065629005432,
+      "eval_mean_token_accuracy": 0.8269920596915684,
+      "eval_num_tokens": 7040482.0,
+      "eval_runtime": 88.9016,
+      "eval_samples_per_second": 15.973,
+      "eval_steps_per_second": 2.002,
+      "step": 3040
+    },
+    {
+      "entropy": 0.12565911412239075,
+      "epoch": 7.3741701870850935,
+      "grad_norm": 0.5728679895401001,
+      "learning_rate": 7.381306715921025e-05,
+      "loss": 0.09626244902610778,
+      "mean_token_accuracy": 0.9661444149911403,
+      "num_tokens": 7086261.0,
+      "step": 3060
+    },
+    {
+      "epoch": 7.3741701870850935,
+      "eval_entropy": 0.2762492011939542,
+      "eval_loss": 0.8961160778999329,
+      "eval_mean_token_accuracy": 0.826369910092836,
+      "eval_num_tokens": 7086261.0,
+      "eval_runtime": 88.5204,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 3060
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.995756309171016e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3080/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3080/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3080/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3080/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3080/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3080/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3080/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3080/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3080/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3080/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3080/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3080/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3080/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3080/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..2a7b4c5a16d20bf7d57b86c0b8f3b2930f26c3d3
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3080/trainer_state.json
@@ -0,0 +1,3268 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 7.422450211225105,
+  "eval_steps": 20,
+  "global_step": 3080,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    },
+    {
+      "entropy": 0.1326121488586068,
+      "epoch": 7.181050090525045,
+      "grad_norm": 0.4843117296695709,
+      "learning_rate": 8.411870386507724e-05,
+      "loss": 0.09537842869758606,
+      "mean_token_accuracy": 0.9645151920616627,
+      "num_tokens": 6898444.0,
+      "step": 2980
+    },
+    {
+      "epoch": 7.181050090525045,
+      "eval_entropy": 0.27999477783280813,
+      "eval_loss": 0.9031037092208862,
+      "eval_mean_token_accuracy": 0.8261882204018282,
+      "eval_num_tokens": 6898444.0,
+      "eval_runtime": 88.5602,
+      "eval_samples_per_second": 16.034,
+      "eval_steps_per_second": 2.01,
+      "step": 2980
+    },
+    {
+      "entropy": 0.1156751093454659,
+      "epoch": 7.229330114665057,
+      "grad_norm": 0.5297922492027283,
+      "learning_rate": 8.149629351430326e-05,
+      "loss": 0.09203402996063233,
+      "mean_token_accuracy": 0.9689733698964119,
+      "num_tokens": 6947106.0,
+      "step": 3000
+    },
+    {
+      "epoch": 7.229330114665057,
+      "eval_entropy": 0.2740225845508361,
+      "eval_loss": 0.8876440525054932,
+      "eval_mean_token_accuracy": 0.8265116857678703,
+      "eval_num_tokens": 6947106.0,
+      "eval_runtime": 87.7971,
+      "eval_samples_per_second": 16.174,
+      "eval_steps_per_second": 2.027,
+      "step": 3000
+    },
+    {
+      "entropy": 0.12116114404052496,
+      "epoch": 7.27761013880507,
+      "grad_norm": 0.5467398166656494,
+      "learning_rate": 7.890406302404618e-05,
+      "loss": 0.0933056652545929,
+      "mean_token_accuracy": 0.9676455594599247,
+      "num_tokens": 6992784.0,
+      "step": 3020
+    },
+    {
+      "epoch": 7.27761013880507,
+      "eval_entropy": 0.27238342793804876,
+      "eval_loss": 0.9166117310523987,
+      "eval_mean_token_accuracy": 0.825499015242866,
+      "eval_num_tokens": 6992784.0,
+      "eval_runtime": 88.3276,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 3020
+    },
+    {
+      "entropy": 0.11715119928121567,
+      "epoch": 7.325890162945082,
+      "grad_norm": 0.4976702332496643,
+      "learning_rate": 7.634274596487176e-05,
+      "loss": 0.09449659585952759,
+      "mean_token_accuracy": 0.966506478190422,
+      "num_tokens": 7040482.0,
+      "step": 3040
+    },
+    {
+      "epoch": 7.325890162945082,
+      "eval_entropy": 0.2788975556077582,
+      "eval_loss": 0.8833065629005432,
+      "eval_mean_token_accuracy": 0.8269920596915684,
+      "eval_num_tokens": 7040482.0,
+      "eval_runtime": 88.9016,
+      "eval_samples_per_second": 15.973,
+      "eval_steps_per_second": 2.002,
+      "step": 3040
+    },
+    {
+      "entropy": 0.12565911412239075,
+      "epoch": 7.3741701870850935,
+      "grad_norm": 0.5728679895401001,
+      "learning_rate": 7.381306715921025e-05,
+      "loss": 0.09626244902610778,
+      "mean_token_accuracy": 0.9661444149911403,
+      "num_tokens": 7086261.0,
+      "step": 3060
+    },
+    {
+      "epoch": 7.3741701870850935,
+      "eval_entropy": 0.2762492011939542,
+      "eval_loss": 0.8961160778999329,
+      "eval_mean_token_accuracy": 0.826369910092836,
+      "eval_num_tokens": 7086261.0,
+      "eval_runtime": 88.5204,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 3060
+    },
+    {
+      "entropy": 0.12258567921817302,
+      "epoch": 7.422450211225105,
+      "grad_norm": 0.468985915184021,
+      "learning_rate": 7.131574247624031e-05,
+      "loss": 0.09120266437530518,
+      "mean_token_accuracy": 0.9674938447773457,
+      "num_tokens": 7133247.0,
+      "step": 3080
+    },
+    {
+      "epoch": 7.422450211225105,
+      "eval_entropy": 0.2688512086366,
+      "eval_loss": 0.9048654437065125,
+      "eval_mean_token_accuracy": 0.8265582302982888,
+      "eval_num_tokens": 7133247.0,
+      "eval_runtime": 88.164,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 3080
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.015762537055703e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3100/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3100/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3100/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3100/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3100/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3100/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3100/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3100/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3100/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3100/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3100/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3100/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3100/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3100/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..7dfd497e9c3fd8375fa6b3b75d267b555dac6400
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3100/trainer_state.json
@@ -0,0 +1,3289 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 7.470730235365117,
+  "eval_steps": 20,
+  "global_step": 3100,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    },
+    {
+      "entropy": 0.1326121488586068,
+      "epoch": 7.181050090525045,
+      "grad_norm": 0.4843117296695709,
+      "learning_rate": 8.411870386507724e-05,
+      "loss": 0.09537842869758606,
+      "mean_token_accuracy": 0.9645151920616627,
+      "num_tokens": 6898444.0,
+      "step": 2980
+    },
+    {
+      "epoch": 7.181050090525045,
+      "eval_entropy": 0.27999477783280813,
+      "eval_loss": 0.9031037092208862,
+      "eval_mean_token_accuracy": 0.8261882204018282,
+      "eval_num_tokens": 6898444.0,
+      "eval_runtime": 88.5602,
+      "eval_samples_per_second": 16.034,
+      "eval_steps_per_second": 2.01,
+      "step": 2980
+    },
+    {
+      "entropy": 0.1156751093454659,
+      "epoch": 7.229330114665057,
+      "grad_norm": 0.5297922492027283,
+      "learning_rate": 8.149629351430326e-05,
+      "loss": 0.09203402996063233,
+      "mean_token_accuracy": 0.9689733698964119,
+      "num_tokens": 6947106.0,
+      "step": 3000
+    },
+    {
+      "epoch": 7.229330114665057,
+      "eval_entropy": 0.2740225845508361,
+      "eval_loss": 0.8876440525054932,
+      "eval_mean_token_accuracy": 0.8265116857678703,
+      "eval_num_tokens": 6947106.0,
+      "eval_runtime": 87.7971,
+      "eval_samples_per_second": 16.174,
+      "eval_steps_per_second": 2.027,
+      "step": 3000
+    },
+    {
+      "entropy": 0.12116114404052496,
+      "epoch": 7.27761013880507,
+      "grad_norm": 0.5467398166656494,
+      "learning_rate": 7.890406302404618e-05,
+      "loss": 0.0933056652545929,
+      "mean_token_accuracy": 0.9676455594599247,
+      "num_tokens": 6992784.0,
+      "step": 3020
+    },
+    {
+      "epoch": 7.27761013880507,
+      "eval_entropy": 0.27238342793804876,
+      "eval_loss": 0.9166117310523987,
+      "eval_mean_token_accuracy": 0.825499015242866,
+      "eval_num_tokens": 6992784.0,
+      "eval_runtime": 88.3276,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 3020
+    },
+    {
+      "entropy": 0.11715119928121567,
+      "epoch": 7.325890162945082,
+      "grad_norm": 0.4976702332496643,
+      "learning_rate": 7.634274596487176e-05,
+      "loss": 0.09449659585952759,
+      "mean_token_accuracy": 0.966506478190422,
+      "num_tokens": 7040482.0,
+      "step": 3040
+    },
+    {
+      "epoch": 7.325890162945082,
+      "eval_entropy": 0.2788975556077582,
+      "eval_loss": 0.8833065629005432,
+      "eval_mean_token_accuracy": 0.8269920596915684,
+      "eval_num_tokens": 7040482.0,
+      "eval_runtime": 88.9016,
+      "eval_samples_per_second": 15.973,
+      "eval_steps_per_second": 2.002,
+      "step": 3040
+    },
+    {
+      "entropy": 0.12565911412239075,
+      "epoch": 7.3741701870850935,
+      "grad_norm": 0.5728679895401001,
+      "learning_rate": 7.381306715921025e-05,
+      "loss": 0.09626244902610778,
+      "mean_token_accuracy": 0.9661444149911403,
+      "num_tokens": 7086261.0,
+      "step": 3060
+    },
+    {
+      "epoch": 7.3741701870850935,
+      "eval_entropy": 0.2762492011939542,
+      "eval_loss": 0.8961160778999329,
+      "eval_mean_token_accuracy": 0.826369910092836,
+      "eval_num_tokens": 7086261.0,
+      "eval_runtime": 88.5204,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 3060
+    },
+    {
+      "entropy": 0.12258567921817302,
+      "epoch": 7.422450211225105,
+      "grad_norm": 0.468985915184021,
+      "learning_rate": 7.131574247624031e-05,
+      "loss": 0.09120266437530518,
+      "mean_token_accuracy": 0.9674938447773457,
+      "num_tokens": 7133247.0,
+      "step": 3080
+    },
+    {
+      "epoch": 7.422450211225105,
+      "eval_entropy": 0.2688512086366,
+      "eval_loss": 0.9048654437065125,
+      "eval_mean_token_accuracy": 0.8265582302982888,
+      "eval_num_tokens": 7133247.0,
+      "eval_runtime": 88.164,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 3080
+    },
+    {
+      "entropy": 0.12070536343380808,
+      "epoch": 7.470730235365117,
+      "grad_norm": 0.5310841798782349,
+      "learning_rate": 6.885147862930688e-05,
+      "loss": 0.09386247992515565,
+      "mean_token_accuracy": 0.96664222702384,
+      "num_tokens": 7177810.0,
+      "step": 3100
+    },
+    {
+      "epoch": 7.470730235365117,
+      "eval_entropy": 0.26289632874593305,
+      "eval_loss": 0.902837336063385,
+      "eval_mean_token_accuracy": 0.828052184554968,
+      "eval_num_tokens": 7177810.0,
+      "eval_runtime": 88.1201,
+      "eval_samples_per_second": 16.114,
+      "eval_steps_per_second": 2.02,
+      "step": 3100
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.034539812194898e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3120/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3120/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3120/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3120/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3120/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3120/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3120/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3120/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3120/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3120/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3120/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3120/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3120/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3120/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..042fc9852a9ce37e7c10e7a57550f3ecf23579bf
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3120/trainer_state.json
@@ -0,0 +1,3310 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 7.51901025950513,
+  "eval_steps": 20,
+  "global_step": 3120,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    },
+    {
+      "entropy": 0.1326121488586068,
+      "epoch": 7.181050090525045,
+      "grad_norm": 0.4843117296695709,
+      "learning_rate": 8.411870386507724e-05,
+      "loss": 0.09537842869758606,
+      "mean_token_accuracy": 0.9645151920616627,
+      "num_tokens": 6898444.0,
+      "step": 2980
+    },
+    {
+      "epoch": 7.181050090525045,
+      "eval_entropy": 0.27999477783280813,
+      "eval_loss": 0.9031037092208862,
+      "eval_mean_token_accuracy": 0.8261882204018282,
+      "eval_num_tokens": 6898444.0,
+      "eval_runtime": 88.5602,
+      "eval_samples_per_second": 16.034,
+      "eval_steps_per_second": 2.01,
+      "step": 2980
+    },
+    {
+      "entropy": 0.1156751093454659,
+      "epoch": 7.229330114665057,
+      "grad_norm": 0.5297922492027283,
+      "learning_rate": 8.149629351430326e-05,
+      "loss": 0.09203402996063233,
+      "mean_token_accuracy": 0.9689733698964119,
+      "num_tokens": 6947106.0,
+      "step": 3000
+    },
+    {
+      "epoch": 7.229330114665057,
+      "eval_entropy": 0.2740225845508361,
+      "eval_loss": 0.8876440525054932,
+      "eval_mean_token_accuracy": 0.8265116857678703,
+      "eval_num_tokens": 6947106.0,
+      "eval_runtime": 87.7971,
+      "eval_samples_per_second": 16.174,
+      "eval_steps_per_second": 2.027,
+      "step": 3000
+    },
+    {
+      "entropy": 0.12116114404052496,
+      "epoch": 7.27761013880507,
+      "grad_norm": 0.5467398166656494,
+      "learning_rate": 7.890406302404618e-05,
+      "loss": 0.0933056652545929,
+      "mean_token_accuracy": 0.9676455594599247,
+      "num_tokens": 6992784.0,
+      "step": 3020
+    },
+    {
+      "epoch": 7.27761013880507,
+      "eval_entropy": 0.27238342793804876,
+      "eval_loss": 0.9166117310523987,
+      "eval_mean_token_accuracy": 0.825499015242866,
+      "eval_num_tokens": 6992784.0,
+      "eval_runtime": 88.3276,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 3020
+    },
+    {
+      "entropy": 0.11715119928121567,
+      "epoch": 7.325890162945082,
+      "grad_norm": 0.4976702332496643,
+      "learning_rate": 7.634274596487176e-05,
+      "loss": 0.09449659585952759,
+      "mean_token_accuracy": 0.966506478190422,
+      "num_tokens": 7040482.0,
+      "step": 3040
+    },
+    {
+      "epoch": 7.325890162945082,
+      "eval_entropy": 0.2788975556077582,
+      "eval_loss": 0.8833065629005432,
+      "eval_mean_token_accuracy": 0.8269920596915684,
+      "eval_num_tokens": 7040482.0,
+      "eval_runtime": 88.9016,
+      "eval_samples_per_second": 15.973,
+      "eval_steps_per_second": 2.002,
+      "step": 3040
+    },
+    {
+      "entropy": 0.12565911412239075,
+      "epoch": 7.3741701870850935,
+      "grad_norm": 0.5728679895401001,
+      "learning_rate": 7.381306715921025e-05,
+      "loss": 0.09626244902610778,
+      "mean_token_accuracy": 0.9661444149911403,
+      "num_tokens": 7086261.0,
+      "step": 3060
+    },
+    {
+      "epoch": 7.3741701870850935,
+      "eval_entropy": 0.2762492011939542,
+      "eval_loss": 0.8961160778999329,
+      "eval_mean_token_accuracy": 0.826369910092836,
+      "eval_num_tokens": 7086261.0,
+      "eval_runtime": 88.5204,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 3060
+    },
+    {
+      "entropy": 0.12258567921817302,
+      "epoch": 7.422450211225105,
+      "grad_norm": 0.468985915184021,
+      "learning_rate": 7.131574247624031e-05,
+      "loss": 0.09120266437530518,
+      "mean_token_accuracy": 0.9674938447773457,
+      "num_tokens": 7133247.0,
+      "step": 3080
+    },
+    {
+      "epoch": 7.422450211225105,
+      "eval_entropy": 0.2688512086366,
+      "eval_loss": 0.9048654437065125,
+      "eval_mean_token_accuracy": 0.8265582302982888,
+      "eval_num_tokens": 7133247.0,
+      "eval_runtime": 88.164,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 3080
+    },
+    {
+      "entropy": 0.12070536343380808,
+      "epoch": 7.470730235365117,
+      "grad_norm": 0.5310841798782349,
+      "learning_rate": 6.885147862930688e-05,
+      "loss": 0.09386247992515565,
+      "mean_token_accuracy": 0.96664222702384,
+      "num_tokens": 7177810.0,
+      "step": 3100
+    },
+    {
+      "epoch": 7.470730235365117,
+      "eval_entropy": 0.26289632874593305,
+      "eval_loss": 0.902837336063385,
+      "eval_mean_token_accuracy": 0.828052184554968,
+      "eval_num_tokens": 7177810.0,
+      "eval_runtime": 88.1201,
+      "eval_samples_per_second": 16.114,
+      "eval_steps_per_second": 2.02,
+      "step": 3100
+    },
+    {
+      "entropy": 0.11322665181942285,
+      "epoch": 7.51901025950513,
+      "grad_norm": 0.6846067905426025,
+      "learning_rate": 6.642097297592908e-05,
+      "loss": 0.08772538304328918,
+      "mean_token_accuracy": 0.9691362135112286,
+      "num_tokens": 7225302.0,
+      "step": 3120
+    },
+    {
+      "epoch": 7.51901025950513,
+      "eval_entropy": 0.2670605613776807,
+      "eval_loss": 0.9130566120147705,
+      "eval_mean_token_accuracy": 0.826366903741708,
+      "eval_num_tokens": 7225302.0,
+      "eval_runtime": 88.6035,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3120
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.0549048135791616e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3140/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3140/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3140/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3140/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3140/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3140/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3140/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3140/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3140/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3140/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3140/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3140/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3140/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3140/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..5bcb1354e8685276cb0a128d96fe86a3c28424c5
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3140/trainer_state.json
@@ -0,0 +1,3331 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 7.567290283645142,
+  "eval_steps": 20,
+  "global_step": 3140,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    },
+    {
+      "entropy": 0.1326121488586068,
+      "epoch": 7.181050090525045,
+      "grad_norm": 0.4843117296695709,
+      "learning_rate": 8.411870386507724e-05,
+      "loss": 0.09537842869758606,
+      "mean_token_accuracy": 0.9645151920616627,
+      "num_tokens": 6898444.0,
+      "step": 2980
+    },
+    {
+      "epoch": 7.181050090525045,
+      "eval_entropy": 0.27999477783280813,
+      "eval_loss": 0.9031037092208862,
+      "eval_mean_token_accuracy": 0.8261882204018282,
+      "eval_num_tokens": 6898444.0,
+      "eval_runtime": 88.5602,
+      "eval_samples_per_second": 16.034,
+      "eval_steps_per_second": 2.01,
+      "step": 2980
+    },
+    {
+      "entropy": 0.1156751093454659,
+      "epoch": 7.229330114665057,
+      "grad_norm": 0.5297922492027283,
+      "learning_rate": 8.149629351430326e-05,
+      "loss": 0.09203402996063233,
+      "mean_token_accuracy": 0.9689733698964119,
+      "num_tokens": 6947106.0,
+      "step": 3000
+    },
+    {
+      "epoch": 7.229330114665057,
+      "eval_entropy": 0.2740225845508361,
+      "eval_loss": 0.8876440525054932,
+      "eval_mean_token_accuracy": 0.8265116857678703,
+      "eval_num_tokens": 6947106.0,
+      "eval_runtime": 87.7971,
+      "eval_samples_per_second": 16.174,
+      "eval_steps_per_second": 2.027,
+      "step": 3000
+    },
+    {
+      "entropy": 0.12116114404052496,
+      "epoch": 7.27761013880507,
+      "grad_norm": 0.5467398166656494,
+      "learning_rate": 7.890406302404618e-05,
+      "loss": 0.0933056652545929,
+      "mean_token_accuracy": 0.9676455594599247,
+      "num_tokens": 6992784.0,
+      "step": 3020
+    },
+    {
+      "epoch": 7.27761013880507,
+      "eval_entropy": 0.27238342793804876,
+      "eval_loss": 0.9166117310523987,
+      "eval_mean_token_accuracy": 0.825499015242866,
+      "eval_num_tokens": 6992784.0,
+      "eval_runtime": 88.3276,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 3020
+    },
+    {
+      "entropy": 0.11715119928121567,
+      "epoch": 7.325890162945082,
+      "grad_norm": 0.4976702332496643,
+      "learning_rate": 7.634274596487176e-05,
+      "loss": 0.09449659585952759,
+      "mean_token_accuracy": 0.966506478190422,
+      "num_tokens": 7040482.0,
+      "step": 3040
+    },
+    {
+      "epoch": 7.325890162945082,
+      "eval_entropy": 0.2788975556077582,
+      "eval_loss": 0.8833065629005432,
+      "eval_mean_token_accuracy": 0.8269920596915684,
+      "eval_num_tokens": 7040482.0,
+      "eval_runtime": 88.9016,
+      "eval_samples_per_second": 15.973,
+      "eval_steps_per_second": 2.002,
+      "step": 3040
+    },
+    {
+      "entropy": 0.12565911412239075,
+      "epoch": 7.3741701870850935,
+      "grad_norm": 0.5728679895401001,
+      "learning_rate": 7.381306715921025e-05,
+      "loss": 0.09626244902610778,
+      "mean_token_accuracy": 0.9661444149911403,
+      "num_tokens": 7086261.0,
+      "step": 3060
+    },
+    {
+      "epoch": 7.3741701870850935,
+      "eval_entropy": 0.2762492011939542,
+      "eval_loss": 0.8961160778999329,
+      "eval_mean_token_accuracy": 0.826369910092836,
+      "eval_num_tokens": 7086261.0,
+      "eval_runtime": 88.5204,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 3060
+    },
+    {
+      "entropy": 0.12258567921817302,
+      "epoch": 7.422450211225105,
+      "grad_norm": 0.468985915184021,
+      "learning_rate": 7.131574247624031e-05,
+      "loss": 0.09120266437530518,
+      "mean_token_accuracy": 0.9674938447773457,
+      "num_tokens": 7133247.0,
+      "step": 3080
+    },
+    {
+      "epoch": 7.422450211225105,
+      "eval_entropy": 0.2688512086366,
+      "eval_loss": 0.9048654437065125,
+      "eval_mean_token_accuracy": 0.8265582302982888,
+      "eval_num_tokens": 7133247.0,
+      "eval_runtime": 88.164,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 3080
+    },
+    {
+      "entropy": 0.12070536343380808,
+      "epoch": 7.470730235365117,
+      "grad_norm": 0.5310841798782349,
+      "learning_rate": 6.885147862930688e-05,
+      "loss": 0.09386247992515565,
+      "mean_token_accuracy": 0.96664222702384,
+      "num_tokens": 7177810.0,
+      "step": 3100
+    },
+    {
+      "epoch": 7.470730235365117,
+      "eval_entropy": 0.26289632874593305,
+      "eval_loss": 0.902837336063385,
+      "eval_mean_token_accuracy": 0.828052184554968,
+      "eval_num_tokens": 7177810.0,
+      "eval_runtime": 88.1201,
+      "eval_samples_per_second": 16.114,
+      "eval_steps_per_second": 2.02,
+      "step": 3100
+    },
+    {
+      "entropy": 0.11322665181942285,
+      "epoch": 7.51901025950513,
+      "grad_norm": 0.6846067905426025,
+      "learning_rate": 6.642097297592908e-05,
+      "loss": 0.08772538304328918,
+      "mean_token_accuracy": 0.9691362135112286,
+      "num_tokens": 7225302.0,
+      "step": 3120
+    },
+    {
+      "epoch": 7.51901025950513,
+      "eval_entropy": 0.2670605613776807,
+      "eval_loss": 0.9130566120147705,
+      "eval_mean_token_accuracy": 0.826366903741708,
+      "eval_num_tokens": 7225302.0,
+      "eval_runtime": 88.6035,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3120
+    },
+    {
+      "entropy": 0.10985926580615342,
+      "epoch": 7.567290283645142,
+      "grad_norm": 0.4799748361110687,
+      "learning_rate": 6.402491332045683e-05,
+      "loss": 0.08672296404838561,
+      "mean_token_accuracy": 0.9698824048042297,
+      "num_tokens": 7274881.0,
+      "step": 3140
+    },
+    {
+      "epoch": 7.567290283645142,
+      "eval_entropy": 0.26657010866015146,
+      "eval_loss": 0.9056380987167358,
+      "eval_mean_token_accuracy": 0.8272326461384806,
+      "eval_num_tokens": 7274881.0,
+      "eval_runtime": 88.2967,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3140
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.076369819776307e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3160/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3160/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3160/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3160/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3160/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3160/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3160/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3160/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3160/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3160/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3160/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3160/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3160/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3160/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..775448264fd381dfe599d9c8c3bfac8b25ae5684
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3160/trainer_state.json
@@ -0,0 +1,3352 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 7.615570307785154,
+  "eval_steps": 20,
+  "global_step": 3160,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    },
+    {
+      "entropy": 0.1326121488586068,
+      "epoch": 7.181050090525045,
+      "grad_norm": 0.4843117296695709,
+      "learning_rate": 8.411870386507724e-05,
+      "loss": 0.09537842869758606,
+      "mean_token_accuracy": 0.9645151920616627,
+      "num_tokens": 6898444.0,
+      "step": 2980
+    },
+    {
+      "epoch": 7.181050090525045,
+      "eval_entropy": 0.27999477783280813,
+      "eval_loss": 0.9031037092208862,
+      "eval_mean_token_accuracy": 0.8261882204018282,
+      "eval_num_tokens": 6898444.0,
+      "eval_runtime": 88.5602,
+      "eval_samples_per_second": 16.034,
+      "eval_steps_per_second": 2.01,
+      "step": 2980
+    },
+    {
+      "entropy": 0.1156751093454659,
+      "epoch": 7.229330114665057,
+      "grad_norm": 0.5297922492027283,
+      "learning_rate": 8.149629351430326e-05,
+      "loss": 0.09203402996063233,
+      "mean_token_accuracy": 0.9689733698964119,
+      "num_tokens": 6947106.0,
+      "step": 3000
+    },
+    {
+      "epoch": 7.229330114665057,
+      "eval_entropy": 0.2740225845508361,
+      "eval_loss": 0.8876440525054932,
+      "eval_mean_token_accuracy": 0.8265116857678703,
+      "eval_num_tokens": 6947106.0,
+      "eval_runtime": 87.7971,
+      "eval_samples_per_second": 16.174,
+      "eval_steps_per_second": 2.027,
+      "step": 3000
+    },
+    {
+      "entropy": 0.12116114404052496,
+      "epoch": 7.27761013880507,
+      "grad_norm": 0.5467398166656494,
+      "learning_rate": 7.890406302404618e-05,
+      "loss": 0.0933056652545929,
+      "mean_token_accuracy": 0.9676455594599247,
+      "num_tokens": 6992784.0,
+      "step": 3020
+    },
+    {
+      "epoch": 7.27761013880507,
+      "eval_entropy": 0.27238342793804876,
+      "eval_loss": 0.9166117310523987,
+      "eval_mean_token_accuracy": 0.825499015242866,
+      "eval_num_tokens": 6992784.0,
+      "eval_runtime": 88.3276,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 3020
+    },
+    {
+      "entropy": 0.11715119928121567,
+      "epoch": 7.325890162945082,
+      "grad_norm": 0.4976702332496643,
+      "learning_rate": 7.634274596487176e-05,
+      "loss": 0.09449659585952759,
+      "mean_token_accuracy": 0.966506478190422,
+      "num_tokens": 7040482.0,
+      "step": 3040
+    },
+    {
+      "epoch": 7.325890162945082,
+      "eval_entropy": 0.2788975556077582,
+      "eval_loss": 0.8833065629005432,
+      "eval_mean_token_accuracy": 0.8269920596915684,
+      "eval_num_tokens": 7040482.0,
+      "eval_runtime": 88.9016,
+      "eval_samples_per_second": 15.973,
+      "eval_steps_per_second": 2.002,
+      "step": 3040
+    },
+    {
+      "entropy": 0.12565911412239075,
+      "epoch": 7.3741701870850935,
+      "grad_norm": 0.5728679895401001,
+      "learning_rate": 7.381306715921025e-05,
+      "loss": 0.09626244902610778,
+      "mean_token_accuracy": 0.9661444149911403,
+      "num_tokens": 7086261.0,
+      "step": 3060
+    },
+    {
+      "epoch": 7.3741701870850935,
+      "eval_entropy": 0.2762492011939542,
+      "eval_loss": 0.8961160778999329,
+      "eval_mean_token_accuracy": 0.826369910092836,
+      "eval_num_tokens": 7086261.0,
+      "eval_runtime": 88.5204,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 3060
+    },
+    {
+      "entropy": 0.12258567921817302,
+      "epoch": 7.422450211225105,
+      "grad_norm": 0.468985915184021,
+      "learning_rate": 7.131574247624031e-05,
+      "loss": 0.09120266437530518,
+      "mean_token_accuracy": 0.9674938447773457,
+      "num_tokens": 7133247.0,
+      "step": 3080
+    },
+    {
+      "epoch": 7.422450211225105,
+      "eval_entropy": 0.2688512086366,
+      "eval_loss": 0.9048654437065125,
+      "eval_mean_token_accuracy": 0.8265582302982888,
+      "eval_num_tokens": 7133247.0,
+      "eval_runtime": 88.164,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 3080
+    },
+    {
+      "entropy": 0.12070536343380808,
+      "epoch": 7.470730235365117,
+      "grad_norm": 0.5310841798782349,
+      "learning_rate": 6.885147862930688e-05,
+      "loss": 0.09386247992515565,
+      "mean_token_accuracy": 0.96664222702384,
+      "num_tokens": 7177810.0,
+      "step": 3100
+    },
+    {
+      "epoch": 7.470730235365117,
+      "eval_entropy": 0.26289632874593305,
+      "eval_loss": 0.902837336063385,
+      "eval_mean_token_accuracy": 0.828052184554968,
+      "eval_num_tokens": 7177810.0,
+      "eval_runtime": 88.1201,
+      "eval_samples_per_second": 16.114,
+      "eval_steps_per_second": 2.02,
+      "step": 3100
+    },
+    {
+      "entropy": 0.11322665181942285,
+      "epoch": 7.51901025950513,
+      "grad_norm": 0.6846067905426025,
+      "learning_rate": 6.642097297592908e-05,
+      "loss": 0.08772538304328918,
+      "mean_token_accuracy": 0.9691362135112286,
+      "num_tokens": 7225302.0,
+      "step": 3120
+    },
+    {
+      "epoch": 7.51901025950513,
+      "eval_entropy": 0.2670605613776807,
+      "eval_loss": 0.9130566120147705,
+      "eval_mean_token_accuracy": 0.826366903741708,
+      "eval_num_tokens": 7225302.0,
+      "eval_runtime": 88.6035,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3120
+    },
+    {
+      "entropy": 0.10985926580615342,
+      "epoch": 7.567290283645142,
+      "grad_norm": 0.4799748361110687,
+      "learning_rate": 6.402491332045683e-05,
+      "loss": 0.08672296404838561,
+      "mean_token_accuracy": 0.9698824048042297,
+      "num_tokens": 7274881.0,
+      "step": 3140
+    },
+    {
+      "epoch": 7.567290283645142,
+      "eval_entropy": 0.26657010866015146,
+      "eval_loss": 0.9056380987167358,
+      "eval_mean_token_accuracy": 0.8272326461384806,
+      "eval_num_tokens": 7274881.0,
+      "eval_runtime": 88.2967,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3140
+    },
+    {
+      "entropy": 0.11890473761595785,
+      "epoch": 7.615570307785154,
+      "grad_norm": 0.4081912934780121,
+      "learning_rate": 6.166397771942987e-05,
+      "loss": 0.09396827220916748,
+      "mean_token_accuracy": 0.967512384802103,
+      "num_tokens": 7321163.0,
+      "step": 3160
+    },
+    {
+      "epoch": 7.615570307785154,
+      "eval_entropy": 0.26364776575833226,
+      "eval_loss": 0.9204733967781067,
+      "eval_mean_token_accuracy": 0.827419729714983,
+      "eval_num_tokens": 7321163.0,
+      "eval_runtime": 88.3441,
+      "eval_samples_per_second": 16.074,
+      "eval_steps_per_second": 2.015,
+      "step": 3160
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.095971659926753e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3180/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3180/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3180/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3180/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3180/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3180/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3180/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3180/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3180/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3180/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3180/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3180/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3180/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3180/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..2c70abb0b86d1b62a66f938d28608c353d94b669
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3180/trainer_state.json
@@ -0,0 +1,3373 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 7.663850331925166,
+  "eval_steps": 20,
+  "global_step": 3180,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    },
+    {
+      "entropy": 0.1326121488586068,
+      "epoch": 7.181050090525045,
+      "grad_norm": 0.4843117296695709,
+      "learning_rate": 8.411870386507724e-05,
+      "loss": 0.09537842869758606,
+      "mean_token_accuracy": 0.9645151920616627,
+      "num_tokens": 6898444.0,
+      "step": 2980
+    },
+    {
+      "epoch": 7.181050090525045,
+      "eval_entropy": 0.27999477783280813,
+      "eval_loss": 0.9031037092208862,
+      "eval_mean_token_accuracy": 0.8261882204018282,
+      "eval_num_tokens": 6898444.0,
+      "eval_runtime": 88.5602,
+      "eval_samples_per_second": 16.034,
+      "eval_steps_per_second": 2.01,
+      "step": 2980
+    },
+    {
+      "entropy": 0.1156751093454659,
+      "epoch": 7.229330114665057,
+      "grad_norm": 0.5297922492027283,
+      "learning_rate": 8.149629351430326e-05,
+      "loss": 0.09203402996063233,
+      "mean_token_accuracy": 0.9689733698964119,
+      "num_tokens": 6947106.0,
+      "step": 3000
+    },
+    {
+      "epoch": 7.229330114665057,
+      "eval_entropy": 0.2740225845508361,
+      "eval_loss": 0.8876440525054932,
+      "eval_mean_token_accuracy": 0.8265116857678703,
+      "eval_num_tokens": 6947106.0,
+      "eval_runtime": 87.7971,
+      "eval_samples_per_second": 16.174,
+      "eval_steps_per_second": 2.027,
+      "step": 3000
+    },
+    {
+      "entropy": 0.12116114404052496,
+      "epoch": 7.27761013880507,
+      "grad_norm": 0.5467398166656494,
+      "learning_rate": 7.890406302404618e-05,
+      "loss": 0.0933056652545929,
+      "mean_token_accuracy": 0.9676455594599247,
+      "num_tokens": 6992784.0,
+      "step": 3020
+    },
+    {
+      "epoch": 7.27761013880507,
+      "eval_entropy": 0.27238342793804876,
+      "eval_loss": 0.9166117310523987,
+      "eval_mean_token_accuracy": 0.825499015242866,
+      "eval_num_tokens": 6992784.0,
+      "eval_runtime": 88.3276,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 3020
+    },
+    {
+      "entropy": 0.11715119928121567,
+      "epoch": 7.325890162945082,
+      "grad_norm": 0.4976702332496643,
+      "learning_rate": 7.634274596487176e-05,
+      "loss": 0.09449659585952759,
+      "mean_token_accuracy": 0.966506478190422,
+      "num_tokens": 7040482.0,
+      "step": 3040
+    },
+    {
+      "epoch": 7.325890162945082,
+      "eval_entropy": 0.2788975556077582,
+      "eval_loss": 0.8833065629005432,
+      "eval_mean_token_accuracy": 0.8269920596915684,
+      "eval_num_tokens": 7040482.0,
+      "eval_runtime": 88.9016,
+      "eval_samples_per_second": 15.973,
+      "eval_steps_per_second": 2.002,
+      "step": 3040
+    },
+    {
+      "entropy": 0.12565911412239075,
+      "epoch": 7.3741701870850935,
+      "grad_norm": 0.5728679895401001,
+      "learning_rate": 7.381306715921025e-05,
+      "loss": 0.09626244902610778,
+      "mean_token_accuracy": 0.9661444149911403,
+      "num_tokens": 7086261.0,
+      "step": 3060
+    },
+    {
+      "epoch": 7.3741701870850935,
+      "eval_entropy": 0.2762492011939542,
+      "eval_loss": 0.8961160778999329,
+      "eval_mean_token_accuracy": 0.826369910092836,
+      "eval_num_tokens": 7086261.0,
+      "eval_runtime": 88.5204,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 3060
+    },
+    {
+      "entropy": 0.12258567921817302,
+      "epoch": 7.422450211225105,
+      "grad_norm": 0.468985915184021,
+      "learning_rate": 7.131574247624031e-05,
+      "loss": 0.09120266437530518,
+      "mean_token_accuracy": 0.9674938447773457,
+      "num_tokens": 7133247.0,
+      "step": 3080
+    },
+    {
+      "epoch": 7.422450211225105,
+      "eval_entropy": 0.2688512086366,
+      "eval_loss": 0.9048654437065125,
+      "eval_mean_token_accuracy": 0.8265582302982888,
+      "eval_num_tokens": 7133247.0,
+      "eval_runtime": 88.164,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 3080
+    },
+    {
+      "entropy": 0.12070536343380808,
+      "epoch": 7.470730235365117,
+      "grad_norm": 0.5310841798782349,
+      "learning_rate": 6.885147862930688e-05,
+      "loss": 0.09386247992515565,
+      "mean_token_accuracy": 0.96664222702384,
+      "num_tokens": 7177810.0,
+      "step": 3100
+    },
+    {
+      "epoch": 7.470730235365117,
+      "eval_entropy": 0.26289632874593305,
+      "eval_loss": 0.902837336063385,
+      "eval_mean_token_accuracy": 0.828052184554968,
+      "eval_num_tokens": 7177810.0,
+      "eval_runtime": 88.1201,
+      "eval_samples_per_second": 16.114,
+      "eval_steps_per_second": 2.02,
+      "step": 3100
+    },
+    {
+      "entropy": 0.11322665181942285,
+      "epoch": 7.51901025950513,
+      "grad_norm": 0.6846067905426025,
+      "learning_rate": 6.642097297592908e-05,
+      "loss": 0.08772538304328918,
+      "mean_token_accuracy": 0.9691362135112286,
+      "num_tokens": 7225302.0,
+      "step": 3120
+    },
+    {
+      "epoch": 7.51901025950513,
+      "eval_entropy": 0.2670605613776807,
+      "eval_loss": 0.9130566120147705,
+      "eval_mean_token_accuracy": 0.826366903741708,
+      "eval_num_tokens": 7225302.0,
+      "eval_runtime": 88.6035,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3120
+    },
+    {
+      "entropy": 0.10985926580615342,
+      "epoch": 7.567290283645142,
+      "grad_norm": 0.4799748361110687,
+      "learning_rate": 6.402491332045683e-05,
+      "loss": 0.08672296404838561,
+      "mean_token_accuracy": 0.9698824048042297,
+      "num_tokens": 7274881.0,
+      "step": 3140
+    },
+    {
+      "epoch": 7.567290283645142,
+      "eval_entropy": 0.26657010866015146,
+      "eval_loss": 0.9056380987167358,
+      "eval_mean_token_accuracy": 0.8272326461384806,
+      "eval_num_tokens": 7274881.0,
+      "eval_runtime": 88.2967,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3140
+    },
+    {
+      "entropy": 0.11890473761595785,
+      "epoch": 7.615570307785154,
+      "grad_norm": 0.4081912934780121,
+      "learning_rate": 6.166397771942987e-05,
+      "loss": 0.09396827220916748,
+      "mean_token_accuracy": 0.967512384802103,
+      "num_tokens": 7321163.0,
+      "step": 3160
+    },
+    {
+      "epoch": 7.615570307785154,
+      "eval_entropy": 0.26364776575833226,
+      "eval_loss": 0.9204733967781067,
+      "eval_mean_token_accuracy": 0.827419729714983,
+      "eval_num_tokens": 7321163.0,
+      "eval_runtime": 88.3441,
+      "eval_samples_per_second": 16.074,
+      "eval_steps_per_second": 2.015,
+      "step": 3160
+    },
+    {
+      "entropy": 0.12209134344011545,
+      "epoch": 7.663850331925166,
+      "grad_norm": 0.5159774422645569,
+      "learning_rate": 5.93388342896962e-05,
+      "loss": 0.09292978644371033,
+      "mean_token_accuracy": 0.9658421002328396,
+      "num_tokens": 7366217.0,
+      "step": 3180
+    },
+    {
+      "epoch": 7.663850331925166,
+      "eval_entropy": 0.2714110068725736,
+      "eval_loss": 0.9143092036247253,
+      "eval_mean_token_accuracy": 0.8262251099843657,
+      "eval_num_tokens": 7366217.0,
+      "eval_runtime": 87.9635,
+      "eval_samples_per_second": 16.143,
+      "eval_steps_per_second": 2.024,
+      "step": 3180
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.115362095643464e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-320/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-320/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-320/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-320/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-320/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-320/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-320/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-320/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-320/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-320/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-320/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-320/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-320/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-320/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..40d6676d519fcec5ea94ebcefa75c8df50db5e20
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-320/trainer_state.json
@@ -0,0 +1,370 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.7724803862401931,
+  "eval_steps": 20,
+  "global_step": 320,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.1722770376474624e+16,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3200/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3200/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3200/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3200/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3200/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3200/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3200/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3200/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3200/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3200/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3200/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3200/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3200/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3200/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..252b2efb2a6db9926dfbdc1e3bccd627e1590a83
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3200/trainer_state.json
@@ -0,0 +1,3394 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 7.712130356065178,
+  "eval_steps": 20,
+  "global_step": 3200,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    },
+    {
+      "entropy": 0.1326121488586068,
+      "epoch": 7.181050090525045,
+      "grad_norm": 0.4843117296695709,
+      "learning_rate": 8.411870386507724e-05,
+      "loss": 0.09537842869758606,
+      "mean_token_accuracy": 0.9645151920616627,
+      "num_tokens": 6898444.0,
+      "step": 2980
+    },
+    {
+      "epoch": 7.181050090525045,
+      "eval_entropy": 0.27999477783280813,
+      "eval_loss": 0.9031037092208862,
+      "eval_mean_token_accuracy": 0.8261882204018282,
+      "eval_num_tokens": 6898444.0,
+      "eval_runtime": 88.5602,
+      "eval_samples_per_second": 16.034,
+      "eval_steps_per_second": 2.01,
+      "step": 2980
+    },
+    {
+      "entropy": 0.1156751093454659,
+      "epoch": 7.229330114665057,
+      "grad_norm": 0.5297922492027283,
+      "learning_rate": 8.149629351430326e-05,
+      "loss": 0.09203402996063233,
+      "mean_token_accuracy": 0.9689733698964119,
+      "num_tokens": 6947106.0,
+      "step": 3000
+    },
+    {
+      "epoch": 7.229330114665057,
+      "eval_entropy": 0.2740225845508361,
+      "eval_loss": 0.8876440525054932,
+      "eval_mean_token_accuracy": 0.8265116857678703,
+      "eval_num_tokens": 6947106.0,
+      "eval_runtime": 87.7971,
+      "eval_samples_per_second": 16.174,
+      "eval_steps_per_second": 2.027,
+      "step": 3000
+    },
+    {
+      "entropy": 0.12116114404052496,
+      "epoch": 7.27761013880507,
+      "grad_norm": 0.5467398166656494,
+      "learning_rate": 7.890406302404618e-05,
+      "loss": 0.0933056652545929,
+      "mean_token_accuracy": 0.9676455594599247,
+      "num_tokens": 6992784.0,
+      "step": 3020
+    },
+    {
+      "epoch": 7.27761013880507,
+      "eval_entropy": 0.27238342793804876,
+      "eval_loss": 0.9166117310523987,
+      "eval_mean_token_accuracy": 0.825499015242866,
+      "eval_num_tokens": 6992784.0,
+      "eval_runtime": 88.3276,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 3020
+    },
+    {
+      "entropy": 0.11715119928121567,
+      "epoch": 7.325890162945082,
+      "grad_norm": 0.4976702332496643,
+      "learning_rate": 7.634274596487176e-05,
+      "loss": 0.09449659585952759,
+      "mean_token_accuracy": 0.966506478190422,
+      "num_tokens": 7040482.0,
+      "step": 3040
+    },
+    {
+      "epoch": 7.325890162945082,
+      "eval_entropy": 0.2788975556077582,
+      "eval_loss": 0.8833065629005432,
+      "eval_mean_token_accuracy": 0.8269920596915684,
+      "eval_num_tokens": 7040482.0,
+      "eval_runtime": 88.9016,
+      "eval_samples_per_second": 15.973,
+      "eval_steps_per_second": 2.002,
+      "step": 3040
+    },
+    {
+      "entropy": 0.12565911412239075,
+      "epoch": 7.3741701870850935,
+      "grad_norm": 0.5728679895401001,
+      "learning_rate": 7.381306715921025e-05,
+      "loss": 0.09626244902610778,
+      "mean_token_accuracy": 0.9661444149911403,
+      "num_tokens": 7086261.0,
+      "step": 3060
+    },
+    {
+      "epoch": 7.3741701870850935,
+      "eval_entropy": 0.2762492011939542,
+      "eval_loss": 0.8961160778999329,
+      "eval_mean_token_accuracy": 0.826369910092836,
+      "eval_num_tokens": 7086261.0,
+      "eval_runtime": 88.5204,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 3060
+    },
+    {
+      "entropy": 0.12258567921817302,
+      "epoch": 7.422450211225105,
+      "grad_norm": 0.468985915184021,
+      "learning_rate": 7.131574247624031e-05,
+      "loss": 0.09120266437530518,
+      "mean_token_accuracy": 0.9674938447773457,
+      "num_tokens": 7133247.0,
+      "step": 3080
+    },
+    {
+      "epoch": 7.422450211225105,
+      "eval_entropy": 0.2688512086366,
+      "eval_loss": 0.9048654437065125,
+      "eval_mean_token_accuracy": 0.8265582302982888,
+      "eval_num_tokens": 7133247.0,
+      "eval_runtime": 88.164,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 3080
+    },
+    {
+      "entropy": 0.12070536343380808,
+      "epoch": 7.470730235365117,
+      "grad_norm": 0.5310841798782349,
+      "learning_rate": 6.885147862930688e-05,
+      "loss": 0.09386247992515565,
+      "mean_token_accuracy": 0.96664222702384,
+      "num_tokens": 7177810.0,
+      "step": 3100
+    },
+    {
+      "epoch": 7.470730235365117,
+      "eval_entropy": 0.26289632874593305,
+      "eval_loss": 0.902837336063385,
+      "eval_mean_token_accuracy": 0.828052184554968,
+      "eval_num_tokens": 7177810.0,
+      "eval_runtime": 88.1201,
+      "eval_samples_per_second": 16.114,
+      "eval_steps_per_second": 2.02,
+      "step": 3100
+    },
+    {
+      "entropy": 0.11322665181942285,
+      "epoch": 7.51901025950513,
+      "grad_norm": 0.6846067905426025,
+      "learning_rate": 6.642097297592908e-05,
+      "loss": 0.08772538304328918,
+      "mean_token_accuracy": 0.9691362135112286,
+      "num_tokens": 7225302.0,
+      "step": 3120
+    },
+    {
+      "epoch": 7.51901025950513,
+      "eval_entropy": 0.2670605613776807,
+      "eval_loss": 0.9130566120147705,
+      "eval_mean_token_accuracy": 0.826366903741708,
+      "eval_num_tokens": 7225302.0,
+      "eval_runtime": 88.6035,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3120
+    },
+    {
+      "entropy": 0.10985926580615342,
+      "epoch": 7.567290283645142,
+      "grad_norm": 0.4799748361110687,
+      "learning_rate": 6.402491332045683e-05,
+      "loss": 0.08672296404838561,
+      "mean_token_accuracy": 0.9698824048042297,
+      "num_tokens": 7274881.0,
+      "step": 3140
+    },
+    {
+      "epoch": 7.567290283645142,
+      "eval_entropy": 0.26657010866015146,
+      "eval_loss": 0.9056380987167358,
+      "eval_mean_token_accuracy": 0.8272326461384806,
+      "eval_num_tokens": 7274881.0,
+      "eval_runtime": 88.2967,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3140
+    },
+    {
+      "entropy": 0.11890473761595785,
+      "epoch": 7.615570307785154,
+      "grad_norm": 0.4081912934780121,
+      "learning_rate": 6.166397771942987e-05,
+      "loss": 0.09396827220916748,
+      "mean_token_accuracy": 0.967512384802103,
+      "num_tokens": 7321163.0,
+      "step": 3160
+    },
+    {
+      "epoch": 7.615570307785154,
+      "eval_entropy": 0.26364776575833226,
+      "eval_loss": 0.9204733967781067,
+      "eval_mean_token_accuracy": 0.827419729714983,
+      "eval_num_tokens": 7321163.0,
+      "eval_runtime": 88.3441,
+      "eval_samples_per_second": 16.074,
+      "eval_steps_per_second": 2.015,
+      "step": 3160
+    },
+    {
+      "entropy": 0.12209134344011545,
+      "epoch": 7.663850331925166,
+      "grad_norm": 0.5159774422645569,
+      "learning_rate": 5.93388342896962e-05,
+      "loss": 0.09292978644371033,
+      "mean_token_accuracy": 0.9658421002328396,
+      "num_tokens": 7366217.0,
+      "step": 3180
+    },
+    {
+      "epoch": 7.663850331925166,
+      "eval_entropy": 0.2714110068725736,
+      "eval_loss": 0.9143092036247253,
+      "eval_mean_token_accuracy": 0.8262251099843657,
+      "eval_num_tokens": 7366217.0,
+      "eval_runtime": 87.9635,
+      "eval_samples_per_second": 16.143,
+      "eval_steps_per_second": 2.024,
+      "step": 3180
+    },
+    {
+      "entropy": 0.11955609126016498,
+      "epoch": 7.712130356065178,
+      "grad_norm": 0.7005976438522339,
+      "learning_rate": 5.705014101934252e-05,
+      "loss": 0.09319776892662049,
+      "mean_token_accuracy": 0.9674513377249241,
+      "num_tokens": 7410887.0,
+      "step": 3200
+    },
+    {
+      "epoch": 7.712130356065178,
+      "eval_entropy": 0.2626291513442993,
+      "eval_loss": 0.91884446144104,
+      "eval_mean_token_accuracy": 0.827563955207889,
+      "eval_num_tokens": 7410887.0,
+      "eval_runtime": 87.8849,
+      "eval_samples_per_second": 16.157,
+      "eval_steps_per_second": 2.025,
+      "step": 3200
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.13355691209386e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3220/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3220/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3220/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3220/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3220/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3220/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3220/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3220/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3220/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3220/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3220/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3220/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3220/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3220/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..7e84d39b83ead9f531b7f0686ab9d5cfe3971e42
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3220/trainer_state.json
@@ -0,0 +1,3415 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 7.76041038020519,
+  "eval_steps": 20,
+  "global_step": 3220,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    },
+    {
+      "entropy": 0.1326121488586068,
+      "epoch": 7.181050090525045,
+      "grad_norm": 0.4843117296695709,
+      "learning_rate": 8.411870386507724e-05,
+      "loss": 0.09537842869758606,
+      "mean_token_accuracy": 0.9645151920616627,
+      "num_tokens": 6898444.0,
+      "step": 2980
+    },
+    {
+      "epoch": 7.181050090525045,
+      "eval_entropy": 0.27999477783280813,
+      "eval_loss": 0.9031037092208862,
+      "eval_mean_token_accuracy": 0.8261882204018282,
+      "eval_num_tokens": 6898444.0,
+      "eval_runtime": 88.5602,
+      "eval_samples_per_second": 16.034,
+      "eval_steps_per_second": 2.01,
+      "step": 2980
+    },
+    {
+      "entropy": 0.1156751093454659,
+      "epoch": 7.229330114665057,
+      "grad_norm": 0.5297922492027283,
+      "learning_rate": 8.149629351430326e-05,
+      "loss": 0.09203402996063233,
+      "mean_token_accuracy": 0.9689733698964119,
+      "num_tokens": 6947106.0,
+      "step": 3000
+    },
+    {
+      "epoch": 7.229330114665057,
+      "eval_entropy": 0.2740225845508361,
+      "eval_loss": 0.8876440525054932,
+      "eval_mean_token_accuracy": 0.8265116857678703,
+      "eval_num_tokens": 6947106.0,
+      "eval_runtime": 87.7971,
+      "eval_samples_per_second": 16.174,
+      "eval_steps_per_second": 2.027,
+      "step": 3000
+    },
+    {
+      "entropy": 0.12116114404052496,
+      "epoch": 7.27761013880507,
+      "grad_norm": 0.5467398166656494,
+      "learning_rate": 7.890406302404618e-05,
+      "loss": 0.0933056652545929,
+      "mean_token_accuracy": 0.9676455594599247,
+      "num_tokens": 6992784.0,
+      "step": 3020
+    },
+    {
+      "epoch": 7.27761013880507,
+      "eval_entropy": 0.27238342793804876,
+      "eval_loss": 0.9166117310523987,
+      "eval_mean_token_accuracy": 0.825499015242866,
+      "eval_num_tokens": 6992784.0,
+      "eval_runtime": 88.3276,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 3020
+    },
+    {
+      "entropy": 0.11715119928121567,
+      "epoch": 7.325890162945082,
+      "grad_norm": 0.4976702332496643,
+      "learning_rate": 7.634274596487176e-05,
+      "loss": 0.09449659585952759,
+      "mean_token_accuracy": 0.966506478190422,
+      "num_tokens": 7040482.0,
+      "step": 3040
+    },
+    {
+      "epoch": 7.325890162945082,
+      "eval_entropy": 0.2788975556077582,
+      "eval_loss": 0.8833065629005432,
+      "eval_mean_token_accuracy": 0.8269920596915684,
+      "eval_num_tokens": 7040482.0,
+      "eval_runtime": 88.9016,
+      "eval_samples_per_second": 15.973,
+      "eval_steps_per_second": 2.002,
+      "step": 3040
+    },
+    {
+      "entropy": 0.12565911412239075,
+      "epoch": 7.3741701870850935,
+      "grad_norm": 0.5728679895401001,
+      "learning_rate": 7.381306715921025e-05,
+      "loss": 0.09626244902610778,
+      "mean_token_accuracy": 0.9661444149911403,
+      "num_tokens": 7086261.0,
+      "step": 3060
+    },
+    {
+      "epoch": 7.3741701870850935,
+      "eval_entropy": 0.2762492011939542,
+      "eval_loss": 0.8961160778999329,
+      "eval_mean_token_accuracy": 0.826369910092836,
+      "eval_num_tokens": 7086261.0,
+      "eval_runtime": 88.5204,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 3060
+    },
+    {
+      "entropy": 0.12258567921817302,
+      "epoch": 7.422450211225105,
+      "grad_norm": 0.468985915184021,
+      "learning_rate": 7.131574247624031e-05,
+      "loss": 0.09120266437530518,
+      "mean_token_accuracy": 0.9674938447773457,
+      "num_tokens": 7133247.0,
+      "step": 3080
+    },
+    {
+      "epoch": 7.422450211225105,
+      "eval_entropy": 0.2688512086366,
+      "eval_loss": 0.9048654437065125,
+      "eval_mean_token_accuracy": 0.8265582302982888,
+      "eval_num_tokens": 7133247.0,
+      "eval_runtime": 88.164,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 3080
+    },
+    {
+      "entropy": 0.12070536343380808,
+      "epoch": 7.470730235365117,
+      "grad_norm": 0.5310841798782349,
+      "learning_rate": 6.885147862930688e-05,
+      "loss": 0.09386247992515565,
+      "mean_token_accuracy": 0.96664222702384,
+      "num_tokens": 7177810.0,
+      "step": 3100
+    },
+    {
+      "epoch": 7.470730235365117,
+      "eval_entropy": 0.26289632874593305,
+      "eval_loss": 0.902837336063385,
+      "eval_mean_token_accuracy": 0.828052184554968,
+      "eval_num_tokens": 7177810.0,
+      "eval_runtime": 88.1201,
+      "eval_samples_per_second": 16.114,
+      "eval_steps_per_second": 2.02,
+      "step": 3100
+    },
+    {
+      "entropy": 0.11322665181942285,
+      "epoch": 7.51901025950513,
+      "grad_norm": 0.6846067905426025,
+      "learning_rate": 6.642097297592908e-05,
+      "loss": 0.08772538304328918,
+      "mean_token_accuracy": 0.9691362135112286,
+      "num_tokens": 7225302.0,
+      "step": 3120
+    },
+    {
+      "epoch": 7.51901025950513,
+      "eval_entropy": 0.2670605613776807,
+      "eval_loss": 0.9130566120147705,
+      "eval_mean_token_accuracy": 0.826366903741708,
+      "eval_num_tokens": 7225302.0,
+      "eval_runtime": 88.6035,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3120
+    },
+    {
+      "entropy": 0.10985926580615342,
+      "epoch": 7.567290283645142,
+      "grad_norm": 0.4799748361110687,
+      "learning_rate": 6.402491332045683e-05,
+      "loss": 0.08672296404838561,
+      "mean_token_accuracy": 0.9698824048042297,
+      "num_tokens": 7274881.0,
+      "step": 3140
+    },
+    {
+      "epoch": 7.567290283645142,
+      "eval_entropy": 0.26657010866015146,
+      "eval_loss": 0.9056380987167358,
+      "eval_mean_token_accuracy": 0.8272326461384806,
+      "eval_num_tokens": 7274881.0,
+      "eval_runtime": 88.2967,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3140
+    },
+    {
+      "entropy": 0.11890473761595785,
+      "epoch": 7.615570307785154,
+      "grad_norm": 0.4081912934780121,
+      "learning_rate": 6.166397771942987e-05,
+      "loss": 0.09396827220916748,
+      "mean_token_accuracy": 0.967512384802103,
+      "num_tokens": 7321163.0,
+      "step": 3160
+    },
+    {
+      "epoch": 7.615570307785154,
+      "eval_entropy": 0.26364776575833226,
+      "eval_loss": 0.9204733967781067,
+      "eval_mean_token_accuracy": 0.827419729714983,
+      "eval_num_tokens": 7321163.0,
+      "eval_runtime": 88.3441,
+      "eval_samples_per_second": 16.074,
+      "eval_steps_per_second": 2.015,
+      "step": 3160
+    },
+    {
+      "entropy": 0.12209134344011545,
+      "epoch": 7.663850331925166,
+      "grad_norm": 0.5159774422645569,
+      "learning_rate": 5.93388342896962e-05,
+      "loss": 0.09292978644371033,
+      "mean_token_accuracy": 0.9658421002328396,
+      "num_tokens": 7366217.0,
+      "step": 3180
+    },
+    {
+      "epoch": 7.663850331925166,
+      "eval_entropy": 0.2714110068725736,
+      "eval_loss": 0.9143092036247253,
+      "eval_mean_token_accuracy": 0.8262251099843657,
+      "eval_num_tokens": 7366217.0,
+      "eval_runtime": 87.9635,
+      "eval_samples_per_second": 16.143,
+      "eval_steps_per_second": 2.024,
+      "step": 3180
+    },
+    {
+      "entropy": 0.11955609126016498,
+      "epoch": 7.712130356065178,
+      "grad_norm": 0.7005976438522339,
+      "learning_rate": 5.705014101934252e-05,
+      "loss": 0.09319776892662049,
+      "mean_token_accuracy": 0.9674513377249241,
+      "num_tokens": 7410887.0,
+      "step": 3200
+    },
+    {
+      "epoch": 7.712130356065178,
+      "eval_entropy": 0.2626291513442993,
+      "eval_loss": 0.91884446144104,
+      "eval_mean_token_accuracy": 0.827563955207889,
+      "eval_num_tokens": 7410887.0,
+      "eval_runtime": 87.8849,
+      "eval_samples_per_second": 16.157,
+      "eval_steps_per_second": 2.025,
+      "step": 3200
+    },
+    {
+      "entropy": 0.11767532899975777,
+      "epoch": 7.76041038020519,
+      "grad_norm": 0.4675362706184387,
+      "learning_rate": 5.479854558149187e-05,
+      "loss": 0.09038087725639343,
+      "mean_token_accuracy": 0.9669680051505566,
+      "num_tokens": 7457578.0,
+      "step": 3220
+    },
+    {
+      "epoch": 7.76041038020519,
+      "eval_entropy": 0.26581037203582486,
+      "eval_loss": 0.9146883487701416,
+      "eval_mean_token_accuracy": 0.8270345812433222,
+      "eval_num_tokens": 7457578.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 3220
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.153286822980096e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3240/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3240/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3240/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3240/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3240/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3240/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3240/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3240/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3240/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3240/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3240/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3240/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3240/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3240/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..67fa39ea74ee718a4081090762b2ef362f896168
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3240/trainer_state.json
@@ -0,0 +1,3436 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 7.808690404345202,
+  "eval_steps": 20,
+  "global_step": 3240,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    },
+    {
+      "entropy": 0.1326121488586068,
+      "epoch": 7.181050090525045,
+      "grad_norm": 0.4843117296695709,
+      "learning_rate": 8.411870386507724e-05,
+      "loss": 0.09537842869758606,
+      "mean_token_accuracy": 0.9645151920616627,
+      "num_tokens": 6898444.0,
+      "step": 2980
+    },
+    {
+      "epoch": 7.181050090525045,
+      "eval_entropy": 0.27999477783280813,
+      "eval_loss": 0.9031037092208862,
+      "eval_mean_token_accuracy": 0.8261882204018282,
+      "eval_num_tokens": 6898444.0,
+      "eval_runtime": 88.5602,
+      "eval_samples_per_second": 16.034,
+      "eval_steps_per_second": 2.01,
+      "step": 2980
+    },
+    {
+      "entropy": 0.1156751093454659,
+      "epoch": 7.229330114665057,
+      "grad_norm": 0.5297922492027283,
+      "learning_rate": 8.149629351430326e-05,
+      "loss": 0.09203402996063233,
+      "mean_token_accuracy": 0.9689733698964119,
+      "num_tokens": 6947106.0,
+      "step": 3000
+    },
+    {
+      "epoch": 7.229330114665057,
+      "eval_entropy": 0.2740225845508361,
+      "eval_loss": 0.8876440525054932,
+      "eval_mean_token_accuracy": 0.8265116857678703,
+      "eval_num_tokens": 6947106.0,
+      "eval_runtime": 87.7971,
+      "eval_samples_per_second": 16.174,
+      "eval_steps_per_second": 2.027,
+      "step": 3000
+    },
+    {
+      "entropy": 0.12116114404052496,
+      "epoch": 7.27761013880507,
+      "grad_norm": 0.5467398166656494,
+      "learning_rate": 7.890406302404618e-05,
+      "loss": 0.0933056652545929,
+      "mean_token_accuracy": 0.9676455594599247,
+      "num_tokens": 6992784.0,
+      "step": 3020
+    },
+    {
+      "epoch": 7.27761013880507,
+      "eval_entropy": 0.27238342793804876,
+      "eval_loss": 0.9166117310523987,
+      "eval_mean_token_accuracy": 0.825499015242866,
+      "eval_num_tokens": 6992784.0,
+      "eval_runtime": 88.3276,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 3020
+    },
+    {
+      "entropy": 0.11715119928121567,
+      "epoch": 7.325890162945082,
+      "grad_norm": 0.4976702332496643,
+      "learning_rate": 7.634274596487176e-05,
+      "loss": 0.09449659585952759,
+      "mean_token_accuracy": 0.966506478190422,
+      "num_tokens": 7040482.0,
+      "step": 3040
+    },
+    {
+      "epoch": 7.325890162945082,
+      "eval_entropy": 0.2788975556077582,
+      "eval_loss": 0.8833065629005432,
+      "eval_mean_token_accuracy": 0.8269920596915684,
+      "eval_num_tokens": 7040482.0,
+      "eval_runtime": 88.9016,
+      "eval_samples_per_second": 15.973,
+      "eval_steps_per_second": 2.002,
+      "step": 3040
+    },
+    {
+      "entropy": 0.12565911412239075,
+      "epoch": 7.3741701870850935,
+      "grad_norm": 0.5728679895401001,
+      "learning_rate": 7.381306715921025e-05,
+      "loss": 0.09626244902610778,
+      "mean_token_accuracy": 0.9661444149911403,
+      "num_tokens": 7086261.0,
+      "step": 3060
+    },
+    {
+      "epoch": 7.3741701870850935,
+      "eval_entropy": 0.2762492011939542,
+      "eval_loss": 0.8961160778999329,
+      "eval_mean_token_accuracy": 0.826369910092836,
+      "eval_num_tokens": 7086261.0,
+      "eval_runtime": 88.5204,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 3060
+    },
+    {
+      "entropy": 0.12258567921817302,
+      "epoch": 7.422450211225105,
+      "grad_norm": 0.468985915184021,
+      "learning_rate": 7.131574247624031e-05,
+      "loss": 0.09120266437530518,
+      "mean_token_accuracy": 0.9674938447773457,
+      "num_tokens": 7133247.0,
+      "step": 3080
+    },
+    {
+      "epoch": 7.422450211225105,
+      "eval_entropy": 0.2688512086366,
+      "eval_loss": 0.9048654437065125,
+      "eval_mean_token_accuracy": 0.8265582302982888,
+      "eval_num_tokens": 7133247.0,
+      "eval_runtime": 88.164,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 3080
+    },
+    {
+      "entropy": 0.12070536343380808,
+      "epoch": 7.470730235365117,
+      "grad_norm": 0.5310841798782349,
+      "learning_rate": 6.885147862930688e-05,
+      "loss": 0.09386247992515565,
+      "mean_token_accuracy": 0.96664222702384,
+      "num_tokens": 7177810.0,
+      "step": 3100
+    },
+    {
+      "epoch": 7.470730235365117,
+      "eval_entropy": 0.26289632874593305,
+      "eval_loss": 0.902837336063385,
+      "eval_mean_token_accuracy": 0.828052184554968,
+      "eval_num_tokens": 7177810.0,
+      "eval_runtime": 88.1201,
+      "eval_samples_per_second": 16.114,
+      "eval_steps_per_second": 2.02,
+      "step": 3100
+    },
+    {
+      "entropy": 0.11322665181942285,
+      "epoch": 7.51901025950513,
+      "grad_norm": 0.6846067905426025,
+      "learning_rate": 6.642097297592908e-05,
+      "loss": 0.08772538304328918,
+      "mean_token_accuracy": 0.9691362135112286,
+      "num_tokens": 7225302.0,
+      "step": 3120
+    },
+    {
+      "epoch": 7.51901025950513,
+      "eval_entropy": 0.2670605613776807,
+      "eval_loss": 0.9130566120147705,
+      "eval_mean_token_accuracy": 0.826366903741708,
+      "eval_num_tokens": 7225302.0,
+      "eval_runtime": 88.6035,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3120
+    },
+    {
+      "entropy": 0.10985926580615342,
+      "epoch": 7.567290283645142,
+      "grad_norm": 0.4799748361110687,
+      "learning_rate": 6.402491332045683e-05,
+      "loss": 0.08672296404838561,
+      "mean_token_accuracy": 0.9698824048042297,
+      "num_tokens": 7274881.0,
+      "step": 3140
+    },
+    {
+      "epoch": 7.567290283645142,
+      "eval_entropy": 0.26657010866015146,
+      "eval_loss": 0.9056380987167358,
+      "eval_mean_token_accuracy": 0.8272326461384806,
+      "eval_num_tokens": 7274881.0,
+      "eval_runtime": 88.2967,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3140
+    },
+    {
+      "entropy": 0.11890473761595785,
+      "epoch": 7.615570307785154,
+      "grad_norm": 0.4081912934780121,
+      "learning_rate": 6.166397771942987e-05,
+      "loss": 0.09396827220916748,
+      "mean_token_accuracy": 0.967512384802103,
+      "num_tokens": 7321163.0,
+      "step": 3160
+    },
+    {
+      "epoch": 7.615570307785154,
+      "eval_entropy": 0.26364776575833226,
+      "eval_loss": 0.9204733967781067,
+      "eval_mean_token_accuracy": 0.827419729714983,
+      "eval_num_tokens": 7321163.0,
+      "eval_runtime": 88.3441,
+      "eval_samples_per_second": 16.074,
+      "eval_steps_per_second": 2.015,
+      "step": 3160
+    },
+    {
+      "entropy": 0.12209134344011545,
+      "epoch": 7.663850331925166,
+      "grad_norm": 0.5159774422645569,
+      "learning_rate": 5.93388342896962e-05,
+      "loss": 0.09292978644371033,
+      "mean_token_accuracy": 0.9658421002328396,
+      "num_tokens": 7366217.0,
+      "step": 3180
+    },
+    {
+      "epoch": 7.663850331925166,
+      "eval_entropy": 0.2714110068725736,
+      "eval_loss": 0.9143092036247253,
+      "eval_mean_token_accuracy": 0.8262251099843657,
+      "eval_num_tokens": 7366217.0,
+      "eval_runtime": 87.9635,
+      "eval_samples_per_second": 16.143,
+      "eval_steps_per_second": 2.024,
+      "step": 3180
+    },
+    {
+      "entropy": 0.11955609126016498,
+      "epoch": 7.712130356065178,
+      "grad_norm": 0.7005976438522339,
+      "learning_rate": 5.705014101934252e-05,
+      "loss": 0.09319776892662049,
+      "mean_token_accuracy": 0.9674513377249241,
+      "num_tokens": 7410887.0,
+      "step": 3200
+    },
+    {
+      "epoch": 7.712130356065178,
+      "eval_entropy": 0.2626291513442993,
+      "eval_loss": 0.91884446144104,
+      "eval_mean_token_accuracy": 0.827563955207889,
+      "eval_num_tokens": 7410887.0,
+      "eval_runtime": 87.8849,
+      "eval_samples_per_second": 16.157,
+      "eval_steps_per_second": 2.025,
+      "step": 3200
+    },
+    {
+      "entropy": 0.11767532899975777,
+      "epoch": 7.76041038020519,
+      "grad_norm": 0.4675362706184387,
+      "learning_rate": 5.479854558149187e-05,
+      "loss": 0.09038087725639343,
+      "mean_token_accuracy": 0.9669680051505566,
+      "num_tokens": 7457578.0,
+      "step": 3220
+    },
+    {
+      "epoch": 7.76041038020519,
+      "eval_entropy": 0.26581037203582486,
+      "eval_loss": 0.9146883487701416,
+      "eval_mean_token_accuracy": 0.8270345812433222,
+      "eval_num_tokens": 7457578.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 3220
+    },
+    {
+      "entropy": 0.11383061073720455,
+      "epoch": 7.808690404345202,
+      "grad_norm": 0.541247546672821,
+      "learning_rate": 5.258468515101918e-05,
+      "loss": 0.08942405581474304,
+      "mean_token_accuracy": 0.968398705124855,
+      "num_tokens": 7505761.0,
+      "step": 3240
+    },
+    {
+      "epoch": 7.808690404345202,
+      "eval_entropy": 0.26687998960861997,
+      "eval_loss": 0.9037085175514221,
+      "eval_mean_token_accuracy": 0.8274977863504646,
+      "eval_num_tokens": 7505761.0,
+      "eval_runtime": 88.5163,
+      "eval_samples_per_second": 16.042,
+      "eval_steps_per_second": 2.011,
+      "step": 3240
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.174001825895731e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3260/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3260/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3260/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3260/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3260/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3260/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3260/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3260/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3260/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3260/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3260/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3260/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3260/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3260/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..adb1dabaf9974867e381829bd49c87124e1d9622
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3260/trainer_state.json
@@ -0,0 +1,3457 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 7.856970428485214,
+  "eval_steps": 20,
+  "global_step": 3260,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    },
+    {
+      "entropy": 0.1326121488586068,
+      "epoch": 7.181050090525045,
+      "grad_norm": 0.4843117296695709,
+      "learning_rate": 8.411870386507724e-05,
+      "loss": 0.09537842869758606,
+      "mean_token_accuracy": 0.9645151920616627,
+      "num_tokens": 6898444.0,
+      "step": 2980
+    },
+    {
+      "epoch": 7.181050090525045,
+      "eval_entropy": 0.27999477783280813,
+      "eval_loss": 0.9031037092208862,
+      "eval_mean_token_accuracy": 0.8261882204018282,
+      "eval_num_tokens": 6898444.0,
+      "eval_runtime": 88.5602,
+      "eval_samples_per_second": 16.034,
+      "eval_steps_per_second": 2.01,
+      "step": 2980
+    },
+    {
+      "entropy": 0.1156751093454659,
+      "epoch": 7.229330114665057,
+      "grad_norm": 0.5297922492027283,
+      "learning_rate": 8.149629351430326e-05,
+      "loss": 0.09203402996063233,
+      "mean_token_accuracy": 0.9689733698964119,
+      "num_tokens": 6947106.0,
+      "step": 3000
+    },
+    {
+      "epoch": 7.229330114665057,
+      "eval_entropy": 0.2740225845508361,
+      "eval_loss": 0.8876440525054932,
+      "eval_mean_token_accuracy": 0.8265116857678703,
+      "eval_num_tokens": 6947106.0,
+      "eval_runtime": 87.7971,
+      "eval_samples_per_second": 16.174,
+      "eval_steps_per_second": 2.027,
+      "step": 3000
+    },
+    {
+      "entropy": 0.12116114404052496,
+      "epoch": 7.27761013880507,
+      "grad_norm": 0.5467398166656494,
+      "learning_rate": 7.890406302404618e-05,
+      "loss": 0.0933056652545929,
+      "mean_token_accuracy": 0.9676455594599247,
+      "num_tokens": 6992784.0,
+      "step": 3020
+    },
+    {
+      "epoch": 7.27761013880507,
+      "eval_entropy": 0.27238342793804876,
+      "eval_loss": 0.9166117310523987,
+      "eval_mean_token_accuracy": 0.825499015242866,
+      "eval_num_tokens": 6992784.0,
+      "eval_runtime": 88.3276,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 3020
+    },
+    {
+      "entropy": 0.11715119928121567,
+      "epoch": 7.325890162945082,
+      "grad_norm": 0.4976702332496643,
+      "learning_rate": 7.634274596487176e-05,
+      "loss": 0.09449659585952759,
+      "mean_token_accuracy": 0.966506478190422,
+      "num_tokens": 7040482.0,
+      "step": 3040
+    },
+    {
+      "epoch": 7.325890162945082,
+      "eval_entropy": 0.2788975556077582,
+      "eval_loss": 0.8833065629005432,
+      "eval_mean_token_accuracy": 0.8269920596915684,
+      "eval_num_tokens": 7040482.0,
+      "eval_runtime": 88.9016,
+      "eval_samples_per_second": 15.973,
+      "eval_steps_per_second": 2.002,
+      "step": 3040
+    },
+    {
+      "entropy": 0.12565911412239075,
+      "epoch": 7.3741701870850935,
+      "grad_norm": 0.5728679895401001,
+      "learning_rate": 7.381306715921025e-05,
+      "loss": 0.09626244902610778,
+      "mean_token_accuracy": 0.9661444149911403,
+      "num_tokens": 7086261.0,
+      "step": 3060
+    },
+    {
+      "epoch": 7.3741701870850935,
+      "eval_entropy": 0.2762492011939542,
+      "eval_loss": 0.8961160778999329,
+      "eval_mean_token_accuracy": 0.826369910092836,
+      "eval_num_tokens": 7086261.0,
+      "eval_runtime": 88.5204,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 3060
+    },
+    {
+      "entropy": 0.12258567921817302,
+      "epoch": 7.422450211225105,
+      "grad_norm": 0.468985915184021,
+      "learning_rate": 7.131574247624031e-05,
+      "loss": 0.09120266437530518,
+      "mean_token_accuracy": 0.9674938447773457,
+      "num_tokens": 7133247.0,
+      "step": 3080
+    },
+    {
+      "epoch": 7.422450211225105,
+      "eval_entropy": 0.2688512086366,
+      "eval_loss": 0.9048654437065125,
+      "eval_mean_token_accuracy": 0.8265582302982888,
+      "eval_num_tokens": 7133247.0,
+      "eval_runtime": 88.164,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 3080
+    },
+    {
+      "entropy": 0.12070536343380808,
+      "epoch": 7.470730235365117,
+      "grad_norm": 0.5310841798782349,
+      "learning_rate": 6.885147862930688e-05,
+      "loss": 0.09386247992515565,
+      "mean_token_accuracy": 0.96664222702384,
+      "num_tokens": 7177810.0,
+      "step": 3100
+    },
+    {
+      "epoch": 7.470730235365117,
+      "eval_entropy": 0.26289632874593305,
+      "eval_loss": 0.902837336063385,
+      "eval_mean_token_accuracy": 0.828052184554968,
+      "eval_num_tokens": 7177810.0,
+      "eval_runtime": 88.1201,
+      "eval_samples_per_second": 16.114,
+      "eval_steps_per_second": 2.02,
+      "step": 3100
+    },
+    {
+      "entropy": 0.11322665181942285,
+      "epoch": 7.51901025950513,
+      "grad_norm": 0.6846067905426025,
+      "learning_rate": 6.642097297592908e-05,
+      "loss": 0.08772538304328918,
+      "mean_token_accuracy": 0.9691362135112286,
+      "num_tokens": 7225302.0,
+      "step": 3120
+    },
+    {
+      "epoch": 7.51901025950513,
+      "eval_entropy": 0.2670605613776807,
+      "eval_loss": 0.9130566120147705,
+      "eval_mean_token_accuracy": 0.826366903741708,
+      "eval_num_tokens": 7225302.0,
+      "eval_runtime": 88.6035,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3120
+    },
+    {
+      "entropy": 0.10985926580615342,
+      "epoch": 7.567290283645142,
+      "grad_norm": 0.4799748361110687,
+      "learning_rate": 6.402491332045683e-05,
+      "loss": 0.08672296404838561,
+      "mean_token_accuracy": 0.9698824048042297,
+      "num_tokens": 7274881.0,
+      "step": 3140
+    },
+    {
+      "epoch": 7.567290283645142,
+      "eval_entropy": 0.26657010866015146,
+      "eval_loss": 0.9056380987167358,
+      "eval_mean_token_accuracy": 0.8272326461384806,
+      "eval_num_tokens": 7274881.0,
+      "eval_runtime": 88.2967,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3140
+    },
+    {
+      "entropy": 0.11890473761595785,
+      "epoch": 7.615570307785154,
+      "grad_norm": 0.4081912934780121,
+      "learning_rate": 6.166397771942987e-05,
+      "loss": 0.09396827220916748,
+      "mean_token_accuracy": 0.967512384802103,
+      "num_tokens": 7321163.0,
+      "step": 3160
+    },
+    {
+      "epoch": 7.615570307785154,
+      "eval_entropy": 0.26364776575833226,
+      "eval_loss": 0.9204733967781067,
+      "eval_mean_token_accuracy": 0.827419729714983,
+      "eval_num_tokens": 7321163.0,
+      "eval_runtime": 88.3441,
+      "eval_samples_per_second": 16.074,
+      "eval_steps_per_second": 2.015,
+      "step": 3160
+    },
+    {
+      "entropy": 0.12209134344011545,
+      "epoch": 7.663850331925166,
+      "grad_norm": 0.5159774422645569,
+      "learning_rate": 5.93388342896962e-05,
+      "loss": 0.09292978644371033,
+      "mean_token_accuracy": 0.9658421002328396,
+      "num_tokens": 7366217.0,
+      "step": 3180
+    },
+    {
+      "epoch": 7.663850331925166,
+      "eval_entropy": 0.2714110068725736,
+      "eval_loss": 0.9143092036247253,
+      "eval_mean_token_accuracy": 0.8262251099843657,
+      "eval_num_tokens": 7366217.0,
+      "eval_runtime": 87.9635,
+      "eval_samples_per_second": 16.143,
+      "eval_steps_per_second": 2.024,
+      "step": 3180
+    },
+    {
+      "entropy": 0.11955609126016498,
+      "epoch": 7.712130356065178,
+      "grad_norm": 0.7005976438522339,
+      "learning_rate": 5.705014101934252e-05,
+      "loss": 0.09319776892662049,
+      "mean_token_accuracy": 0.9674513377249241,
+      "num_tokens": 7410887.0,
+      "step": 3200
+    },
+    {
+      "epoch": 7.712130356065178,
+      "eval_entropy": 0.2626291513442993,
+      "eval_loss": 0.91884446144104,
+      "eval_mean_token_accuracy": 0.827563955207889,
+      "eval_num_tokens": 7410887.0,
+      "eval_runtime": 87.8849,
+      "eval_samples_per_second": 16.157,
+      "eval_steps_per_second": 2.025,
+      "step": 3200
+    },
+    {
+      "entropy": 0.11767532899975777,
+      "epoch": 7.76041038020519,
+      "grad_norm": 0.4675362706184387,
+      "learning_rate": 5.479854558149187e-05,
+      "loss": 0.09038087725639343,
+      "mean_token_accuracy": 0.9669680051505566,
+      "num_tokens": 7457578.0,
+      "step": 3220
+    },
+    {
+      "epoch": 7.76041038020519,
+      "eval_entropy": 0.26581037203582486,
+      "eval_loss": 0.9146883487701416,
+      "eval_mean_token_accuracy": 0.8270345812433222,
+      "eval_num_tokens": 7457578.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 3220
+    },
+    {
+      "entropy": 0.11383061073720455,
+      "epoch": 7.808690404345202,
+      "grad_norm": 0.541247546672821,
+      "learning_rate": 5.258468515101918e-05,
+      "loss": 0.08942405581474304,
+      "mean_token_accuracy": 0.968398705124855,
+      "num_tokens": 7505761.0,
+      "step": 3240
+    },
+    {
+      "epoch": 7.808690404345202,
+      "eval_entropy": 0.26687998960861997,
+      "eval_loss": 0.9037085175514221,
+      "eval_mean_token_accuracy": 0.8274977863504646,
+      "eval_num_tokens": 7505761.0,
+      "eval_runtime": 88.5163,
+      "eval_samples_per_second": 16.042,
+      "eval_steps_per_second": 2.011,
+      "step": 3240
+    },
+    {
+      "entropy": 0.12004058873280883,
+      "epoch": 7.856970428485214,
+      "grad_norm": 0.5289552211761475,
+      "learning_rate": 5.040918622423899e-05,
+      "loss": 0.09064805507659912,
+      "mean_token_accuracy": 0.9673330299556255,
+      "num_tokens": 7552557.0,
+      "step": 3260
+    },
+    {
+      "epoch": 7.856970428485214,
+      "eval_entropy": 0.26288379251622085,
+      "eval_loss": 0.9146178364753723,
+      "eval_mean_token_accuracy": 0.8276662448149049,
+      "eval_num_tokens": 7552557.0,
+      "eval_runtime": 88.0723,
+      "eval_samples_per_second": 16.123,
+      "eval_steps_per_second": 2.021,
+      "step": 3260
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.1936536662649446e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3280/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3280/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3280/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3280/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3280/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3280/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3280/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3280/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3280/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3280/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3280/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3280/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3280/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3280/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..fb825e240dda35ae0a0bf35dc9686f1079adf38c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3280/trainer_state.json
@@ -0,0 +1,3478 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 7.905250452625227,
+  "eval_steps": 20,
+  "global_step": 3280,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    },
+    {
+      "entropy": 0.1326121488586068,
+      "epoch": 7.181050090525045,
+      "grad_norm": 0.4843117296695709,
+      "learning_rate": 8.411870386507724e-05,
+      "loss": 0.09537842869758606,
+      "mean_token_accuracy": 0.9645151920616627,
+      "num_tokens": 6898444.0,
+      "step": 2980
+    },
+    {
+      "epoch": 7.181050090525045,
+      "eval_entropy": 0.27999477783280813,
+      "eval_loss": 0.9031037092208862,
+      "eval_mean_token_accuracy": 0.8261882204018282,
+      "eval_num_tokens": 6898444.0,
+      "eval_runtime": 88.5602,
+      "eval_samples_per_second": 16.034,
+      "eval_steps_per_second": 2.01,
+      "step": 2980
+    },
+    {
+      "entropy": 0.1156751093454659,
+      "epoch": 7.229330114665057,
+      "grad_norm": 0.5297922492027283,
+      "learning_rate": 8.149629351430326e-05,
+      "loss": 0.09203402996063233,
+      "mean_token_accuracy": 0.9689733698964119,
+      "num_tokens": 6947106.0,
+      "step": 3000
+    },
+    {
+      "epoch": 7.229330114665057,
+      "eval_entropy": 0.2740225845508361,
+      "eval_loss": 0.8876440525054932,
+      "eval_mean_token_accuracy": 0.8265116857678703,
+      "eval_num_tokens": 6947106.0,
+      "eval_runtime": 87.7971,
+      "eval_samples_per_second": 16.174,
+      "eval_steps_per_second": 2.027,
+      "step": 3000
+    },
+    {
+      "entropy": 0.12116114404052496,
+      "epoch": 7.27761013880507,
+      "grad_norm": 0.5467398166656494,
+      "learning_rate": 7.890406302404618e-05,
+      "loss": 0.0933056652545929,
+      "mean_token_accuracy": 0.9676455594599247,
+      "num_tokens": 6992784.0,
+      "step": 3020
+    },
+    {
+      "epoch": 7.27761013880507,
+      "eval_entropy": 0.27238342793804876,
+      "eval_loss": 0.9166117310523987,
+      "eval_mean_token_accuracy": 0.825499015242866,
+      "eval_num_tokens": 6992784.0,
+      "eval_runtime": 88.3276,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 3020
+    },
+    {
+      "entropy": 0.11715119928121567,
+      "epoch": 7.325890162945082,
+      "grad_norm": 0.4976702332496643,
+      "learning_rate": 7.634274596487176e-05,
+      "loss": 0.09449659585952759,
+      "mean_token_accuracy": 0.966506478190422,
+      "num_tokens": 7040482.0,
+      "step": 3040
+    },
+    {
+      "epoch": 7.325890162945082,
+      "eval_entropy": 0.2788975556077582,
+      "eval_loss": 0.8833065629005432,
+      "eval_mean_token_accuracy": 0.8269920596915684,
+      "eval_num_tokens": 7040482.0,
+      "eval_runtime": 88.9016,
+      "eval_samples_per_second": 15.973,
+      "eval_steps_per_second": 2.002,
+      "step": 3040
+    },
+    {
+      "entropy": 0.12565911412239075,
+      "epoch": 7.3741701870850935,
+      "grad_norm": 0.5728679895401001,
+      "learning_rate": 7.381306715921025e-05,
+      "loss": 0.09626244902610778,
+      "mean_token_accuracy": 0.9661444149911403,
+      "num_tokens": 7086261.0,
+      "step": 3060
+    },
+    {
+      "epoch": 7.3741701870850935,
+      "eval_entropy": 0.2762492011939542,
+      "eval_loss": 0.8961160778999329,
+      "eval_mean_token_accuracy": 0.826369910092836,
+      "eval_num_tokens": 7086261.0,
+      "eval_runtime": 88.5204,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 3060
+    },
+    {
+      "entropy": 0.12258567921817302,
+      "epoch": 7.422450211225105,
+      "grad_norm": 0.468985915184021,
+      "learning_rate": 7.131574247624031e-05,
+      "loss": 0.09120266437530518,
+      "mean_token_accuracy": 0.9674938447773457,
+      "num_tokens": 7133247.0,
+      "step": 3080
+    },
+    {
+      "epoch": 7.422450211225105,
+      "eval_entropy": 0.2688512086366,
+      "eval_loss": 0.9048654437065125,
+      "eval_mean_token_accuracy": 0.8265582302982888,
+      "eval_num_tokens": 7133247.0,
+      "eval_runtime": 88.164,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 3080
+    },
+    {
+      "entropy": 0.12070536343380808,
+      "epoch": 7.470730235365117,
+      "grad_norm": 0.5310841798782349,
+      "learning_rate": 6.885147862930688e-05,
+      "loss": 0.09386247992515565,
+      "mean_token_accuracy": 0.96664222702384,
+      "num_tokens": 7177810.0,
+      "step": 3100
+    },
+    {
+      "epoch": 7.470730235365117,
+      "eval_entropy": 0.26289632874593305,
+      "eval_loss": 0.902837336063385,
+      "eval_mean_token_accuracy": 0.828052184554968,
+      "eval_num_tokens": 7177810.0,
+      "eval_runtime": 88.1201,
+      "eval_samples_per_second": 16.114,
+      "eval_steps_per_second": 2.02,
+      "step": 3100
+    },
+    {
+      "entropy": 0.11322665181942285,
+      "epoch": 7.51901025950513,
+      "grad_norm": 0.6846067905426025,
+      "learning_rate": 6.642097297592908e-05,
+      "loss": 0.08772538304328918,
+      "mean_token_accuracy": 0.9691362135112286,
+      "num_tokens": 7225302.0,
+      "step": 3120
+    },
+    {
+      "epoch": 7.51901025950513,
+      "eval_entropy": 0.2670605613776807,
+      "eval_loss": 0.9130566120147705,
+      "eval_mean_token_accuracy": 0.826366903741708,
+      "eval_num_tokens": 7225302.0,
+      "eval_runtime": 88.6035,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3120
+    },
+    {
+      "entropy": 0.10985926580615342,
+      "epoch": 7.567290283645142,
+      "grad_norm": 0.4799748361110687,
+      "learning_rate": 6.402491332045683e-05,
+      "loss": 0.08672296404838561,
+      "mean_token_accuracy": 0.9698824048042297,
+      "num_tokens": 7274881.0,
+      "step": 3140
+    },
+    {
+      "epoch": 7.567290283645142,
+      "eval_entropy": 0.26657010866015146,
+      "eval_loss": 0.9056380987167358,
+      "eval_mean_token_accuracy": 0.8272326461384806,
+      "eval_num_tokens": 7274881.0,
+      "eval_runtime": 88.2967,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3140
+    },
+    {
+      "entropy": 0.11890473761595785,
+      "epoch": 7.615570307785154,
+      "grad_norm": 0.4081912934780121,
+      "learning_rate": 6.166397771942987e-05,
+      "loss": 0.09396827220916748,
+      "mean_token_accuracy": 0.967512384802103,
+      "num_tokens": 7321163.0,
+      "step": 3160
+    },
+    {
+      "epoch": 7.615570307785154,
+      "eval_entropy": 0.26364776575833226,
+      "eval_loss": 0.9204733967781067,
+      "eval_mean_token_accuracy": 0.827419729714983,
+      "eval_num_tokens": 7321163.0,
+      "eval_runtime": 88.3441,
+      "eval_samples_per_second": 16.074,
+      "eval_steps_per_second": 2.015,
+      "step": 3160
+    },
+    {
+      "entropy": 0.12209134344011545,
+      "epoch": 7.663850331925166,
+      "grad_norm": 0.5159774422645569,
+      "learning_rate": 5.93388342896962e-05,
+      "loss": 0.09292978644371033,
+      "mean_token_accuracy": 0.9658421002328396,
+      "num_tokens": 7366217.0,
+      "step": 3180
+    },
+    {
+      "epoch": 7.663850331925166,
+      "eval_entropy": 0.2714110068725736,
+      "eval_loss": 0.9143092036247253,
+      "eval_mean_token_accuracy": 0.8262251099843657,
+      "eval_num_tokens": 7366217.0,
+      "eval_runtime": 87.9635,
+      "eval_samples_per_second": 16.143,
+      "eval_steps_per_second": 2.024,
+      "step": 3180
+    },
+    {
+      "entropy": 0.11955609126016498,
+      "epoch": 7.712130356065178,
+      "grad_norm": 0.7005976438522339,
+      "learning_rate": 5.705014101934252e-05,
+      "loss": 0.09319776892662049,
+      "mean_token_accuracy": 0.9674513377249241,
+      "num_tokens": 7410887.0,
+      "step": 3200
+    },
+    {
+      "epoch": 7.712130356065178,
+      "eval_entropy": 0.2626291513442993,
+      "eval_loss": 0.91884446144104,
+      "eval_mean_token_accuracy": 0.827563955207889,
+      "eval_num_tokens": 7410887.0,
+      "eval_runtime": 87.8849,
+      "eval_samples_per_second": 16.157,
+      "eval_steps_per_second": 2.025,
+      "step": 3200
+    },
+    {
+      "entropy": 0.11767532899975777,
+      "epoch": 7.76041038020519,
+      "grad_norm": 0.4675362706184387,
+      "learning_rate": 5.479854558149187e-05,
+      "loss": 0.09038087725639343,
+      "mean_token_accuracy": 0.9669680051505566,
+      "num_tokens": 7457578.0,
+      "step": 3220
+    },
+    {
+      "epoch": 7.76041038020519,
+      "eval_entropy": 0.26581037203582486,
+      "eval_loss": 0.9146883487701416,
+      "eval_mean_token_accuracy": 0.8270345812433222,
+      "eval_num_tokens": 7457578.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 3220
+    },
+    {
+      "entropy": 0.11383061073720455,
+      "epoch": 7.808690404345202,
+      "grad_norm": 0.541247546672821,
+      "learning_rate": 5.258468515101918e-05,
+      "loss": 0.08942405581474304,
+      "mean_token_accuracy": 0.968398705124855,
+      "num_tokens": 7505761.0,
+      "step": 3240
+    },
+    {
+      "epoch": 7.808690404345202,
+      "eval_entropy": 0.26687998960861997,
+      "eval_loss": 0.9037085175514221,
+      "eval_mean_token_accuracy": 0.8274977863504646,
+      "eval_num_tokens": 7505761.0,
+      "eval_runtime": 88.5163,
+      "eval_samples_per_second": 16.042,
+      "eval_steps_per_second": 2.011,
+      "step": 3240
+    },
+    {
+      "entropy": 0.12004058873280883,
+      "epoch": 7.856970428485214,
+      "grad_norm": 0.5289552211761475,
+      "learning_rate": 5.040918622423899e-05,
+      "loss": 0.09064805507659912,
+      "mean_token_accuracy": 0.9673330299556255,
+      "num_tokens": 7552557.0,
+      "step": 3260
+    },
+    {
+      "epoch": 7.856970428485214,
+      "eval_entropy": 0.26288379251622085,
+      "eval_loss": 0.9146178364753723,
+      "eval_mean_token_accuracy": 0.8276662448149049,
+      "eval_num_tokens": 7552557.0,
+      "eval_runtime": 88.0723,
+      "eval_samples_per_second": 16.123,
+      "eval_steps_per_second": 2.021,
+      "step": 3260
+    },
+    {
+      "entropy": 0.1180689456872642,
+      "epoch": 7.905250452625227,
+      "grad_norm": 0.4754861891269684,
+      "learning_rate": 4.827266444161377e-05,
+      "loss": 0.09402655363082886,
+      "mean_token_accuracy": 0.9663361787796021,
+      "num_tokens": 7596998.0,
+      "step": 3280
+    },
+    {
+      "epoch": 7.905250452625227,
+      "eval_entropy": 0.261690996922134,
+      "eval_loss": 0.9120186567306519,
+      "eval_mean_token_accuracy": 0.8278995015647974,
+      "eval_num_tokens": 7596998.0,
+      "eval_runtime": 88.6573,
+      "eval_samples_per_second": 16.017,
+      "eval_steps_per_second": 2.008,
+      "step": 3280
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.212524801463931e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3300/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3300/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3300/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3300/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3300/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3300/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3300/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3300/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3300/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3300/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3300/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3300/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3300/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3300/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..495293dc25eda8fe690801b8cc9f4600865c1517
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3300/trainer_state.json
@@ -0,0 +1,3499 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 7.953530476765239,
+  "eval_steps": 20,
+  "global_step": 3300,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    },
+    {
+      "entropy": 0.1326121488586068,
+      "epoch": 7.181050090525045,
+      "grad_norm": 0.4843117296695709,
+      "learning_rate": 8.411870386507724e-05,
+      "loss": 0.09537842869758606,
+      "mean_token_accuracy": 0.9645151920616627,
+      "num_tokens": 6898444.0,
+      "step": 2980
+    },
+    {
+      "epoch": 7.181050090525045,
+      "eval_entropy": 0.27999477783280813,
+      "eval_loss": 0.9031037092208862,
+      "eval_mean_token_accuracy": 0.8261882204018282,
+      "eval_num_tokens": 6898444.0,
+      "eval_runtime": 88.5602,
+      "eval_samples_per_second": 16.034,
+      "eval_steps_per_second": 2.01,
+      "step": 2980
+    },
+    {
+      "entropy": 0.1156751093454659,
+      "epoch": 7.229330114665057,
+      "grad_norm": 0.5297922492027283,
+      "learning_rate": 8.149629351430326e-05,
+      "loss": 0.09203402996063233,
+      "mean_token_accuracy": 0.9689733698964119,
+      "num_tokens": 6947106.0,
+      "step": 3000
+    },
+    {
+      "epoch": 7.229330114665057,
+      "eval_entropy": 0.2740225845508361,
+      "eval_loss": 0.8876440525054932,
+      "eval_mean_token_accuracy": 0.8265116857678703,
+      "eval_num_tokens": 6947106.0,
+      "eval_runtime": 87.7971,
+      "eval_samples_per_second": 16.174,
+      "eval_steps_per_second": 2.027,
+      "step": 3000
+    },
+    {
+      "entropy": 0.12116114404052496,
+      "epoch": 7.27761013880507,
+      "grad_norm": 0.5467398166656494,
+      "learning_rate": 7.890406302404618e-05,
+      "loss": 0.0933056652545929,
+      "mean_token_accuracy": 0.9676455594599247,
+      "num_tokens": 6992784.0,
+      "step": 3020
+    },
+    {
+      "epoch": 7.27761013880507,
+      "eval_entropy": 0.27238342793804876,
+      "eval_loss": 0.9166117310523987,
+      "eval_mean_token_accuracy": 0.825499015242866,
+      "eval_num_tokens": 6992784.0,
+      "eval_runtime": 88.3276,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 3020
+    },
+    {
+      "entropy": 0.11715119928121567,
+      "epoch": 7.325890162945082,
+      "grad_norm": 0.4976702332496643,
+      "learning_rate": 7.634274596487176e-05,
+      "loss": 0.09449659585952759,
+      "mean_token_accuracy": 0.966506478190422,
+      "num_tokens": 7040482.0,
+      "step": 3040
+    },
+    {
+      "epoch": 7.325890162945082,
+      "eval_entropy": 0.2788975556077582,
+      "eval_loss": 0.8833065629005432,
+      "eval_mean_token_accuracy": 0.8269920596915684,
+      "eval_num_tokens": 7040482.0,
+      "eval_runtime": 88.9016,
+      "eval_samples_per_second": 15.973,
+      "eval_steps_per_second": 2.002,
+      "step": 3040
+    },
+    {
+      "entropy": 0.12565911412239075,
+      "epoch": 7.3741701870850935,
+      "grad_norm": 0.5728679895401001,
+      "learning_rate": 7.381306715921025e-05,
+      "loss": 0.09626244902610778,
+      "mean_token_accuracy": 0.9661444149911403,
+      "num_tokens": 7086261.0,
+      "step": 3060
+    },
+    {
+      "epoch": 7.3741701870850935,
+      "eval_entropy": 0.2762492011939542,
+      "eval_loss": 0.8961160778999329,
+      "eval_mean_token_accuracy": 0.826369910092836,
+      "eval_num_tokens": 7086261.0,
+      "eval_runtime": 88.5204,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 3060
+    },
+    {
+      "entropy": 0.12258567921817302,
+      "epoch": 7.422450211225105,
+      "grad_norm": 0.468985915184021,
+      "learning_rate": 7.131574247624031e-05,
+      "loss": 0.09120266437530518,
+      "mean_token_accuracy": 0.9674938447773457,
+      "num_tokens": 7133247.0,
+      "step": 3080
+    },
+    {
+      "epoch": 7.422450211225105,
+      "eval_entropy": 0.2688512086366,
+      "eval_loss": 0.9048654437065125,
+      "eval_mean_token_accuracy": 0.8265582302982888,
+      "eval_num_tokens": 7133247.0,
+      "eval_runtime": 88.164,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 3080
+    },
+    {
+      "entropy": 0.12070536343380808,
+      "epoch": 7.470730235365117,
+      "grad_norm": 0.5310841798782349,
+      "learning_rate": 6.885147862930688e-05,
+      "loss": 0.09386247992515565,
+      "mean_token_accuracy": 0.96664222702384,
+      "num_tokens": 7177810.0,
+      "step": 3100
+    },
+    {
+      "epoch": 7.470730235365117,
+      "eval_entropy": 0.26289632874593305,
+      "eval_loss": 0.902837336063385,
+      "eval_mean_token_accuracy": 0.828052184554968,
+      "eval_num_tokens": 7177810.0,
+      "eval_runtime": 88.1201,
+      "eval_samples_per_second": 16.114,
+      "eval_steps_per_second": 2.02,
+      "step": 3100
+    },
+    {
+      "entropy": 0.11322665181942285,
+      "epoch": 7.51901025950513,
+      "grad_norm": 0.6846067905426025,
+      "learning_rate": 6.642097297592908e-05,
+      "loss": 0.08772538304328918,
+      "mean_token_accuracy": 0.9691362135112286,
+      "num_tokens": 7225302.0,
+      "step": 3120
+    },
+    {
+      "epoch": 7.51901025950513,
+      "eval_entropy": 0.2670605613776807,
+      "eval_loss": 0.9130566120147705,
+      "eval_mean_token_accuracy": 0.826366903741708,
+      "eval_num_tokens": 7225302.0,
+      "eval_runtime": 88.6035,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3120
+    },
+    {
+      "entropy": 0.10985926580615342,
+      "epoch": 7.567290283645142,
+      "grad_norm": 0.4799748361110687,
+      "learning_rate": 6.402491332045683e-05,
+      "loss": 0.08672296404838561,
+      "mean_token_accuracy": 0.9698824048042297,
+      "num_tokens": 7274881.0,
+      "step": 3140
+    },
+    {
+      "epoch": 7.567290283645142,
+      "eval_entropy": 0.26657010866015146,
+      "eval_loss": 0.9056380987167358,
+      "eval_mean_token_accuracy": 0.8272326461384806,
+      "eval_num_tokens": 7274881.0,
+      "eval_runtime": 88.2967,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3140
+    },
+    {
+      "entropy": 0.11890473761595785,
+      "epoch": 7.615570307785154,
+      "grad_norm": 0.4081912934780121,
+      "learning_rate": 6.166397771942987e-05,
+      "loss": 0.09396827220916748,
+      "mean_token_accuracy": 0.967512384802103,
+      "num_tokens": 7321163.0,
+      "step": 3160
+    },
+    {
+      "epoch": 7.615570307785154,
+      "eval_entropy": 0.26364776575833226,
+      "eval_loss": 0.9204733967781067,
+      "eval_mean_token_accuracy": 0.827419729714983,
+      "eval_num_tokens": 7321163.0,
+      "eval_runtime": 88.3441,
+      "eval_samples_per_second": 16.074,
+      "eval_steps_per_second": 2.015,
+      "step": 3160
+    },
+    {
+      "entropy": 0.12209134344011545,
+      "epoch": 7.663850331925166,
+      "grad_norm": 0.5159774422645569,
+      "learning_rate": 5.93388342896962e-05,
+      "loss": 0.09292978644371033,
+      "mean_token_accuracy": 0.9658421002328396,
+      "num_tokens": 7366217.0,
+      "step": 3180
+    },
+    {
+      "epoch": 7.663850331925166,
+      "eval_entropy": 0.2714110068725736,
+      "eval_loss": 0.9143092036247253,
+      "eval_mean_token_accuracy": 0.8262251099843657,
+      "eval_num_tokens": 7366217.0,
+      "eval_runtime": 87.9635,
+      "eval_samples_per_second": 16.143,
+      "eval_steps_per_second": 2.024,
+      "step": 3180
+    },
+    {
+      "entropy": 0.11955609126016498,
+      "epoch": 7.712130356065178,
+      "grad_norm": 0.7005976438522339,
+      "learning_rate": 5.705014101934252e-05,
+      "loss": 0.09319776892662049,
+      "mean_token_accuracy": 0.9674513377249241,
+      "num_tokens": 7410887.0,
+      "step": 3200
+    },
+    {
+      "epoch": 7.712130356065178,
+      "eval_entropy": 0.2626291513442993,
+      "eval_loss": 0.91884446144104,
+      "eval_mean_token_accuracy": 0.827563955207889,
+      "eval_num_tokens": 7410887.0,
+      "eval_runtime": 87.8849,
+      "eval_samples_per_second": 16.157,
+      "eval_steps_per_second": 2.025,
+      "step": 3200
+    },
+    {
+      "entropy": 0.11767532899975777,
+      "epoch": 7.76041038020519,
+      "grad_norm": 0.4675362706184387,
+      "learning_rate": 5.479854558149187e-05,
+      "loss": 0.09038087725639343,
+      "mean_token_accuracy": 0.9669680051505566,
+      "num_tokens": 7457578.0,
+      "step": 3220
+    },
+    {
+      "epoch": 7.76041038020519,
+      "eval_entropy": 0.26581037203582486,
+      "eval_loss": 0.9146883487701416,
+      "eval_mean_token_accuracy": 0.8270345812433222,
+      "eval_num_tokens": 7457578.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 3220
+    },
+    {
+      "entropy": 0.11383061073720455,
+      "epoch": 7.808690404345202,
+      "grad_norm": 0.541247546672821,
+      "learning_rate": 5.258468515101918e-05,
+      "loss": 0.08942405581474304,
+      "mean_token_accuracy": 0.968398705124855,
+      "num_tokens": 7505761.0,
+      "step": 3240
+    },
+    {
+      "epoch": 7.808690404345202,
+      "eval_entropy": 0.26687998960861997,
+      "eval_loss": 0.9037085175514221,
+      "eval_mean_token_accuracy": 0.8274977863504646,
+      "eval_num_tokens": 7505761.0,
+      "eval_runtime": 88.5163,
+      "eval_samples_per_second": 16.042,
+      "eval_steps_per_second": 2.011,
+      "step": 3240
+    },
+    {
+      "entropy": 0.12004058873280883,
+      "epoch": 7.856970428485214,
+      "grad_norm": 0.5289552211761475,
+      "learning_rate": 5.040918622423899e-05,
+      "loss": 0.09064805507659912,
+      "mean_token_accuracy": 0.9673330299556255,
+      "num_tokens": 7552557.0,
+      "step": 3260
+    },
+    {
+      "epoch": 7.856970428485214,
+      "eval_entropy": 0.26288379251622085,
+      "eval_loss": 0.9146178364753723,
+      "eval_mean_token_accuracy": 0.8276662448149049,
+      "eval_num_tokens": 7552557.0,
+      "eval_runtime": 88.0723,
+      "eval_samples_per_second": 16.123,
+      "eval_steps_per_second": 2.021,
+      "step": 3260
+    },
+    {
+      "entropy": 0.1180689456872642,
+      "epoch": 7.905250452625227,
+      "grad_norm": 0.4754861891269684,
+      "learning_rate": 4.827266444161377e-05,
+      "loss": 0.09402655363082886,
+      "mean_token_accuracy": 0.9663361787796021,
+      "num_tokens": 7596998.0,
+      "step": 3280
+    },
+    {
+      "epoch": 7.905250452625227,
+      "eval_entropy": 0.261690996922134,
+      "eval_loss": 0.9120186567306519,
+      "eval_mean_token_accuracy": 0.8278995015647974,
+      "eval_num_tokens": 7596998.0,
+      "eval_runtime": 88.6573,
+      "eval_samples_per_second": 16.017,
+      "eval_steps_per_second": 2.008,
+      "step": 3280
+    },
+    {
+      "entropy": 0.1162164019420743,
+      "epoch": 7.953530476765239,
+      "grad_norm": 0.6203398108482361,
+      "learning_rate": 4.61757244135355e-05,
+      "loss": 0.08832768201828003,
+      "mean_token_accuracy": 0.9694220684468746,
+      "num_tokens": 7643288.0,
+      "step": 3300
+    },
+    {
+      "epoch": 7.953530476765239,
+      "eval_entropy": 0.26158678129817664,
+      "eval_loss": 0.9301361441612244,
+      "eval_mean_token_accuracy": 0.8269603885291668,
+      "eval_num_tokens": 7643288.0,
+      "eval_runtime": 88.7064,
+      "eval_samples_per_second": 16.008,
+      "eval_steps_per_second": 2.007,
+      "step": 3300
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.2321661154712986e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3320/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3320/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3320/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3320/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3320/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3320/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3320/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3320/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3320/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3320/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3320/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3320/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3320/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3320/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..4c53da8266592955dba5e41ee9086fc0d2dc93c6
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3320/trainer_state.json
@@ -0,0 +1,3520 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 8.0,
+  "eval_steps": 20,
+  "global_step": 3320,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    },
+    {
+      "entropy": 0.1326121488586068,
+      "epoch": 7.181050090525045,
+      "grad_norm": 0.4843117296695709,
+      "learning_rate": 8.411870386507724e-05,
+      "loss": 0.09537842869758606,
+      "mean_token_accuracy": 0.9645151920616627,
+      "num_tokens": 6898444.0,
+      "step": 2980
+    },
+    {
+      "epoch": 7.181050090525045,
+      "eval_entropy": 0.27999477783280813,
+      "eval_loss": 0.9031037092208862,
+      "eval_mean_token_accuracy": 0.8261882204018282,
+      "eval_num_tokens": 6898444.0,
+      "eval_runtime": 88.5602,
+      "eval_samples_per_second": 16.034,
+      "eval_steps_per_second": 2.01,
+      "step": 2980
+    },
+    {
+      "entropy": 0.1156751093454659,
+      "epoch": 7.229330114665057,
+      "grad_norm": 0.5297922492027283,
+      "learning_rate": 8.149629351430326e-05,
+      "loss": 0.09203402996063233,
+      "mean_token_accuracy": 0.9689733698964119,
+      "num_tokens": 6947106.0,
+      "step": 3000
+    },
+    {
+      "epoch": 7.229330114665057,
+      "eval_entropy": 0.2740225845508361,
+      "eval_loss": 0.8876440525054932,
+      "eval_mean_token_accuracy": 0.8265116857678703,
+      "eval_num_tokens": 6947106.0,
+      "eval_runtime": 87.7971,
+      "eval_samples_per_second": 16.174,
+      "eval_steps_per_second": 2.027,
+      "step": 3000
+    },
+    {
+      "entropy": 0.12116114404052496,
+      "epoch": 7.27761013880507,
+      "grad_norm": 0.5467398166656494,
+      "learning_rate": 7.890406302404618e-05,
+      "loss": 0.0933056652545929,
+      "mean_token_accuracy": 0.9676455594599247,
+      "num_tokens": 6992784.0,
+      "step": 3020
+    },
+    {
+      "epoch": 7.27761013880507,
+      "eval_entropy": 0.27238342793804876,
+      "eval_loss": 0.9166117310523987,
+      "eval_mean_token_accuracy": 0.825499015242866,
+      "eval_num_tokens": 6992784.0,
+      "eval_runtime": 88.3276,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 3020
+    },
+    {
+      "entropy": 0.11715119928121567,
+      "epoch": 7.325890162945082,
+      "grad_norm": 0.4976702332496643,
+      "learning_rate": 7.634274596487176e-05,
+      "loss": 0.09449659585952759,
+      "mean_token_accuracy": 0.966506478190422,
+      "num_tokens": 7040482.0,
+      "step": 3040
+    },
+    {
+      "epoch": 7.325890162945082,
+      "eval_entropy": 0.2788975556077582,
+      "eval_loss": 0.8833065629005432,
+      "eval_mean_token_accuracy": 0.8269920596915684,
+      "eval_num_tokens": 7040482.0,
+      "eval_runtime": 88.9016,
+      "eval_samples_per_second": 15.973,
+      "eval_steps_per_second": 2.002,
+      "step": 3040
+    },
+    {
+      "entropy": 0.12565911412239075,
+      "epoch": 7.3741701870850935,
+      "grad_norm": 0.5728679895401001,
+      "learning_rate": 7.381306715921025e-05,
+      "loss": 0.09626244902610778,
+      "mean_token_accuracy": 0.9661444149911403,
+      "num_tokens": 7086261.0,
+      "step": 3060
+    },
+    {
+      "epoch": 7.3741701870850935,
+      "eval_entropy": 0.2762492011939542,
+      "eval_loss": 0.8961160778999329,
+      "eval_mean_token_accuracy": 0.826369910092836,
+      "eval_num_tokens": 7086261.0,
+      "eval_runtime": 88.5204,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 3060
+    },
+    {
+      "entropy": 0.12258567921817302,
+      "epoch": 7.422450211225105,
+      "grad_norm": 0.468985915184021,
+      "learning_rate": 7.131574247624031e-05,
+      "loss": 0.09120266437530518,
+      "mean_token_accuracy": 0.9674938447773457,
+      "num_tokens": 7133247.0,
+      "step": 3080
+    },
+    {
+      "epoch": 7.422450211225105,
+      "eval_entropy": 0.2688512086366,
+      "eval_loss": 0.9048654437065125,
+      "eval_mean_token_accuracy": 0.8265582302982888,
+      "eval_num_tokens": 7133247.0,
+      "eval_runtime": 88.164,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 3080
+    },
+    {
+      "entropy": 0.12070536343380808,
+      "epoch": 7.470730235365117,
+      "grad_norm": 0.5310841798782349,
+      "learning_rate": 6.885147862930688e-05,
+      "loss": 0.09386247992515565,
+      "mean_token_accuracy": 0.96664222702384,
+      "num_tokens": 7177810.0,
+      "step": 3100
+    },
+    {
+      "epoch": 7.470730235365117,
+      "eval_entropy": 0.26289632874593305,
+      "eval_loss": 0.902837336063385,
+      "eval_mean_token_accuracy": 0.828052184554968,
+      "eval_num_tokens": 7177810.0,
+      "eval_runtime": 88.1201,
+      "eval_samples_per_second": 16.114,
+      "eval_steps_per_second": 2.02,
+      "step": 3100
+    },
+    {
+      "entropy": 0.11322665181942285,
+      "epoch": 7.51901025950513,
+      "grad_norm": 0.6846067905426025,
+      "learning_rate": 6.642097297592908e-05,
+      "loss": 0.08772538304328918,
+      "mean_token_accuracy": 0.9691362135112286,
+      "num_tokens": 7225302.0,
+      "step": 3120
+    },
+    {
+      "epoch": 7.51901025950513,
+      "eval_entropy": 0.2670605613776807,
+      "eval_loss": 0.9130566120147705,
+      "eval_mean_token_accuracy": 0.826366903741708,
+      "eval_num_tokens": 7225302.0,
+      "eval_runtime": 88.6035,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3120
+    },
+    {
+      "entropy": 0.10985926580615342,
+      "epoch": 7.567290283645142,
+      "grad_norm": 0.4799748361110687,
+      "learning_rate": 6.402491332045683e-05,
+      "loss": 0.08672296404838561,
+      "mean_token_accuracy": 0.9698824048042297,
+      "num_tokens": 7274881.0,
+      "step": 3140
+    },
+    {
+      "epoch": 7.567290283645142,
+      "eval_entropy": 0.26657010866015146,
+      "eval_loss": 0.9056380987167358,
+      "eval_mean_token_accuracy": 0.8272326461384806,
+      "eval_num_tokens": 7274881.0,
+      "eval_runtime": 88.2967,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3140
+    },
+    {
+      "entropy": 0.11890473761595785,
+      "epoch": 7.615570307785154,
+      "grad_norm": 0.4081912934780121,
+      "learning_rate": 6.166397771942987e-05,
+      "loss": 0.09396827220916748,
+      "mean_token_accuracy": 0.967512384802103,
+      "num_tokens": 7321163.0,
+      "step": 3160
+    },
+    {
+      "epoch": 7.615570307785154,
+      "eval_entropy": 0.26364776575833226,
+      "eval_loss": 0.9204733967781067,
+      "eval_mean_token_accuracy": 0.827419729714983,
+      "eval_num_tokens": 7321163.0,
+      "eval_runtime": 88.3441,
+      "eval_samples_per_second": 16.074,
+      "eval_steps_per_second": 2.015,
+      "step": 3160
+    },
+    {
+      "entropy": 0.12209134344011545,
+      "epoch": 7.663850331925166,
+      "grad_norm": 0.5159774422645569,
+      "learning_rate": 5.93388342896962e-05,
+      "loss": 0.09292978644371033,
+      "mean_token_accuracy": 0.9658421002328396,
+      "num_tokens": 7366217.0,
+      "step": 3180
+    },
+    {
+      "epoch": 7.663850331925166,
+      "eval_entropy": 0.2714110068725736,
+      "eval_loss": 0.9143092036247253,
+      "eval_mean_token_accuracy": 0.8262251099843657,
+      "eval_num_tokens": 7366217.0,
+      "eval_runtime": 87.9635,
+      "eval_samples_per_second": 16.143,
+      "eval_steps_per_second": 2.024,
+      "step": 3180
+    },
+    {
+      "entropy": 0.11955609126016498,
+      "epoch": 7.712130356065178,
+      "grad_norm": 0.7005976438522339,
+      "learning_rate": 5.705014101934252e-05,
+      "loss": 0.09319776892662049,
+      "mean_token_accuracy": 0.9674513377249241,
+      "num_tokens": 7410887.0,
+      "step": 3200
+    },
+    {
+      "epoch": 7.712130356065178,
+      "eval_entropy": 0.2626291513442993,
+      "eval_loss": 0.91884446144104,
+      "eval_mean_token_accuracy": 0.827563955207889,
+      "eval_num_tokens": 7410887.0,
+      "eval_runtime": 87.8849,
+      "eval_samples_per_second": 16.157,
+      "eval_steps_per_second": 2.025,
+      "step": 3200
+    },
+    {
+      "entropy": 0.11767532899975777,
+      "epoch": 7.76041038020519,
+      "grad_norm": 0.4675362706184387,
+      "learning_rate": 5.479854558149187e-05,
+      "loss": 0.09038087725639343,
+      "mean_token_accuracy": 0.9669680051505566,
+      "num_tokens": 7457578.0,
+      "step": 3220
+    },
+    {
+      "epoch": 7.76041038020519,
+      "eval_entropy": 0.26581037203582486,
+      "eval_loss": 0.9146883487701416,
+      "eval_mean_token_accuracy": 0.8270345812433222,
+      "eval_num_tokens": 7457578.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 3220
+    },
+    {
+      "entropy": 0.11383061073720455,
+      "epoch": 7.808690404345202,
+      "grad_norm": 0.541247546672821,
+      "learning_rate": 5.258468515101918e-05,
+      "loss": 0.08942405581474304,
+      "mean_token_accuracy": 0.968398705124855,
+      "num_tokens": 7505761.0,
+      "step": 3240
+    },
+    {
+      "epoch": 7.808690404345202,
+      "eval_entropy": 0.26687998960861997,
+      "eval_loss": 0.9037085175514221,
+      "eval_mean_token_accuracy": 0.8274977863504646,
+      "eval_num_tokens": 7505761.0,
+      "eval_runtime": 88.5163,
+      "eval_samples_per_second": 16.042,
+      "eval_steps_per_second": 2.011,
+      "step": 3240
+    },
+    {
+      "entropy": 0.12004058873280883,
+      "epoch": 7.856970428485214,
+      "grad_norm": 0.5289552211761475,
+      "learning_rate": 5.040918622423899e-05,
+      "loss": 0.09064805507659912,
+      "mean_token_accuracy": 0.9673330299556255,
+      "num_tokens": 7552557.0,
+      "step": 3260
+    },
+    {
+      "epoch": 7.856970428485214,
+      "eval_entropy": 0.26288379251622085,
+      "eval_loss": 0.9146178364753723,
+      "eval_mean_token_accuracy": 0.8276662448149049,
+      "eval_num_tokens": 7552557.0,
+      "eval_runtime": 88.0723,
+      "eval_samples_per_second": 16.123,
+      "eval_steps_per_second": 2.021,
+      "step": 3260
+    },
+    {
+      "entropy": 0.1180689456872642,
+      "epoch": 7.905250452625227,
+      "grad_norm": 0.4754861891269684,
+      "learning_rate": 4.827266444161377e-05,
+      "loss": 0.09402655363082886,
+      "mean_token_accuracy": 0.9663361787796021,
+      "num_tokens": 7596998.0,
+      "step": 3280
+    },
+    {
+      "epoch": 7.905250452625227,
+      "eval_entropy": 0.261690996922134,
+      "eval_loss": 0.9120186567306519,
+      "eval_mean_token_accuracy": 0.8278995015647974,
+      "eval_num_tokens": 7596998.0,
+      "eval_runtime": 88.6573,
+      "eval_samples_per_second": 16.017,
+      "eval_steps_per_second": 2.008,
+      "step": 3280
+    },
+    {
+      "entropy": 0.1162164019420743,
+      "epoch": 7.953530476765239,
+      "grad_norm": 0.6203398108482361,
+      "learning_rate": 4.61757244135355e-05,
+      "loss": 0.08832768201828003,
+      "mean_token_accuracy": 0.9694220684468746,
+      "num_tokens": 7643288.0,
+      "step": 3300
+    },
+    {
+      "epoch": 7.953530476765239,
+      "eval_entropy": 0.26158678129817664,
+      "eval_loss": 0.9301361441612244,
+      "eval_mean_token_accuracy": 0.8269603885291668,
+      "eval_num_tokens": 7643288.0,
+      "eval_runtime": 88.7064,
+      "eval_samples_per_second": 16.008,
+      "eval_steps_per_second": 2.007,
+      "step": 3300
+    },
+    {
+      "entropy": 0.10877202204489089,
+      "epoch": 8.0,
+      "grad_norm": 3.058673143386841,
+      "learning_rate": 4.411895954922763e-05,
+      "loss": 0.09159615635871887,
+      "mean_token_accuracy": 0.9686878834451947,
+      "num_tokens": 7688608.0,
+      "step": 3320
+    },
+    {
+      "epoch": 8.0,
+      "eval_entropy": 0.2590438057197614,
+      "eval_loss": 0.9262065887451172,
+      "eval_mean_token_accuracy": 0.8277404546067956,
+      "eval_num_tokens": 7688608.0,
+      "eval_runtime": 87.7557,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 3320
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.250573432851456e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3340/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3340/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3340/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3340/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3340/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3340/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3340/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3340/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3340/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3340/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3340/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3340/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3340/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3340/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..fa6ab53c1fe48c20eba37b9d5db217a69501c1b9
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3340/trainer_state.json
@@ -0,0 +1,3541 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 8.048280024140013,
+  "eval_steps": 20,
+  "global_step": 3340,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    },
+    {
+      "entropy": 0.1326121488586068,
+      "epoch": 7.181050090525045,
+      "grad_norm": 0.4843117296695709,
+      "learning_rate": 8.411870386507724e-05,
+      "loss": 0.09537842869758606,
+      "mean_token_accuracy": 0.9645151920616627,
+      "num_tokens": 6898444.0,
+      "step": 2980
+    },
+    {
+      "epoch": 7.181050090525045,
+      "eval_entropy": 0.27999477783280813,
+      "eval_loss": 0.9031037092208862,
+      "eval_mean_token_accuracy": 0.8261882204018282,
+      "eval_num_tokens": 6898444.0,
+      "eval_runtime": 88.5602,
+      "eval_samples_per_second": 16.034,
+      "eval_steps_per_second": 2.01,
+      "step": 2980
+    },
+    {
+      "entropy": 0.1156751093454659,
+      "epoch": 7.229330114665057,
+      "grad_norm": 0.5297922492027283,
+      "learning_rate": 8.149629351430326e-05,
+      "loss": 0.09203402996063233,
+      "mean_token_accuracy": 0.9689733698964119,
+      "num_tokens": 6947106.0,
+      "step": 3000
+    },
+    {
+      "epoch": 7.229330114665057,
+      "eval_entropy": 0.2740225845508361,
+      "eval_loss": 0.8876440525054932,
+      "eval_mean_token_accuracy": 0.8265116857678703,
+      "eval_num_tokens": 6947106.0,
+      "eval_runtime": 87.7971,
+      "eval_samples_per_second": 16.174,
+      "eval_steps_per_second": 2.027,
+      "step": 3000
+    },
+    {
+      "entropy": 0.12116114404052496,
+      "epoch": 7.27761013880507,
+      "grad_norm": 0.5467398166656494,
+      "learning_rate": 7.890406302404618e-05,
+      "loss": 0.0933056652545929,
+      "mean_token_accuracy": 0.9676455594599247,
+      "num_tokens": 6992784.0,
+      "step": 3020
+    },
+    {
+      "epoch": 7.27761013880507,
+      "eval_entropy": 0.27238342793804876,
+      "eval_loss": 0.9166117310523987,
+      "eval_mean_token_accuracy": 0.825499015242866,
+      "eval_num_tokens": 6992784.0,
+      "eval_runtime": 88.3276,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 3020
+    },
+    {
+      "entropy": 0.11715119928121567,
+      "epoch": 7.325890162945082,
+      "grad_norm": 0.4976702332496643,
+      "learning_rate": 7.634274596487176e-05,
+      "loss": 0.09449659585952759,
+      "mean_token_accuracy": 0.966506478190422,
+      "num_tokens": 7040482.0,
+      "step": 3040
+    },
+    {
+      "epoch": 7.325890162945082,
+      "eval_entropy": 0.2788975556077582,
+      "eval_loss": 0.8833065629005432,
+      "eval_mean_token_accuracy": 0.8269920596915684,
+      "eval_num_tokens": 7040482.0,
+      "eval_runtime": 88.9016,
+      "eval_samples_per_second": 15.973,
+      "eval_steps_per_second": 2.002,
+      "step": 3040
+    },
+    {
+      "entropy": 0.12565911412239075,
+      "epoch": 7.3741701870850935,
+      "grad_norm": 0.5728679895401001,
+      "learning_rate": 7.381306715921025e-05,
+      "loss": 0.09626244902610778,
+      "mean_token_accuracy": 0.9661444149911403,
+      "num_tokens": 7086261.0,
+      "step": 3060
+    },
+    {
+      "epoch": 7.3741701870850935,
+      "eval_entropy": 0.2762492011939542,
+      "eval_loss": 0.8961160778999329,
+      "eval_mean_token_accuracy": 0.826369910092836,
+      "eval_num_tokens": 7086261.0,
+      "eval_runtime": 88.5204,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 3060
+    },
+    {
+      "entropy": 0.12258567921817302,
+      "epoch": 7.422450211225105,
+      "grad_norm": 0.468985915184021,
+      "learning_rate": 7.131574247624031e-05,
+      "loss": 0.09120266437530518,
+      "mean_token_accuracy": 0.9674938447773457,
+      "num_tokens": 7133247.0,
+      "step": 3080
+    },
+    {
+      "epoch": 7.422450211225105,
+      "eval_entropy": 0.2688512086366,
+      "eval_loss": 0.9048654437065125,
+      "eval_mean_token_accuracy": 0.8265582302982888,
+      "eval_num_tokens": 7133247.0,
+      "eval_runtime": 88.164,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 3080
+    },
+    {
+      "entropy": 0.12070536343380808,
+      "epoch": 7.470730235365117,
+      "grad_norm": 0.5310841798782349,
+      "learning_rate": 6.885147862930688e-05,
+      "loss": 0.09386247992515565,
+      "mean_token_accuracy": 0.96664222702384,
+      "num_tokens": 7177810.0,
+      "step": 3100
+    },
+    {
+      "epoch": 7.470730235365117,
+      "eval_entropy": 0.26289632874593305,
+      "eval_loss": 0.902837336063385,
+      "eval_mean_token_accuracy": 0.828052184554968,
+      "eval_num_tokens": 7177810.0,
+      "eval_runtime": 88.1201,
+      "eval_samples_per_second": 16.114,
+      "eval_steps_per_second": 2.02,
+      "step": 3100
+    },
+    {
+      "entropy": 0.11322665181942285,
+      "epoch": 7.51901025950513,
+      "grad_norm": 0.6846067905426025,
+      "learning_rate": 6.642097297592908e-05,
+      "loss": 0.08772538304328918,
+      "mean_token_accuracy": 0.9691362135112286,
+      "num_tokens": 7225302.0,
+      "step": 3120
+    },
+    {
+      "epoch": 7.51901025950513,
+      "eval_entropy": 0.2670605613776807,
+      "eval_loss": 0.9130566120147705,
+      "eval_mean_token_accuracy": 0.826366903741708,
+      "eval_num_tokens": 7225302.0,
+      "eval_runtime": 88.6035,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3120
+    },
+    {
+      "entropy": 0.10985926580615342,
+      "epoch": 7.567290283645142,
+      "grad_norm": 0.4799748361110687,
+      "learning_rate": 6.402491332045683e-05,
+      "loss": 0.08672296404838561,
+      "mean_token_accuracy": 0.9698824048042297,
+      "num_tokens": 7274881.0,
+      "step": 3140
+    },
+    {
+      "epoch": 7.567290283645142,
+      "eval_entropy": 0.26657010866015146,
+      "eval_loss": 0.9056380987167358,
+      "eval_mean_token_accuracy": 0.8272326461384806,
+      "eval_num_tokens": 7274881.0,
+      "eval_runtime": 88.2967,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3140
+    },
+    {
+      "entropy": 0.11890473761595785,
+      "epoch": 7.615570307785154,
+      "grad_norm": 0.4081912934780121,
+      "learning_rate": 6.166397771942987e-05,
+      "loss": 0.09396827220916748,
+      "mean_token_accuracy": 0.967512384802103,
+      "num_tokens": 7321163.0,
+      "step": 3160
+    },
+    {
+      "epoch": 7.615570307785154,
+      "eval_entropy": 0.26364776575833226,
+      "eval_loss": 0.9204733967781067,
+      "eval_mean_token_accuracy": 0.827419729714983,
+      "eval_num_tokens": 7321163.0,
+      "eval_runtime": 88.3441,
+      "eval_samples_per_second": 16.074,
+      "eval_steps_per_second": 2.015,
+      "step": 3160
+    },
+    {
+      "entropy": 0.12209134344011545,
+      "epoch": 7.663850331925166,
+      "grad_norm": 0.5159774422645569,
+      "learning_rate": 5.93388342896962e-05,
+      "loss": 0.09292978644371033,
+      "mean_token_accuracy": 0.9658421002328396,
+      "num_tokens": 7366217.0,
+      "step": 3180
+    },
+    {
+      "epoch": 7.663850331925166,
+      "eval_entropy": 0.2714110068725736,
+      "eval_loss": 0.9143092036247253,
+      "eval_mean_token_accuracy": 0.8262251099843657,
+      "eval_num_tokens": 7366217.0,
+      "eval_runtime": 87.9635,
+      "eval_samples_per_second": 16.143,
+      "eval_steps_per_second": 2.024,
+      "step": 3180
+    },
+    {
+      "entropy": 0.11955609126016498,
+      "epoch": 7.712130356065178,
+      "grad_norm": 0.7005976438522339,
+      "learning_rate": 5.705014101934252e-05,
+      "loss": 0.09319776892662049,
+      "mean_token_accuracy": 0.9674513377249241,
+      "num_tokens": 7410887.0,
+      "step": 3200
+    },
+    {
+      "epoch": 7.712130356065178,
+      "eval_entropy": 0.2626291513442993,
+      "eval_loss": 0.91884446144104,
+      "eval_mean_token_accuracy": 0.827563955207889,
+      "eval_num_tokens": 7410887.0,
+      "eval_runtime": 87.8849,
+      "eval_samples_per_second": 16.157,
+      "eval_steps_per_second": 2.025,
+      "step": 3200
+    },
+    {
+      "entropy": 0.11767532899975777,
+      "epoch": 7.76041038020519,
+      "grad_norm": 0.4675362706184387,
+      "learning_rate": 5.479854558149187e-05,
+      "loss": 0.09038087725639343,
+      "mean_token_accuracy": 0.9669680051505566,
+      "num_tokens": 7457578.0,
+      "step": 3220
+    },
+    {
+      "epoch": 7.76041038020519,
+      "eval_entropy": 0.26581037203582486,
+      "eval_loss": 0.9146883487701416,
+      "eval_mean_token_accuracy": 0.8270345812433222,
+      "eval_num_tokens": 7457578.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 3220
+    },
+    {
+      "entropy": 0.11383061073720455,
+      "epoch": 7.808690404345202,
+      "grad_norm": 0.541247546672821,
+      "learning_rate": 5.258468515101918e-05,
+      "loss": 0.08942405581474304,
+      "mean_token_accuracy": 0.968398705124855,
+      "num_tokens": 7505761.0,
+      "step": 3240
+    },
+    {
+      "epoch": 7.808690404345202,
+      "eval_entropy": 0.26687998960861997,
+      "eval_loss": 0.9037085175514221,
+      "eval_mean_token_accuracy": 0.8274977863504646,
+      "eval_num_tokens": 7505761.0,
+      "eval_runtime": 88.5163,
+      "eval_samples_per_second": 16.042,
+      "eval_steps_per_second": 2.011,
+      "step": 3240
+    },
+    {
+      "entropy": 0.12004058873280883,
+      "epoch": 7.856970428485214,
+      "grad_norm": 0.5289552211761475,
+      "learning_rate": 5.040918622423899e-05,
+      "loss": 0.09064805507659912,
+      "mean_token_accuracy": 0.9673330299556255,
+      "num_tokens": 7552557.0,
+      "step": 3260
+    },
+    {
+      "epoch": 7.856970428485214,
+      "eval_entropy": 0.26288379251622085,
+      "eval_loss": 0.9146178364753723,
+      "eval_mean_token_accuracy": 0.8276662448149049,
+      "eval_num_tokens": 7552557.0,
+      "eval_runtime": 88.0723,
+      "eval_samples_per_second": 16.123,
+      "eval_steps_per_second": 2.021,
+      "step": 3260
+    },
+    {
+      "entropy": 0.1180689456872642,
+      "epoch": 7.905250452625227,
+      "grad_norm": 0.4754861891269684,
+      "learning_rate": 4.827266444161377e-05,
+      "loss": 0.09402655363082886,
+      "mean_token_accuracy": 0.9663361787796021,
+      "num_tokens": 7596998.0,
+      "step": 3280
+    },
+    {
+      "epoch": 7.905250452625227,
+      "eval_entropy": 0.261690996922134,
+      "eval_loss": 0.9120186567306519,
+      "eval_mean_token_accuracy": 0.8278995015647974,
+      "eval_num_tokens": 7596998.0,
+      "eval_runtime": 88.6573,
+      "eval_samples_per_second": 16.017,
+      "eval_steps_per_second": 2.008,
+      "step": 3280
+    },
+    {
+      "entropy": 0.1162164019420743,
+      "epoch": 7.953530476765239,
+      "grad_norm": 0.6203398108482361,
+      "learning_rate": 4.61757244135355e-05,
+      "loss": 0.08832768201828003,
+      "mean_token_accuracy": 0.9694220684468746,
+      "num_tokens": 7643288.0,
+      "step": 3300
+    },
+    {
+      "epoch": 7.953530476765239,
+      "eval_entropy": 0.26158678129817664,
+      "eval_loss": 0.9301361441612244,
+      "eval_mean_token_accuracy": 0.8269603885291668,
+      "eval_num_tokens": 7643288.0,
+      "eval_runtime": 88.7064,
+      "eval_samples_per_second": 16.008,
+      "eval_steps_per_second": 2.007,
+      "step": 3300
+    },
+    {
+      "entropy": 0.10877202204489089,
+      "epoch": 8.0,
+      "grad_norm": 3.058673143386841,
+      "learning_rate": 4.411895954922763e-05,
+      "loss": 0.09159615635871887,
+      "mean_token_accuracy": 0.9686878834451947,
+      "num_tokens": 7688608.0,
+      "step": 3320
+    },
+    {
+      "epoch": 8.0,
+      "eval_entropy": 0.2590438057197614,
+      "eval_loss": 0.9262065887451172,
+      "eval_mean_token_accuracy": 0.8277404546067956,
+      "eval_num_tokens": 7688608.0,
+      "eval_runtime": 87.7557,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 3320
+    },
+    {
+      "entropy": 0.09969092034734786,
+      "epoch": 8.048280024140013,
+      "grad_norm": 0.2561911642551422,
+      "learning_rate": 4.2102951888817644e-05,
+      "loss": 0.07228735685348511,
+      "mean_token_accuracy": 0.9743556551635265,
+      "num_tokens": 7732856.0,
+      "step": 3340
+    },
+    {
+      "epoch": 8.048280024140013,
+      "eval_entropy": 0.2462534454933713,
+      "eval_loss": 0.9755983352661133,
+      "eval_mean_token_accuracy": 0.827510209565752,
+      "eval_num_tokens": 7732856.0,
+      "eval_runtime": 88.0297,
+      "eval_samples_per_second": 16.131,
+      "eval_steps_per_second": 2.022,
+      "step": 3340
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.2692989533782426e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3360/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3360/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3360/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3360/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3360/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3360/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3360/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3360/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3360/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3360/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3360/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3360/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3360/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3360/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..cb9b866c360a800b59d15a38bc7131adbeb4bbc7
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3360/trainer_state.json
@@ -0,0 +1,3562 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 8.096560048280024,
+  "eval_steps": 20,
+  "global_step": 3360,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    },
+    {
+      "entropy": 0.1326121488586068,
+      "epoch": 7.181050090525045,
+      "grad_norm": 0.4843117296695709,
+      "learning_rate": 8.411870386507724e-05,
+      "loss": 0.09537842869758606,
+      "mean_token_accuracy": 0.9645151920616627,
+      "num_tokens": 6898444.0,
+      "step": 2980
+    },
+    {
+      "epoch": 7.181050090525045,
+      "eval_entropy": 0.27999477783280813,
+      "eval_loss": 0.9031037092208862,
+      "eval_mean_token_accuracy": 0.8261882204018282,
+      "eval_num_tokens": 6898444.0,
+      "eval_runtime": 88.5602,
+      "eval_samples_per_second": 16.034,
+      "eval_steps_per_second": 2.01,
+      "step": 2980
+    },
+    {
+      "entropy": 0.1156751093454659,
+      "epoch": 7.229330114665057,
+      "grad_norm": 0.5297922492027283,
+      "learning_rate": 8.149629351430326e-05,
+      "loss": 0.09203402996063233,
+      "mean_token_accuracy": 0.9689733698964119,
+      "num_tokens": 6947106.0,
+      "step": 3000
+    },
+    {
+      "epoch": 7.229330114665057,
+      "eval_entropy": 0.2740225845508361,
+      "eval_loss": 0.8876440525054932,
+      "eval_mean_token_accuracy": 0.8265116857678703,
+      "eval_num_tokens": 6947106.0,
+      "eval_runtime": 87.7971,
+      "eval_samples_per_second": 16.174,
+      "eval_steps_per_second": 2.027,
+      "step": 3000
+    },
+    {
+      "entropy": 0.12116114404052496,
+      "epoch": 7.27761013880507,
+      "grad_norm": 0.5467398166656494,
+      "learning_rate": 7.890406302404618e-05,
+      "loss": 0.0933056652545929,
+      "mean_token_accuracy": 0.9676455594599247,
+      "num_tokens": 6992784.0,
+      "step": 3020
+    },
+    {
+      "epoch": 7.27761013880507,
+      "eval_entropy": 0.27238342793804876,
+      "eval_loss": 0.9166117310523987,
+      "eval_mean_token_accuracy": 0.825499015242866,
+      "eval_num_tokens": 6992784.0,
+      "eval_runtime": 88.3276,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 3020
+    },
+    {
+      "entropy": 0.11715119928121567,
+      "epoch": 7.325890162945082,
+      "grad_norm": 0.4976702332496643,
+      "learning_rate": 7.634274596487176e-05,
+      "loss": 0.09449659585952759,
+      "mean_token_accuracy": 0.966506478190422,
+      "num_tokens": 7040482.0,
+      "step": 3040
+    },
+    {
+      "epoch": 7.325890162945082,
+      "eval_entropy": 0.2788975556077582,
+      "eval_loss": 0.8833065629005432,
+      "eval_mean_token_accuracy": 0.8269920596915684,
+      "eval_num_tokens": 7040482.0,
+      "eval_runtime": 88.9016,
+      "eval_samples_per_second": 15.973,
+      "eval_steps_per_second": 2.002,
+      "step": 3040
+    },
+    {
+      "entropy": 0.12565911412239075,
+      "epoch": 7.3741701870850935,
+      "grad_norm": 0.5728679895401001,
+      "learning_rate": 7.381306715921025e-05,
+      "loss": 0.09626244902610778,
+      "mean_token_accuracy": 0.9661444149911403,
+      "num_tokens": 7086261.0,
+      "step": 3060
+    },
+    {
+      "epoch": 7.3741701870850935,
+      "eval_entropy": 0.2762492011939542,
+      "eval_loss": 0.8961160778999329,
+      "eval_mean_token_accuracy": 0.826369910092836,
+      "eval_num_tokens": 7086261.0,
+      "eval_runtime": 88.5204,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 3060
+    },
+    {
+      "entropy": 0.12258567921817302,
+      "epoch": 7.422450211225105,
+      "grad_norm": 0.468985915184021,
+      "learning_rate": 7.131574247624031e-05,
+      "loss": 0.09120266437530518,
+      "mean_token_accuracy": 0.9674938447773457,
+      "num_tokens": 7133247.0,
+      "step": 3080
+    },
+    {
+      "epoch": 7.422450211225105,
+      "eval_entropy": 0.2688512086366,
+      "eval_loss": 0.9048654437065125,
+      "eval_mean_token_accuracy": 0.8265582302982888,
+      "eval_num_tokens": 7133247.0,
+      "eval_runtime": 88.164,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 3080
+    },
+    {
+      "entropy": 0.12070536343380808,
+      "epoch": 7.470730235365117,
+      "grad_norm": 0.5310841798782349,
+      "learning_rate": 6.885147862930688e-05,
+      "loss": 0.09386247992515565,
+      "mean_token_accuracy": 0.96664222702384,
+      "num_tokens": 7177810.0,
+      "step": 3100
+    },
+    {
+      "epoch": 7.470730235365117,
+      "eval_entropy": 0.26289632874593305,
+      "eval_loss": 0.902837336063385,
+      "eval_mean_token_accuracy": 0.828052184554968,
+      "eval_num_tokens": 7177810.0,
+      "eval_runtime": 88.1201,
+      "eval_samples_per_second": 16.114,
+      "eval_steps_per_second": 2.02,
+      "step": 3100
+    },
+    {
+      "entropy": 0.11322665181942285,
+      "epoch": 7.51901025950513,
+      "grad_norm": 0.6846067905426025,
+      "learning_rate": 6.642097297592908e-05,
+      "loss": 0.08772538304328918,
+      "mean_token_accuracy": 0.9691362135112286,
+      "num_tokens": 7225302.0,
+      "step": 3120
+    },
+    {
+      "epoch": 7.51901025950513,
+      "eval_entropy": 0.2670605613776807,
+      "eval_loss": 0.9130566120147705,
+      "eval_mean_token_accuracy": 0.826366903741708,
+      "eval_num_tokens": 7225302.0,
+      "eval_runtime": 88.6035,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3120
+    },
+    {
+      "entropy": 0.10985926580615342,
+      "epoch": 7.567290283645142,
+      "grad_norm": 0.4799748361110687,
+      "learning_rate": 6.402491332045683e-05,
+      "loss": 0.08672296404838561,
+      "mean_token_accuracy": 0.9698824048042297,
+      "num_tokens": 7274881.0,
+      "step": 3140
+    },
+    {
+      "epoch": 7.567290283645142,
+      "eval_entropy": 0.26657010866015146,
+      "eval_loss": 0.9056380987167358,
+      "eval_mean_token_accuracy": 0.8272326461384806,
+      "eval_num_tokens": 7274881.0,
+      "eval_runtime": 88.2967,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3140
+    },
+    {
+      "entropy": 0.11890473761595785,
+      "epoch": 7.615570307785154,
+      "grad_norm": 0.4081912934780121,
+      "learning_rate": 6.166397771942987e-05,
+      "loss": 0.09396827220916748,
+      "mean_token_accuracy": 0.967512384802103,
+      "num_tokens": 7321163.0,
+      "step": 3160
+    },
+    {
+      "epoch": 7.615570307785154,
+      "eval_entropy": 0.26364776575833226,
+      "eval_loss": 0.9204733967781067,
+      "eval_mean_token_accuracy": 0.827419729714983,
+      "eval_num_tokens": 7321163.0,
+      "eval_runtime": 88.3441,
+      "eval_samples_per_second": 16.074,
+      "eval_steps_per_second": 2.015,
+      "step": 3160
+    },
+    {
+      "entropy": 0.12209134344011545,
+      "epoch": 7.663850331925166,
+      "grad_norm": 0.5159774422645569,
+      "learning_rate": 5.93388342896962e-05,
+      "loss": 0.09292978644371033,
+      "mean_token_accuracy": 0.9658421002328396,
+      "num_tokens": 7366217.0,
+      "step": 3180
+    },
+    {
+      "epoch": 7.663850331925166,
+      "eval_entropy": 0.2714110068725736,
+      "eval_loss": 0.9143092036247253,
+      "eval_mean_token_accuracy": 0.8262251099843657,
+      "eval_num_tokens": 7366217.0,
+      "eval_runtime": 87.9635,
+      "eval_samples_per_second": 16.143,
+      "eval_steps_per_second": 2.024,
+      "step": 3180
+    },
+    {
+      "entropy": 0.11955609126016498,
+      "epoch": 7.712130356065178,
+      "grad_norm": 0.7005976438522339,
+      "learning_rate": 5.705014101934252e-05,
+      "loss": 0.09319776892662049,
+      "mean_token_accuracy": 0.9674513377249241,
+      "num_tokens": 7410887.0,
+      "step": 3200
+    },
+    {
+      "epoch": 7.712130356065178,
+      "eval_entropy": 0.2626291513442993,
+      "eval_loss": 0.91884446144104,
+      "eval_mean_token_accuracy": 0.827563955207889,
+      "eval_num_tokens": 7410887.0,
+      "eval_runtime": 87.8849,
+      "eval_samples_per_second": 16.157,
+      "eval_steps_per_second": 2.025,
+      "step": 3200
+    },
+    {
+      "entropy": 0.11767532899975777,
+      "epoch": 7.76041038020519,
+      "grad_norm": 0.4675362706184387,
+      "learning_rate": 5.479854558149187e-05,
+      "loss": 0.09038087725639343,
+      "mean_token_accuracy": 0.9669680051505566,
+      "num_tokens": 7457578.0,
+      "step": 3220
+    },
+    {
+      "epoch": 7.76041038020519,
+      "eval_entropy": 0.26581037203582486,
+      "eval_loss": 0.9146883487701416,
+      "eval_mean_token_accuracy": 0.8270345812433222,
+      "eval_num_tokens": 7457578.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 3220
+    },
+    {
+      "entropy": 0.11383061073720455,
+      "epoch": 7.808690404345202,
+      "grad_norm": 0.541247546672821,
+      "learning_rate": 5.258468515101918e-05,
+      "loss": 0.08942405581474304,
+      "mean_token_accuracy": 0.968398705124855,
+      "num_tokens": 7505761.0,
+      "step": 3240
+    },
+    {
+      "epoch": 7.808690404345202,
+      "eval_entropy": 0.26687998960861997,
+      "eval_loss": 0.9037085175514221,
+      "eval_mean_token_accuracy": 0.8274977863504646,
+      "eval_num_tokens": 7505761.0,
+      "eval_runtime": 88.5163,
+      "eval_samples_per_second": 16.042,
+      "eval_steps_per_second": 2.011,
+      "step": 3240
+    },
+    {
+      "entropy": 0.12004058873280883,
+      "epoch": 7.856970428485214,
+      "grad_norm": 0.5289552211761475,
+      "learning_rate": 5.040918622423899e-05,
+      "loss": 0.09064805507659912,
+      "mean_token_accuracy": 0.9673330299556255,
+      "num_tokens": 7552557.0,
+      "step": 3260
+    },
+    {
+      "epoch": 7.856970428485214,
+      "eval_entropy": 0.26288379251622085,
+      "eval_loss": 0.9146178364753723,
+      "eval_mean_token_accuracy": 0.8276662448149049,
+      "eval_num_tokens": 7552557.0,
+      "eval_runtime": 88.0723,
+      "eval_samples_per_second": 16.123,
+      "eval_steps_per_second": 2.021,
+      "step": 3260
+    },
+    {
+      "entropy": 0.1180689456872642,
+      "epoch": 7.905250452625227,
+      "grad_norm": 0.4754861891269684,
+      "learning_rate": 4.827266444161377e-05,
+      "loss": 0.09402655363082886,
+      "mean_token_accuracy": 0.9663361787796021,
+      "num_tokens": 7596998.0,
+      "step": 3280
+    },
+    {
+      "epoch": 7.905250452625227,
+      "eval_entropy": 0.261690996922134,
+      "eval_loss": 0.9120186567306519,
+      "eval_mean_token_accuracy": 0.8278995015647974,
+      "eval_num_tokens": 7596998.0,
+      "eval_runtime": 88.6573,
+      "eval_samples_per_second": 16.017,
+      "eval_steps_per_second": 2.008,
+      "step": 3280
+    },
+    {
+      "entropy": 0.1162164019420743,
+      "epoch": 7.953530476765239,
+      "grad_norm": 0.6203398108482361,
+      "learning_rate": 4.61757244135355e-05,
+      "loss": 0.08832768201828003,
+      "mean_token_accuracy": 0.9694220684468746,
+      "num_tokens": 7643288.0,
+      "step": 3300
+    },
+    {
+      "epoch": 7.953530476765239,
+      "eval_entropy": 0.26158678129817664,
+      "eval_loss": 0.9301361441612244,
+      "eval_mean_token_accuracy": 0.8269603885291668,
+      "eval_num_tokens": 7643288.0,
+      "eval_runtime": 88.7064,
+      "eval_samples_per_second": 16.008,
+      "eval_steps_per_second": 2.007,
+      "step": 3300
+    },
+    {
+      "entropy": 0.10877202204489089,
+      "epoch": 8.0,
+      "grad_norm": 3.058673143386841,
+      "learning_rate": 4.411895954922763e-05,
+      "loss": 0.09159615635871887,
+      "mean_token_accuracy": 0.9686878834451947,
+      "num_tokens": 7688608.0,
+      "step": 3320
+    },
+    {
+      "epoch": 8.0,
+      "eval_entropy": 0.2590438057197614,
+      "eval_loss": 0.9262065887451172,
+      "eval_mean_token_accuracy": 0.8277404546067956,
+      "eval_num_tokens": 7688608.0,
+      "eval_runtime": 87.7557,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 3320
+    },
+    {
+      "entropy": 0.09969092034734786,
+      "epoch": 8.048280024140013,
+      "grad_norm": 0.2561911642551422,
+      "learning_rate": 4.2102951888817644e-05,
+      "loss": 0.07228735685348511,
+      "mean_token_accuracy": 0.9743556551635265,
+      "num_tokens": 7732856.0,
+      "step": 3340
+    },
+    {
+      "epoch": 8.048280024140013,
+      "eval_entropy": 0.2462534454933713,
+      "eval_loss": 0.9755983352661133,
+      "eval_mean_token_accuracy": 0.827510209565752,
+      "eval_num_tokens": 7732856.0,
+      "eval_runtime": 88.0297,
+      "eval_samples_per_second": 16.131,
+      "eval_steps_per_second": 2.022,
+      "step": 3340
+    },
+    {
+      "entropy": 0.09477624730207027,
+      "epoch": 8.096560048280024,
+      "grad_norm": 0.3670356571674347,
+      "learning_rate": 4.012827193862633e-05,
+      "loss": 0.0686036467552185,
+      "mean_token_accuracy": 0.9743670664727688,
+      "num_tokens": 7778585.0,
+      "step": 3360
+    },
+    {
+      "epoch": 8.096560048280024,
+      "eval_entropy": 0.2440882951356052,
+      "eval_loss": 1.0118423700332642,
+      "eval_mean_token_accuracy": 0.8271241070849172,
+      "eval_num_tokens": 7778585.0,
+      "eval_runtime": 88.1364,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 3360
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.287402541359309e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3380/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3380/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3380/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3380/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3380/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3380/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3380/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3380/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3380/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3380/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3380/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3380/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3380/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3380/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..a2fc8f1636d418550b9924543d042452222d6c21
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3380/trainer_state.json
@@ -0,0 +1,3583 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 8.144840072420036,
+  "eval_steps": 20,
+  "global_step": 3380,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    },
+    {
+      "entropy": 0.1326121488586068,
+      "epoch": 7.181050090525045,
+      "grad_norm": 0.4843117296695709,
+      "learning_rate": 8.411870386507724e-05,
+      "loss": 0.09537842869758606,
+      "mean_token_accuracy": 0.9645151920616627,
+      "num_tokens": 6898444.0,
+      "step": 2980
+    },
+    {
+      "epoch": 7.181050090525045,
+      "eval_entropy": 0.27999477783280813,
+      "eval_loss": 0.9031037092208862,
+      "eval_mean_token_accuracy": 0.8261882204018282,
+      "eval_num_tokens": 6898444.0,
+      "eval_runtime": 88.5602,
+      "eval_samples_per_second": 16.034,
+      "eval_steps_per_second": 2.01,
+      "step": 2980
+    },
+    {
+      "entropy": 0.1156751093454659,
+      "epoch": 7.229330114665057,
+      "grad_norm": 0.5297922492027283,
+      "learning_rate": 8.149629351430326e-05,
+      "loss": 0.09203402996063233,
+      "mean_token_accuracy": 0.9689733698964119,
+      "num_tokens": 6947106.0,
+      "step": 3000
+    },
+    {
+      "epoch": 7.229330114665057,
+      "eval_entropy": 0.2740225845508361,
+      "eval_loss": 0.8876440525054932,
+      "eval_mean_token_accuracy": 0.8265116857678703,
+      "eval_num_tokens": 6947106.0,
+      "eval_runtime": 87.7971,
+      "eval_samples_per_second": 16.174,
+      "eval_steps_per_second": 2.027,
+      "step": 3000
+    },
+    {
+      "entropy": 0.12116114404052496,
+      "epoch": 7.27761013880507,
+      "grad_norm": 0.5467398166656494,
+      "learning_rate": 7.890406302404618e-05,
+      "loss": 0.0933056652545929,
+      "mean_token_accuracy": 0.9676455594599247,
+      "num_tokens": 6992784.0,
+      "step": 3020
+    },
+    {
+      "epoch": 7.27761013880507,
+      "eval_entropy": 0.27238342793804876,
+      "eval_loss": 0.9166117310523987,
+      "eval_mean_token_accuracy": 0.825499015242866,
+      "eval_num_tokens": 6992784.0,
+      "eval_runtime": 88.3276,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 3020
+    },
+    {
+      "entropy": 0.11715119928121567,
+      "epoch": 7.325890162945082,
+      "grad_norm": 0.4976702332496643,
+      "learning_rate": 7.634274596487176e-05,
+      "loss": 0.09449659585952759,
+      "mean_token_accuracy": 0.966506478190422,
+      "num_tokens": 7040482.0,
+      "step": 3040
+    },
+    {
+      "epoch": 7.325890162945082,
+      "eval_entropy": 0.2788975556077582,
+      "eval_loss": 0.8833065629005432,
+      "eval_mean_token_accuracy": 0.8269920596915684,
+      "eval_num_tokens": 7040482.0,
+      "eval_runtime": 88.9016,
+      "eval_samples_per_second": 15.973,
+      "eval_steps_per_second": 2.002,
+      "step": 3040
+    },
+    {
+      "entropy": 0.12565911412239075,
+      "epoch": 7.3741701870850935,
+      "grad_norm": 0.5728679895401001,
+      "learning_rate": 7.381306715921025e-05,
+      "loss": 0.09626244902610778,
+      "mean_token_accuracy": 0.9661444149911403,
+      "num_tokens": 7086261.0,
+      "step": 3060
+    },
+    {
+      "epoch": 7.3741701870850935,
+      "eval_entropy": 0.2762492011939542,
+      "eval_loss": 0.8961160778999329,
+      "eval_mean_token_accuracy": 0.826369910092836,
+      "eval_num_tokens": 7086261.0,
+      "eval_runtime": 88.5204,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 3060
+    },
+    {
+      "entropy": 0.12258567921817302,
+      "epoch": 7.422450211225105,
+      "grad_norm": 0.468985915184021,
+      "learning_rate": 7.131574247624031e-05,
+      "loss": 0.09120266437530518,
+      "mean_token_accuracy": 0.9674938447773457,
+      "num_tokens": 7133247.0,
+      "step": 3080
+    },
+    {
+      "epoch": 7.422450211225105,
+      "eval_entropy": 0.2688512086366,
+      "eval_loss": 0.9048654437065125,
+      "eval_mean_token_accuracy": 0.8265582302982888,
+      "eval_num_tokens": 7133247.0,
+      "eval_runtime": 88.164,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 3080
+    },
+    {
+      "entropy": 0.12070536343380808,
+      "epoch": 7.470730235365117,
+      "grad_norm": 0.5310841798782349,
+      "learning_rate": 6.885147862930688e-05,
+      "loss": 0.09386247992515565,
+      "mean_token_accuracy": 0.96664222702384,
+      "num_tokens": 7177810.0,
+      "step": 3100
+    },
+    {
+      "epoch": 7.470730235365117,
+      "eval_entropy": 0.26289632874593305,
+      "eval_loss": 0.902837336063385,
+      "eval_mean_token_accuracy": 0.828052184554968,
+      "eval_num_tokens": 7177810.0,
+      "eval_runtime": 88.1201,
+      "eval_samples_per_second": 16.114,
+      "eval_steps_per_second": 2.02,
+      "step": 3100
+    },
+    {
+      "entropy": 0.11322665181942285,
+      "epoch": 7.51901025950513,
+      "grad_norm": 0.6846067905426025,
+      "learning_rate": 6.642097297592908e-05,
+      "loss": 0.08772538304328918,
+      "mean_token_accuracy": 0.9691362135112286,
+      "num_tokens": 7225302.0,
+      "step": 3120
+    },
+    {
+      "epoch": 7.51901025950513,
+      "eval_entropy": 0.2670605613776807,
+      "eval_loss": 0.9130566120147705,
+      "eval_mean_token_accuracy": 0.826366903741708,
+      "eval_num_tokens": 7225302.0,
+      "eval_runtime": 88.6035,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3120
+    },
+    {
+      "entropy": 0.10985926580615342,
+      "epoch": 7.567290283645142,
+      "grad_norm": 0.4799748361110687,
+      "learning_rate": 6.402491332045683e-05,
+      "loss": 0.08672296404838561,
+      "mean_token_accuracy": 0.9698824048042297,
+      "num_tokens": 7274881.0,
+      "step": 3140
+    },
+    {
+      "epoch": 7.567290283645142,
+      "eval_entropy": 0.26657010866015146,
+      "eval_loss": 0.9056380987167358,
+      "eval_mean_token_accuracy": 0.8272326461384806,
+      "eval_num_tokens": 7274881.0,
+      "eval_runtime": 88.2967,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3140
+    },
+    {
+      "entropy": 0.11890473761595785,
+      "epoch": 7.615570307785154,
+      "grad_norm": 0.4081912934780121,
+      "learning_rate": 6.166397771942987e-05,
+      "loss": 0.09396827220916748,
+      "mean_token_accuracy": 0.967512384802103,
+      "num_tokens": 7321163.0,
+      "step": 3160
+    },
+    {
+      "epoch": 7.615570307785154,
+      "eval_entropy": 0.26364776575833226,
+      "eval_loss": 0.9204733967781067,
+      "eval_mean_token_accuracy": 0.827419729714983,
+      "eval_num_tokens": 7321163.0,
+      "eval_runtime": 88.3441,
+      "eval_samples_per_second": 16.074,
+      "eval_steps_per_second": 2.015,
+      "step": 3160
+    },
+    {
+      "entropy": 0.12209134344011545,
+      "epoch": 7.663850331925166,
+      "grad_norm": 0.5159774422645569,
+      "learning_rate": 5.93388342896962e-05,
+      "loss": 0.09292978644371033,
+      "mean_token_accuracy": 0.9658421002328396,
+      "num_tokens": 7366217.0,
+      "step": 3180
+    },
+    {
+      "epoch": 7.663850331925166,
+      "eval_entropy": 0.2714110068725736,
+      "eval_loss": 0.9143092036247253,
+      "eval_mean_token_accuracy": 0.8262251099843657,
+      "eval_num_tokens": 7366217.0,
+      "eval_runtime": 87.9635,
+      "eval_samples_per_second": 16.143,
+      "eval_steps_per_second": 2.024,
+      "step": 3180
+    },
+    {
+      "entropy": 0.11955609126016498,
+      "epoch": 7.712130356065178,
+      "grad_norm": 0.7005976438522339,
+      "learning_rate": 5.705014101934252e-05,
+      "loss": 0.09319776892662049,
+      "mean_token_accuracy": 0.9674513377249241,
+      "num_tokens": 7410887.0,
+      "step": 3200
+    },
+    {
+      "epoch": 7.712130356065178,
+      "eval_entropy": 0.2626291513442993,
+      "eval_loss": 0.91884446144104,
+      "eval_mean_token_accuracy": 0.827563955207889,
+      "eval_num_tokens": 7410887.0,
+      "eval_runtime": 87.8849,
+      "eval_samples_per_second": 16.157,
+      "eval_steps_per_second": 2.025,
+      "step": 3200
+    },
+    {
+      "entropy": 0.11767532899975777,
+      "epoch": 7.76041038020519,
+      "grad_norm": 0.4675362706184387,
+      "learning_rate": 5.479854558149187e-05,
+      "loss": 0.09038087725639343,
+      "mean_token_accuracy": 0.9669680051505566,
+      "num_tokens": 7457578.0,
+      "step": 3220
+    },
+    {
+      "epoch": 7.76041038020519,
+      "eval_entropy": 0.26581037203582486,
+      "eval_loss": 0.9146883487701416,
+      "eval_mean_token_accuracy": 0.8270345812433222,
+      "eval_num_tokens": 7457578.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 3220
+    },
+    {
+      "entropy": 0.11383061073720455,
+      "epoch": 7.808690404345202,
+      "grad_norm": 0.541247546672821,
+      "learning_rate": 5.258468515101918e-05,
+      "loss": 0.08942405581474304,
+      "mean_token_accuracy": 0.968398705124855,
+      "num_tokens": 7505761.0,
+      "step": 3240
+    },
+    {
+      "epoch": 7.808690404345202,
+      "eval_entropy": 0.26687998960861997,
+      "eval_loss": 0.9037085175514221,
+      "eval_mean_token_accuracy": 0.8274977863504646,
+      "eval_num_tokens": 7505761.0,
+      "eval_runtime": 88.5163,
+      "eval_samples_per_second": 16.042,
+      "eval_steps_per_second": 2.011,
+      "step": 3240
+    },
+    {
+      "entropy": 0.12004058873280883,
+      "epoch": 7.856970428485214,
+      "grad_norm": 0.5289552211761475,
+      "learning_rate": 5.040918622423899e-05,
+      "loss": 0.09064805507659912,
+      "mean_token_accuracy": 0.9673330299556255,
+      "num_tokens": 7552557.0,
+      "step": 3260
+    },
+    {
+      "epoch": 7.856970428485214,
+      "eval_entropy": 0.26288379251622085,
+      "eval_loss": 0.9146178364753723,
+      "eval_mean_token_accuracy": 0.8276662448149049,
+      "eval_num_tokens": 7552557.0,
+      "eval_runtime": 88.0723,
+      "eval_samples_per_second": 16.123,
+      "eval_steps_per_second": 2.021,
+      "step": 3260
+    },
+    {
+      "entropy": 0.1180689456872642,
+      "epoch": 7.905250452625227,
+      "grad_norm": 0.4754861891269684,
+      "learning_rate": 4.827266444161377e-05,
+      "loss": 0.09402655363082886,
+      "mean_token_accuracy": 0.9663361787796021,
+      "num_tokens": 7596998.0,
+      "step": 3280
+    },
+    {
+      "epoch": 7.905250452625227,
+      "eval_entropy": 0.261690996922134,
+      "eval_loss": 0.9120186567306519,
+      "eval_mean_token_accuracy": 0.8278995015647974,
+      "eval_num_tokens": 7596998.0,
+      "eval_runtime": 88.6573,
+      "eval_samples_per_second": 16.017,
+      "eval_steps_per_second": 2.008,
+      "step": 3280
+    },
+    {
+      "entropy": 0.1162164019420743,
+      "epoch": 7.953530476765239,
+      "grad_norm": 0.6203398108482361,
+      "learning_rate": 4.61757244135355e-05,
+      "loss": 0.08832768201828003,
+      "mean_token_accuracy": 0.9694220684468746,
+      "num_tokens": 7643288.0,
+      "step": 3300
+    },
+    {
+      "epoch": 7.953530476765239,
+      "eval_entropy": 0.26158678129817664,
+      "eval_loss": 0.9301361441612244,
+      "eval_mean_token_accuracy": 0.8269603885291668,
+      "eval_num_tokens": 7643288.0,
+      "eval_runtime": 88.7064,
+      "eval_samples_per_second": 16.008,
+      "eval_steps_per_second": 2.007,
+      "step": 3300
+    },
+    {
+      "entropy": 0.10877202204489089,
+      "epoch": 8.0,
+      "grad_norm": 3.058673143386841,
+      "learning_rate": 4.411895954922763e-05,
+      "loss": 0.09159615635871887,
+      "mean_token_accuracy": 0.9686878834451947,
+      "num_tokens": 7688608.0,
+      "step": 3320
+    },
+    {
+      "epoch": 8.0,
+      "eval_entropy": 0.2590438057197614,
+      "eval_loss": 0.9262065887451172,
+      "eval_mean_token_accuracy": 0.8277404546067956,
+      "eval_num_tokens": 7688608.0,
+      "eval_runtime": 87.7557,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 3320
+    },
+    {
+      "entropy": 0.09969092034734786,
+      "epoch": 8.048280024140013,
+      "grad_norm": 0.2561911642551422,
+      "learning_rate": 4.2102951888817644e-05,
+      "loss": 0.07228735685348511,
+      "mean_token_accuracy": 0.9743556551635265,
+      "num_tokens": 7732856.0,
+      "step": 3340
+    },
+    {
+      "epoch": 8.048280024140013,
+      "eval_entropy": 0.2462534454933713,
+      "eval_loss": 0.9755983352661133,
+      "eval_mean_token_accuracy": 0.827510209565752,
+      "eval_num_tokens": 7732856.0,
+      "eval_runtime": 88.0297,
+      "eval_samples_per_second": 16.131,
+      "eval_steps_per_second": 2.022,
+      "step": 3340
+    },
+    {
+      "entropy": 0.09477624730207027,
+      "epoch": 8.096560048280024,
+      "grad_norm": 0.3670356571674347,
+      "learning_rate": 4.012827193862633e-05,
+      "loss": 0.0686036467552185,
+      "mean_token_accuracy": 0.9743670664727688,
+      "num_tokens": 7778585.0,
+      "step": 3360
+    },
+    {
+      "epoch": 8.096560048280024,
+      "eval_entropy": 0.2440882951356052,
+      "eval_loss": 1.0118423700332642,
+      "eval_mean_token_accuracy": 0.8271241070849172,
+      "eval_num_tokens": 7778585.0,
+      "eval_runtime": 88.1364,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 3360
+    },
+    {
+      "entropy": 0.09417980071157217,
+      "epoch": 8.144840072420036,
+      "grad_norm": 0.5785111784934998,
+      "learning_rate": 3.819547850972133e-05,
+      "loss": 0.07100698947906495,
+      "mean_token_accuracy": 0.9737804099917412,
+      "num_tokens": 7824287.0,
+      "step": 3380
+    },
+    {
+      "epoch": 8.144840072420036,
+      "eval_entropy": 0.24603172860453637,
+      "eval_loss": 1.0021991729736328,
+      "eval_mean_token_accuracy": 0.827162122793412,
+      "eval_num_tokens": 7824287.0,
+      "eval_runtime": 88.4534,
+      "eval_samples_per_second": 16.054,
+      "eval_steps_per_second": 2.012,
+      "step": 3380
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.3070622764999066e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-340/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-340/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-340/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-340/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-340/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-340/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-340/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-340/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-340/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-340/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-340/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-340/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-340/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-340/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..74f3aae32ad3b234a7b009ba629139b2fe08d2e5
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-340/trainer_state.json
@@ -0,0 +1,391 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.8207604103802052,
+  "eval_steps": 20,
+  "global_step": 340,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.378479694237696e+16,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3400/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3400/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3400/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3400/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3400/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3400/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3400/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3400/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3400/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3400/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3400/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3400/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3400/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3400/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..08fc4c3570c19f6e0a2c0d8a76c14e4ef01f07f0
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3400/trainer_state.json
@@ -0,0 +1,3604 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 8.193120096560047,
+  "eval_steps": 20,
+  "global_step": 3400,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    },
+    {
+      "entropy": 0.1326121488586068,
+      "epoch": 7.181050090525045,
+      "grad_norm": 0.4843117296695709,
+      "learning_rate": 8.411870386507724e-05,
+      "loss": 0.09537842869758606,
+      "mean_token_accuracy": 0.9645151920616627,
+      "num_tokens": 6898444.0,
+      "step": 2980
+    },
+    {
+      "epoch": 7.181050090525045,
+      "eval_entropy": 0.27999477783280813,
+      "eval_loss": 0.9031037092208862,
+      "eval_mean_token_accuracy": 0.8261882204018282,
+      "eval_num_tokens": 6898444.0,
+      "eval_runtime": 88.5602,
+      "eval_samples_per_second": 16.034,
+      "eval_steps_per_second": 2.01,
+      "step": 2980
+    },
+    {
+      "entropy": 0.1156751093454659,
+      "epoch": 7.229330114665057,
+      "grad_norm": 0.5297922492027283,
+      "learning_rate": 8.149629351430326e-05,
+      "loss": 0.09203402996063233,
+      "mean_token_accuracy": 0.9689733698964119,
+      "num_tokens": 6947106.0,
+      "step": 3000
+    },
+    {
+      "epoch": 7.229330114665057,
+      "eval_entropy": 0.2740225845508361,
+      "eval_loss": 0.8876440525054932,
+      "eval_mean_token_accuracy": 0.8265116857678703,
+      "eval_num_tokens": 6947106.0,
+      "eval_runtime": 87.7971,
+      "eval_samples_per_second": 16.174,
+      "eval_steps_per_second": 2.027,
+      "step": 3000
+    },
+    {
+      "entropy": 0.12116114404052496,
+      "epoch": 7.27761013880507,
+      "grad_norm": 0.5467398166656494,
+      "learning_rate": 7.890406302404618e-05,
+      "loss": 0.0933056652545929,
+      "mean_token_accuracy": 0.9676455594599247,
+      "num_tokens": 6992784.0,
+      "step": 3020
+    },
+    {
+      "epoch": 7.27761013880507,
+      "eval_entropy": 0.27238342793804876,
+      "eval_loss": 0.9166117310523987,
+      "eval_mean_token_accuracy": 0.825499015242866,
+      "eval_num_tokens": 6992784.0,
+      "eval_runtime": 88.3276,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 3020
+    },
+    {
+      "entropy": 0.11715119928121567,
+      "epoch": 7.325890162945082,
+      "grad_norm": 0.4976702332496643,
+      "learning_rate": 7.634274596487176e-05,
+      "loss": 0.09449659585952759,
+      "mean_token_accuracy": 0.966506478190422,
+      "num_tokens": 7040482.0,
+      "step": 3040
+    },
+    {
+      "epoch": 7.325890162945082,
+      "eval_entropy": 0.2788975556077582,
+      "eval_loss": 0.8833065629005432,
+      "eval_mean_token_accuracy": 0.8269920596915684,
+      "eval_num_tokens": 7040482.0,
+      "eval_runtime": 88.9016,
+      "eval_samples_per_second": 15.973,
+      "eval_steps_per_second": 2.002,
+      "step": 3040
+    },
+    {
+      "entropy": 0.12565911412239075,
+      "epoch": 7.3741701870850935,
+      "grad_norm": 0.5728679895401001,
+      "learning_rate": 7.381306715921025e-05,
+      "loss": 0.09626244902610778,
+      "mean_token_accuracy": 0.9661444149911403,
+      "num_tokens": 7086261.0,
+      "step": 3060
+    },
+    {
+      "epoch": 7.3741701870850935,
+      "eval_entropy": 0.2762492011939542,
+      "eval_loss": 0.8961160778999329,
+      "eval_mean_token_accuracy": 0.826369910092836,
+      "eval_num_tokens": 7086261.0,
+      "eval_runtime": 88.5204,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 3060
+    },
+    {
+      "entropy": 0.12258567921817302,
+      "epoch": 7.422450211225105,
+      "grad_norm": 0.468985915184021,
+      "learning_rate": 7.131574247624031e-05,
+      "loss": 0.09120266437530518,
+      "mean_token_accuracy": 0.9674938447773457,
+      "num_tokens": 7133247.0,
+      "step": 3080
+    },
+    {
+      "epoch": 7.422450211225105,
+      "eval_entropy": 0.2688512086366,
+      "eval_loss": 0.9048654437065125,
+      "eval_mean_token_accuracy": 0.8265582302982888,
+      "eval_num_tokens": 7133247.0,
+      "eval_runtime": 88.164,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 3080
+    },
+    {
+      "entropy": 0.12070536343380808,
+      "epoch": 7.470730235365117,
+      "grad_norm": 0.5310841798782349,
+      "learning_rate": 6.885147862930688e-05,
+      "loss": 0.09386247992515565,
+      "mean_token_accuracy": 0.96664222702384,
+      "num_tokens": 7177810.0,
+      "step": 3100
+    },
+    {
+      "epoch": 7.470730235365117,
+      "eval_entropy": 0.26289632874593305,
+      "eval_loss": 0.902837336063385,
+      "eval_mean_token_accuracy": 0.828052184554968,
+      "eval_num_tokens": 7177810.0,
+      "eval_runtime": 88.1201,
+      "eval_samples_per_second": 16.114,
+      "eval_steps_per_second": 2.02,
+      "step": 3100
+    },
+    {
+      "entropy": 0.11322665181942285,
+      "epoch": 7.51901025950513,
+      "grad_norm": 0.6846067905426025,
+      "learning_rate": 6.642097297592908e-05,
+      "loss": 0.08772538304328918,
+      "mean_token_accuracy": 0.9691362135112286,
+      "num_tokens": 7225302.0,
+      "step": 3120
+    },
+    {
+      "epoch": 7.51901025950513,
+      "eval_entropy": 0.2670605613776807,
+      "eval_loss": 0.9130566120147705,
+      "eval_mean_token_accuracy": 0.826366903741708,
+      "eval_num_tokens": 7225302.0,
+      "eval_runtime": 88.6035,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3120
+    },
+    {
+      "entropy": 0.10985926580615342,
+      "epoch": 7.567290283645142,
+      "grad_norm": 0.4799748361110687,
+      "learning_rate": 6.402491332045683e-05,
+      "loss": 0.08672296404838561,
+      "mean_token_accuracy": 0.9698824048042297,
+      "num_tokens": 7274881.0,
+      "step": 3140
+    },
+    {
+      "epoch": 7.567290283645142,
+      "eval_entropy": 0.26657010866015146,
+      "eval_loss": 0.9056380987167358,
+      "eval_mean_token_accuracy": 0.8272326461384806,
+      "eval_num_tokens": 7274881.0,
+      "eval_runtime": 88.2967,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3140
+    },
+    {
+      "entropy": 0.11890473761595785,
+      "epoch": 7.615570307785154,
+      "grad_norm": 0.4081912934780121,
+      "learning_rate": 6.166397771942987e-05,
+      "loss": 0.09396827220916748,
+      "mean_token_accuracy": 0.967512384802103,
+      "num_tokens": 7321163.0,
+      "step": 3160
+    },
+    {
+      "epoch": 7.615570307785154,
+      "eval_entropy": 0.26364776575833226,
+      "eval_loss": 0.9204733967781067,
+      "eval_mean_token_accuracy": 0.827419729714983,
+      "eval_num_tokens": 7321163.0,
+      "eval_runtime": 88.3441,
+      "eval_samples_per_second": 16.074,
+      "eval_steps_per_second": 2.015,
+      "step": 3160
+    },
+    {
+      "entropy": 0.12209134344011545,
+      "epoch": 7.663850331925166,
+      "grad_norm": 0.5159774422645569,
+      "learning_rate": 5.93388342896962e-05,
+      "loss": 0.09292978644371033,
+      "mean_token_accuracy": 0.9658421002328396,
+      "num_tokens": 7366217.0,
+      "step": 3180
+    },
+    {
+      "epoch": 7.663850331925166,
+      "eval_entropy": 0.2714110068725736,
+      "eval_loss": 0.9143092036247253,
+      "eval_mean_token_accuracy": 0.8262251099843657,
+      "eval_num_tokens": 7366217.0,
+      "eval_runtime": 87.9635,
+      "eval_samples_per_second": 16.143,
+      "eval_steps_per_second": 2.024,
+      "step": 3180
+    },
+    {
+      "entropy": 0.11955609126016498,
+      "epoch": 7.712130356065178,
+      "grad_norm": 0.7005976438522339,
+      "learning_rate": 5.705014101934252e-05,
+      "loss": 0.09319776892662049,
+      "mean_token_accuracy": 0.9674513377249241,
+      "num_tokens": 7410887.0,
+      "step": 3200
+    },
+    {
+      "epoch": 7.712130356065178,
+      "eval_entropy": 0.2626291513442993,
+      "eval_loss": 0.91884446144104,
+      "eval_mean_token_accuracy": 0.827563955207889,
+      "eval_num_tokens": 7410887.0,
+      "eval_runtime": 87.8849,
+      "eval_samples_per_second": 16.157,
+      "eval_steps_per_second": 2.025,
+      "step": 3200
+    },
+    {
+      "entropy": 0.11767532899975777,
+      "epoch": 7.76041038020519,
+      "grad_norm": 0.4675362706184387,
+      "learning_rate": 5.479854558149187e-05,
+      "loss": 0.09038087725639343,
+      "mean_token_accuracy": 0.9669680051505566,
+      "num_tokens": 7457578.0,
+      "step": 3220
+    },
+    {
+      "epoch": 7.76041038020519,
+      "eval_entropy": 0.26581037203582486,
+      "eval_loss": 0.9146883487701416,
+      "eval_mean_token_accuracy": 0.8270345812433222,
+      "eval_num_tokens": 7457578.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 3220
+    },
+    {
+      "entropy": 0.11383061073720455,
+      "epoch": 7.808690404345202,
+      "grad_norm": 0.541247546672821,
+      "learning_rate": 5.258468515101918e-05,
+      "loss": 0.08942405581474304,
+      "mean_token_accuracy": 0.968398705124855,
+      "num_tokens": 7505761.0,
+      "step": 3240
+    },
+    {
+      "epoch": 7.808690404345202,
+      "eval_entropy": 0.26687998960861997,
+      "eval_loss": 0.9037085175514221,
+      "eval_mean_token_accuracy": 0.8274977863504646,
+      "eval_num_tokens": 7505761.0,
+      "eval_runtime": 88.5163,
+      "eval_samples_per_second": 16.042,
+      "eval_steps_per_second": 2.011,
+      "step": 3240
+    },
+    {
+      "entropy": 0.12004058873280883,
+      "epoch": 7.856970428485214,
+      "grad_norm": 0.5289552211761475,
+      "learning_rate": 5.040918622423899e-05,
+      "loss": 0.09064805507659912,
+      "mean_token_accuracy": 0.9673330299556255,
+      "num_tokens": 7552557.0,
+      "step": 3260
+    },
+    {
+      "epoch": 7.856970428485214,
+      "eval_entropy": 0.26288379251622085,
+      "eval_loss": 0.9146178364753723,
+      "eval_mean_token_accuracy": 0.8276662448149049,
+      "eval_num_tokens": 7552557.0,
+      "eval_runtime": 88.0723,
+      "eval_samples_per_second": 16.123,
+      "eval_steps_per_second": 2.021,
+      "step": 3260
+    },
+    {
+      "entropy": 0.1180689456872642,
+      "epoch": 7.905250452625227,
+      "grad_norm": 0.4754861891269684,
+      "learning_rate": 4.827266444161377e-05,
+      "loss": 0.09402655363082886,
+      "mean_token_accuracy": 0.9663361787796021,
+      "num_tokens": 7596998.0,
+      "step": 3280
+    },
+    {
+      "epoch": 7.905250452625227,
+      "eval_entropy": 0.261690996922134,
+      "eval_loss": 0.9120186567306519,
+      "eval_mean_token_accuracy": 0.8278995015647974,
+      "eval_num_tokens": 7596998.0,
+      "eval_runtime": 88.6573,
+      "eval_samples_per_second": 16.017,
+      "eval_steps_per_second": 2.008,
+      "step": 3280
+    },
+    {
+      "entropy": 0.1162164019420743,
+      "epoch": 7.953530476765239,
+      "grad_norm": 0.6203398108482361,
+      "learning_rate": 4.61757244135355e-05,
+      "loss": 0.08832768201828003,
+      "mean_token_accuracy": 0.9694220684468746,
+      "num_tokens": 7643288.0,
+      "step": 3300
+    },
+    {
+      "epoch": 7.953530476765239,
+      "eval_entropy": 0.26158678129817664,
+      "eval_loss": 0.9301361441612244,
+      "eval_mean_token_accuracy": 0.8269603885291668,
+      "eval_num_tokens": 7643288.0,
+      "eval_runtime": 88.7064,
+      "eval_samples_per_second": 16.008,
+      "eval_steps_per_second": 2.007,
+      "step": 3300
+    },
+    {
+      "entropy": 0.10877202204489089,
+      "epoch": 8.0,
+      "grad_norm": 3.058673143386841,
+      "learning_rate": 4.411895954922763e-05,
+      "loss": 0.09159615635871887,
+      "mean_token_accuracy": 0.9686878834451947,
+      "num_tokens": 7688608.0,
+      "step": 3320
+    },
+    {
+      "epoch": 8.0,
+      "eval_entropy": 0.2590438057197614,
+      "eval_loss": 0.9262065887451172,
+      "eval_mean_token_accuracy": 0.8277404546067956,
+      "eval_num_tokens": 7688608.0,
+      "eval_runtime": 87.7557,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 3320
+    },
+    {
+      "entropy": 0.09969092034734786,
+      "epoch": 8.048280024140013,
+      "grad_norm": 0.2561911642551422,
+      "learning_rate": 4.2102951888817644e-05,
+      "loss": 0.07228735685348511,
+      "mean_token_accuracy": 0.9743556551635265,
+      "num_tokens": 7732856.0,
+      "step": 3340
+    },
+    {
+      "epoch": 8.048280024140013,
+      "eval_entropy": 0.2462534454933713,
+      "eval_loss": 0.9755983352661133,
+      "eval_mean_token_accuracy": 0.827510209565752,
+      "eval_num_tokens": 7732856.0,
+      "eval_runtime": 88.0297,
+      "eval_samples_per_second": 16.131,
+      "eval_steps_per_second": 2.022,
+      "step": 3340
+    },
+    {
+      "entropy": 0.09477624730207027,
+      "epoch": 8.096560048280024,
+      "grad_norm": 0.3670356571674347,
+      "learning_rate": 4.012827193862633e-05,
+      "loss": 0.0686036467552185,
+      "mean_token_accuracy": 0.9743670664727688,
+      "num_tokens": 7778585.0,
+      "step": 3360
+    },
+    {
+      "epoch": 8.096560048280024,
+      "eval_entropy": 0.2440882951356052,
+      "eval_loss": 1.0118423700332642,
+      "eval_mean_token_accuracy": 0.8271241070849172,
+      "eval_num_tokens": 7778585.0,
+      "eval_runtime": 88.1364,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 3360
+    },
+    {
+      "entropy": 0.09417980071157217,
+      "epoch": 8.144840072420036,
+      "grad_norm": 0.5785111784934998,
+      "learning_rate": 3.819547850972133e-05,
+      "loss": 0.07100698947906495,
+      "mean_token_accuracy": 0.9737804099917412,
+      "num_tokens": 7824287.0,
+      "step": 3380
+    },
+    {
+      "epoch": 8.144840072420036,
+      "eval_entropy": 0.24603172860453637,
+      "eval_loss": 1.0021991729736328,
+      "eval_mean_token_accuracy": 0.827162122793412,
+      "eval_num_tokens": 7824287.0,
+      "eval_runtime": 88.4534,
+      "eval_samples_per_second": 16.054,
+      "eval_steps_per_second": 2.012,
+      "step": 3380
+    },
+    {
+      "entropy": 0.087222247896716,
+      "epoch": 8.193120096560047,
+      "grad_norm": 0.44848504662513733,
+      "learning_rate": 3.6305118559780393e-05,
+      "loss": 0.06325234174728393,
+      "mean_token_accuracy": 0.9769335232675076,
+      "num_tokens": 7874784.0,
+      "step": 3400
+    },
+    {
+      "epoch": 8.193120096560047,
+      "eval_entropy": 0.24544863905129807,
+      "eval_loss": 1.005008339881897,
+      "eval_mean_token_accuracy": 0.8273035240976998,
+      "eval_num_tokens": 7874784.0,
+      "eval_runtime": 88.6737,
+      "eval_samples_per_second": 16.014,
+      "eval_steps_per_second": 2.007,
+      "step": 3400
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.3285974584426906e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3420/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3420/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3420/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3420/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3420/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3420/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3420/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3420/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3420/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3420/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3420/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3420/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3420/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3420/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..2db801ca833cb1192fbe01b8cfb97880c3c13676
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3420/trainer_state.json
@@ -0,0 +1,3625 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 8.24140012070006,
+  "eval_steps": 20,
+  "global_step": 3420,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    },
+    {
+      "entropy": 0.1326121488586068,
+      "epoch": 7.181050090525045,
+      "grad_norm": 0.4843117296695709,
+      "learning_rate": 8.411870386507724e-05,
+      "loss": 0.09537842869758606,
+      "mean_token_accuracy": 0.9645151920616627,
+      "num_tokens": 6898444.0,
+      "step": 2980
+    },
+    {
+      "epoch": 7.181050090525045,
+      "eval_entropy": 0.27999477783280813,
+      "eval_loss": 0.9031037092208862,
+      "eval_mean_token_accuracy": 0.8261882204018282,
+      "eval_num_tokens": 6898444.0,
+      "eval_runtime": 88.5602,
+      "eval_samples_per_second": 16.034,
+      "eval_steps_per_second": 2.01,
+      "step": 2980
+    },
+    {
+      "entropy": 0.1156751093454659,
+      "epoch": 7.229330114665057,
+      "grad_norm": 0.5297922492027283,
+      "learning_rate": 8.149629351430326e-05,
+      "loss": 0.09203402996063233,
+      "mean_token_accuracy": 0.9689733698964119,
+      "num_tokens": 6947106.0,
+      "step": 3000
+    },
+    {
+      "epoch": 7.229330114665057,
+      "eval_entropy": 0.2740225845508361,
+      "eval_loss": 0.8876440525054932,
+      "eval_mean_token_accuracy": 0.8265116857678703,
+      "eval_num_tokens": 6947106.0,
+      "eval_runtime": 87.7971,
+      "eval_samples_per_second": 16.174,
+      "eval_steps_per_second": 2.027,
+      "step": 3000
+    },
+    {
+      "entropy": 0.12116114404052496,
+      "epoch": 7.27761013880507,
+      "grad_norm": 0.5467398166656494,
+      "learning_rate": 7.890406302404618e-05,
+      "loss": 0.0933056652545929,
+      "mean_token_accuracy": 0.9676455594599247,
+      "num_tokens": 6992784.0,
+      "step": 3020
+    },
+    {
+      "epoch": 7.27761013880507,
+      "eval_entropy": 0.27238342793804876,
+      "eval_loss": 0.9166117310523987,
+      "eval_mean_token_accuracy": 0.825499015242866,
+      "eval_num_tokens": 6992784.0,
+      "eval_runtime": 88.3276,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 3020
+    },
+    {
+      "entropy": 0.11715119928121567,
+      "epoch": 7.325890162945082,
+      "grad_norm": 0.4976702332496643,
+      "learning_rate": 7.634274596487176e-05,
+      "loss": 0.09449659585952759,
+      "mean_token_accuracy": 0.966506478190422,
+      "num_tokens": 7040482.0,
+      "step": 3040
+    },
+    {
+      "epoch": 7.325890162945082,
+      "eval_entropy": 0.2788975556077582,
+      "eval_loss": 0.8833065629005432,
+      "eval_mean_token_accuracy": 0.8269920596915684,
+      "eval_num_tokens": 7040482.0,
+      "eval_runtime": 88.9016,
+      "eval_samples_per_second": 15.973,
+      "eval_steps_per_second": 2.002,
+      "step": 3040
+    },
+    {
+      "entropy": 0.12565911412239075,
+      "epoch": 7.3741701870850935,
+      "grad_norm": 0.5728679895401001,
+      "learning_rate": 7.381306715921025e-05,
+      "loss": 0.09626244902610778,
+      "mean_token_accuracy": 0.9661444149911403,
+      "num_tokens": 7086261.0,
+      "step": 3060
+    },
+    {
+      "epoch": 7.3741701870850935,
+      "eval_entropy": 0.2762492011939542,
+      "eval_loss": 0.8961160778999329,
+      "eval_mean_token_accuracy": 0.826369910092836,
+      "eval_num_tokens": 7086261.0,
+      "eval_runtime": 88.5204,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 3060
+    },
+    {
+      "entropy": 0.12258567921817302,
+      "epoch": 7.422450211225105,
+      "grad_norm": 0.468985915184021,
+      "learning_rate": 7.131574247624031e-05,
+      "loss": 0.09120266437530518,
+      "mean_token_accuracy": 0.9674938447773457,
+      "num_tokens": 7133247.0,
+      "step": 3080
+    },
+    {
+      "epoch": 7.422450211225105,
+      "eval_entropy": 0.2688512086366,
+      "eval_loss": 0.9048654437065125,
+      "eval_mean_token_accuracy": 0.8265582302982888,
+      "eval_num_tokens": 7133247.0,
+      "eval_runtime": 88.164,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 3080
+    },
+    {
+      "entropy": 0.12070536343380808,
+      "epoch": 7.470730235365117,
+      "grad_norm": 0.5310841798782349,
+      "learning_rate": 6.885147862930688e-05,
+      "loss": 0.09386247992515565,
+      "mean_token_accuracy": 0.96664222702384,
+      "num_tokens": 7177810.0,
+      "step": 3100
+    },
+    {
+      "epoch": 7.470730235365117,
+      "eval_entropy": 0.26289632874593305,
+      "eval_loss": 0.902837336063385,
+      "eval_mean_token_accuracy": 0.828052184554968,
+      "eval_num_tokens": 7177810.0,
+      "eval_runtime": 88.1201,
+      "eval_samples_per_second": 16.114,
+      "eval_steps_per_second": 2.02,
+      "step": 3100
+    },
+    {
+      "entropy": 0.11322665181942285,
+      "epoch": 7.51901025950513,
+      "grad_norm": 0.6846067905426025,
+      "learning_rate": 6.642097297592908e-05,
+      "loss": 0.08772538304328918,
+      "mean_token_accuracy": 0.9691362135112286,
+      "num_tokens": 7225302.0,
+      "step": 3120
+    },
+    {
+      "epoch": 7.51901025950513,
+      "eval_entropy": 0.2670605613776807,
+      "eval_loss": 0.9130566120147705,
+      "eval_mean_token_accuracy": 0.826366903741708,
+      "eval_num_tokens": 7225302.0,
+      "eval_runtime": 88.6035,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3120
+    },
+    {
+      "entropy": 0.10985926580615342,
+      "epoch": 7.567290283645142,
+      "grad_norm": 0.4799748361110687,
+      "learning_rate": 6.402491332045683e-05,
+      "loss": 0.08672296404838561,
+      "mean_token_accuracy": 0.9698824048042297,
+      "num_tokens": 7274881.0,
+      "step": 3140
+    },
+    {
+      "epoch": 7.567290283645142,
+      "eval_entropy": 0.26657010866015146,
+      "eval_loss": 0.9056380987167358,
+      "eval_mean_token_accuracy": 0.8272326461384806,
+      "eval_num_tokens": 7274881.0,
+      "eval_runtime": 88.2967,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3140
+    },
+    {
+      "entropy": 0.11890473761595785,
+      "epoch": 7.615570307785154,
+      "grad_norm": 0.4081912934780121,
+      "learning_rate": 6.166397771942987e-05,
+      "loss": 0.09396827220916748,
+      "mean_token_accuracy": 0.967512384802103,
+      "num_tokens": 7321163.0,
+      "step": 3160
+    },
+    {
+      "epoch": 7.615570307785154,
+      "eval_entropy": 0.26364776575833226,
+      "eval_loss": 0.9204733967781067,
+      "eval_mean_token_accuracy": 0.827419729714983,
+      "eval_num_tokens": 7321163.0,
+      "eval_runtime": 88.3441,
+      "eval_samples_per_second": 16.074,
+      "eval_steps_per_second": 2.015,
+      "step": 3160
+    },
+    {
+      "entropy": 0.12209134344011545,
+      "epoch": 7.663850331925166,
+      "grad_norm": 0.5159774422645569,
+      "learning_rate": 5.93388342896962e-05,
+      "loss": 0.09292978644371033,
+      "mean_token_accuracy": 0.9658421002328396,
+      "num_tokens": 7366217.0,
+      "step": 3180
+    },
+    {
+      "epoch": 7.663850331925166,
+      "eval_entropy": 0.2714110068725736,
+      "eval_loss": 0.9143092036247253,
+      "eval_mean_token_accuracy": 0.8262251099843657,
+      "eval_num_tokens": 7366217.0,
+      "eval_runtime": 87.9635,
+      "eval_samples_per_second": 16.143,
+      "eval_steps_per_second": 2.024,
+      "step": 3180
+    },
+    {
+      "entropy": 0.11955609126016498,
+      "epoch": 7.712130356065178,
+      "grad_norm": 0.7005976438522339,
+      "learning_rate": 5.705014101934252e-05,
+      "loss": 0.09319776892662049,
+      "mean_token_accuracy": 0.9674513377249241,
+      "num_tokens": 7410887.0,
+      "step": 3200
+    },
+    {
+      "epoch": 7.712130356065178,
+      "eval_entropy": 0.2626291513442993,
+      "eval_loss": 0.91884446144104,
+      "eval_mean_token_accuracy": 0.827563955207889,
+      "eval_num_tokens": 7410887.0,
+      "eval_runtime": 87.8849,
+      "eval_samples_per_second": 16.157,
+      "eval_steps_per_second": 2.025,
+      "step": 3200
+    },
+    {
+      "entropy": 0.11767532899975777,
+      "epoch": 7.76041038020519,
+      "grad_norm": 0.4675362706184387,
+      "learning_rate": 5.479854558149187e-05,
+      "loss": 0.09038087725639343,
+      "mean_token_accuracy": 0.9669680051505566,
+      "num_tokens": 7457578.0,
+      "step": 3220
+    },
+    {
+      "epoch": 7.76041038020519,
+      "eval_entropy": 0.26581037203582486,
+      "eval_loss": 0.9146883487701416,
+      "eval_mean_token_accuracy": 0.8270345812433222,
+      "eval_num_tokens": 7457578.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 3220
+    },
+    {
+      "entropy": 0.11383061073720455,
+      "epoch": 7.808690404345202,
+      "grad_norm": 0.541247546672821,
+      "learning_rate": 5.258468515101918e-05,
+      "loss": 0.08942405581474304,
+      "mean_token_accuracy": 0.968398705124855,
+      "num_tokens": 7505761.0,
+      "step": 3240
+    },
+    {
+      "epoch": 7.808690404345202,
+      "eval_entropy": 0.26687998960861997,
+      "eval_loss": 0.9037085175514221,
+      "eval_mean_token_accuracy": 0.8274977863504646,
+      "eval_num_tokens": 7505761.0,
+      "eval_runtime": 88.5163,
+      "eval_samples_per_second": 16.042,
+      "eval_steps_per_second": 2.011,
+      "step": 3240
+    },
+    {
+      "entropy": 0.12004058873280883,
+      "epoch": 7.856970428485214,
+      "grad_norm": 0.5289552211761475,
+      "learning_rate": 5.040918622423899e-05,
+      "loss": 0.09064805507659912,
+      "mean_token_accuracy": 0.9673330299556255,
+      "num_tokens": 7552557.0,
+      "step": 3260
+    },
+    {
+      "epoch": 7.856970428485214,
+      "eval_entropy": 0.26288379251622085,
+      "eval_loss": 0.9146178364753723,
+      "eval_mean_token_accuracy": 0.8276662448149049,
+      "eval_num_tokens": 7552557.0,
+      "eval_runtime": 88.0723,
+      "eval_samples_per_second": 16.123,
+      "eval_steps_per_second": 2.021,
+      "step": 3260
+    },
+    {
+      "entropy": 0.1180689456872642,
+      "epoch": 7.905250452625227,
+      "grad_norm": 0.4754861891269684,
+      "learning_rate": 4.827266444161377e-05,
+      "loss": 0.09402655363082886,
+      "mean_token_accuracy": 0.9663361787796021,
+      "num_tokens": 7596998.0,
+      "step": 3280
+    },
+    {
+      "epoch": 7.905250452625227,
+      "eval_entropy": 0.261690996922134,
+      "eval_loss": 0.9120186567306519,
+      "eval_mean_token_accuracy": 0.8278995015647974,
+      "eval_num_tokens": 7596998.0,
+      "eval_runtime": 88.6573,
+      "eval_samples_per_second": 16.017,
+      "eval_steps_per_second": 2.008,
+      "step": 3280
+    },
+    {
+      "entropy": 0.1162164019420743,
+      "epoch": 7.953530476765239,
+      "grad_norm": 0.6203398108482361,
+      "learning_rate": 4.61757244135355e-05,
+      "loss": 0.08832768201828003,
+      "mean_token_accuracy": 0.9694220684468746,
+      "num_tokens": 7643288.0,
+      "step": 3300
+    },
+    {
+      "epoch": 7.953530476765239,
+      "eval_entropy": 0.26158678129817664,
+      "eval_loss": 0.9301361441612244,
+      "eval_mean_token_accuracy": 0.8269603885291668,
+      "eval_num_tokens": 7643288.0,
+      "eval_runtime": 88.7064,
+      "eval_samples_per_second": 16.008,
+      "eval_steps_per_second": 2.007,
+      "step": 3300
+    },
+    {
+      "entropy": 0.10877202204489089,
+      "epoch": 8.0,
+      "grad_norm": 3.058673143386841,
+      "learning_rate": 4.411895954922763e-05,
+      "loss": 0.09159615635871887,
+      "mean_token_accuracy": 0.9686878834451947,
+      "num_tokens": 7688608.0,
+      "step": 3320
+    },
+    {
+      "epoch": 8.0,
+      "eval_entropy": 0.2590438057197614,
+      "eval_loss": 0.9262065887451172,
+      "eval_mean_token_accuracy": 0.8277404546067956,
+      "eval_num_tokens": 7688608.0,
+      "eval_runtime": 87.7557,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 3320
+    },
+    {
+      "entropy": 0.09969092034734786,
+      "epoch": 8.048280024140013,
+      "grad_norm": 0.2561911642551422,
+      "learning_rate": 4.2102951888817644e-05,
+      "loss": 0.07228735685348511,
+      "mean_token_accuracy": 0.9743556551635265,
+      "num_tokens": 7732856.0,
+      "step": 3340
+    },
+    {
+      "epoch": 8.048280024140013,
+      "eval_entropy": 0.2462534454933713,
+      "eval_loss": 0.9755983352661133,
+      "eval_mean_token_accuracy": 0.827510209565752,
+      "eval_num_tokens": 7732856.0,
+      "eval_runtime": 88.0297,
+      "eval_samples_per_second": 16.131,
+      "eval_steps_per_second": 2.022,
+      "step": 3340
+    },
+    {
+      "entropy": 0.09477624730207027,
+      "epoch": 8.096560048280024,
+      "grad_norm": 0.3670356571674347,
+      "learning_rate": 4.012827193862633e-05,
+      "loss": 0.0686036467552185,
+      "mean_token_accuracy": 0.9743670664727688,
+      "num_tokens": 7778585.0,
+      "step": 3360
+    },
+    {
+      "epoch": 8.096560048280024,
+      "eval_entropy": 0.2440882951356052,
+      "eval_loss": 1.0118423700332642,
+      "eval_mean_token_accuracy": 0.8271241070849172,
+      "eval_num_tokens": 7778585.0,
+      "eval_runtime": 88.1364,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 3360
+    },
+    {
+      "entropy": 0.09417980071157217,
+      "epoch": 8.144840072420036,
+      "grad_norm": 0.5785111784934998,
+      "learning_rate": 3.819547850972133e-05,
+      "loss": 0.07100698947906495,
+      "mean_token_accuracy": 0.9737804099917412,
+      "num_tokens": 7824287.0,
+      "step": 3380
+    },
+    {
+      "epoch": 8.144840072420036,
+      "eval_entropy": 0.24603172860453637,
+      "eval_loss": 1.0021991729736328,
+      "eval_mean_token_accuracy": 0.827162122793412,
+      "eval_num_tokens": 7824287.0,
+      "eval_runtime": 88.4534,
+      "eval_samples_per_second": 16.054,
+      "eval_steps_per_second": 2.012,
+      "step": 3380
+    },
+    {
+      "entropy": 0.087222247896716,
+      "epoch": 8.193120096560047,
+      "grad_norm": 0.44848504662513733,
+      "learning_rate": 3.6305118559780393e-05,
+      "loss": 0.06325234174728393,
+      "mean_token_accuracy": 0.9769335232675076,
+      "num_tokens": 7874784.0,
+      "step": 3400
+    },
+    {
+      "epoch": 8.193120096560047,
+      "eval_entropy": 0.24544863905129807,
+      "eval_loss": 1.005008339881897,
+      "eval_mean_token_accuracy": 0.8273035240976998,
+      "eval_num_tokens": 7874784.0,
+      "eval_runtime": 88.6737,
+      "eval_samples_per_second": 16.014,
+      "eval_steps_per_second": 2.007,
+      "step": 3400
+    },
+    {
+      "entropy": 0.09037275062873959,
+      "epoch": 8.24140012070006,
+      "grad_norm": 0.23529155552387238,
+      "learning_rate": 3.4457727038308435e-05,
+      "loss": 0.06589569449424744,
+      "mean_token_accuracy": 0.9751470290124417,
+      "num_tokens": 7925240.0,
+      "step": 3420
+    },
+    {
+      "epoch": 8.24140012070006,
+      "eval_entropy": 0.24203423494368456,
+      "eval_loss": 1.0052824020385742,
+      "eval_mean_token_accuracy": 0.8275575503874361,
+      "eval_num_tokens": 7925240.0,
+      "eval_runtime": 88.3486,
+      "eval_samples_per_second": 16.073,
+      "eval_steps_per_second": 2.015,
+      "step": 3420
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.349095793743667e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3440/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3440/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3440/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3440/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3440/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3440/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3440/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3440/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3440/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3440/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3440/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3440/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3440/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3440/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..03ca086c00f1238eba904e496343ceb40e61967e
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3440/trainer_state.json
@@ -0,0 +1,3646 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 8.289680144840073,
+  "eval_steps": 20,
+  "global_step": 3440,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    },
+    {
+      "entropy": 0.1326121488586068,
+      "epoch": 7.181050090525045,
+      "grad_norm": 0.4843117296695709,
+      "learning_rate": 8.411870386507724e-05,
+      "loss": 0.09537842869758606,
+      "mean_token_accuracy": 0.9645151920616627,
+      "num_tokens": 6898444.0,
+      "step": 2980
+    },
+    {
+      "epoch": 7.181050090525045,
+      "eval_entropy": 0.27999477783280813,
+      "eval_loss": 0.9031037092208862,
+      "eval_mean_token_accuracy": 0.8261882204018282,
+      "eval_num_tokens": 6898444.0,
+      "eval_runtime": 88.5602,
+      "eval_samples_per_second": 16.034,
+      "eval_steps_per_second": 2.01,
+      "step": 2980
+    },
+    {
+      "entropy": 0.1156751093454659,
+      "epoch": 7.229330114665057,
+      "grad_norm": 0.5297922492027283,
+      "learning_rate": 8.149629351430326e-05,
+      "loss": 0.09203402996063233,
+      "mean_token_accuracy": 0.9689733698964119,
+      "num_tokens": 6947106.0,
+      "step": 3000
+    },
+    {
+      "epoch": 7.229330114665057,
+      "eval_entropy": 0.2740225845508361,
+      "eval_loss": 0.8876440525054932,
+      "eval_mean_token_accuracy": 0.8265116857678703,
+      "eval_num_tokens": 6947106.0,
+      "eval_runtime": 87.7971,
+      "eval_samples_per_second": 16.174,
+      "eval_steps_per_second": 2.027,
+      "step": 3000
+    },
+    {
+      "entropy": 0.12116114404052496,
+      "epoch": 7.27761013880507,
+      "grad_norm": 0.5467398166656494,
+      "learning_rate": 7.890406302404618e-05,
+      "loss": 0.0933056652545929,
+      "mean_token_accuracy": 0.9676455594599247,
+      "num_tokens": 6992784.0,
+      "step": 3020
+    },
+    {
+      "epoch": 7.27761013880507,
+      "eval_entropy": 0.27238342793804876,
+      "eval_loss": 0.9166117310523987,
+      "eval_mean_token_accuracy": 0.825499015242866,
+      "eval_num_tokens": 6992784.0,
+      "eval_runtime": 88.3276,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 3020
+    },
+    {
+      "entropy": 0.11715119928121567,
+      "epoch": 7.325890162945082,
+      "grad_norm": 0.4976702332496643,
+      "learning_rate": 7.634274596487176e-05,
+      "loss": 0.09449659585952759,
+      "mean_token_accuracy": 0.966506478190422,
+      "num_tokens": 7040482.0,
+      "step": 3040
+    },
+    {
+      "epoch": 7.325890162945082,
+      "eval_entropy": 0.2788975556077582,
+      "eval_loss": 0.8833065629005432,
+      "eval_mean_token_accuracy": 0.8269920596915684,
+      "eval_num_tokens": 7040482.0,
+      "eval_runtime": 88.9016,
+      "eval_samples_per_second": 15.973,
+      "eval_steps_per_second": 2.002,
+      "step": 3040
+    },
+    {
+      "entropy": 0.12565911412239075,
+      "epoch": 7.3741701870850935,
+      "grad_norm": 0.5728679895401001,
+      "learning_rate": 7.381306715921025e-05,
+      "loss": 0.09626244902610778,
+      "mean_token_accuracy": 0.9661444149911403,
+      "num_tokens": 7086261.0,
+      "step": 3060
+    },
+    {
+      "epoch": 7.3741701870850935,
+      "eval_entropy": 0.2762492011939542,
+      "eval_loss": 0.8961160778999329,
+      "eval_mean_token_accuracy": 0.826369910092836,
+      "eval_num_tokens": 7086261.0,
+      "eval_runtime": 88.5204,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 3060
+    },
+    {
+      "entropy": 0.12258567921817302,
+      "epoch": 7.422450211225105,
+      "grad_norm": 0.468985915184021,
+      "learning_rate": 7.131574247624031e-05,
+      "loss": 0.09120266437530518,
+      "mean_token_accuracy": 0.9674938447773457,
+      "num_tokens": 7133247.0,
+      "step": 3080
+    },
+    {
+      "epoch": 7.422450211225105,
+      "eval_entropy": 0.2688512086366,
+      "eval_loss": 0.9048654437065125,
+      "eval_mean_token_accuracy": 0.8265582302982888,
+      "eval_num_tokens": 7133247.0,
+      "eval_runtime": 88.164,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 3080
+    },
+    {
+      "entropy": 0.12070536343380808,
+      "epoch": 7.470730235365117,
+      "grad_norm": 0.5310841798782349,
+      "learning_rate": 6.885147862930688e-05,
+      "loss": 0.09386247992515565,
+      "mean_token_accuracy": 0.96664222702384,
+      "num_tokens": 7177810.0,
+      "step": 3100
+    },
+    {
+      "epoch": 7.470730235365117,
+      "eval_entropy": 0.26289632874593305,
+      "eval_loss": 0.902837336063385,
+      "eval_mean_token_accuracy": 0.828052184554968,
+      "eval_num_tokens": 7177810.0,
+      "eval_runtime": 88.1201,
+      "eval_samples_per_second": 16.114,
+      "eval_steps_per_second": 2.02,
+      "step": 3100
+    },
+    {
+      "entropy": 0.11322665181942285,
+      "epoch": 7.51901025950513,
+      "grad_norm": 0.6846067905426025,
+      "learning_rate": 6.642097297592908e-05,
+      "loss": 0.08772538304328918,
+      "mean_token_accuracy": 0.9691362135112286,
+      "num_tokens": 7225302.0,
+      "step": 3120
+    },
+    {
+      "epoch": 7.51901025950513,
+      "eval_entropy": 0.2670605613776807,
+      "eval_loss": 0.9130566120147705,
+      "eval_mean_token_accuracy": 0.826366903741708,
+      "eval_num_tokens": 7225302.0,
+      "eval_runtime": 88.6035,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3120
+    },
+    {
+      "entropy": 0.10985926580615342,
+      "epoch": 7.567290283645142,
+      "grad_norm": 0.4799748361110687,
+      "learning_rate": 6.402491332045683e-05,
+      "loss": 0.08672296404838561,
+      "mean_token_accuracy": 0.9698824048042297,
+      "num_tokens": 7274881.0,
+      "step": 3140
+    },
+    {
+      "epoch": 7.567290283645142,
+      "eval_entropy": 0.26657010866015146,
+      "eval_loss": 0.9056380987167358,
+      "eval_mean_token_accuracy": 0.8272326461384806,
+      "eval_num_tokens": 7274881.0,
+      "eval_runtime": 88.2967,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3140
+    },
+    {
+      "entropy": 0.11890473761595785,
+      "epoch": 7.615570307785154,
+      "grad_norm": 0.4081912934780121,
+      "learning_rate": 6.166397771942987e-05,
+      "loss": 0.09396827220916748,
+      "mean_token_accuracy": 0.967512384802103,
+      "num_tokens": 7321163.0,
+      "step": 3160
+    },
+    {
+      "epoch": 7.615570307785154,
+      "eval_entropy": 0.26364776575833226,
+      "eval_loss": 0.9204733967781067,
+      "eval_mean_token_accuracy": 0.827419729714983,
+      "eval_num_tokens": 7321163.0,
+      "eval_runtime": 88.3441,
+      "eval_samples_per_second": 16.074,
+      "eval_steps_per_second": 2.015,
+      "step": 3160
+    },
+    {
+      "entropy": 0.12209134344011545,
+      "epoch": 7.663850331925166,
+      "grad_norm": 0.5159774422645569,
+      "learning_rate": 5.93388342896962e-05,
+      "loss": 0.09292978644371033,
+      "mean_token_accuracy": 0.9658421002328396,
+      "num_tokens": 7366217.0,
+      "step": 3180
+    },
+    {
+      "epoch": 7.663850331925166,
+      "eval_entropy": 0.2714110068725736,
+      "eval_loss": 0.9143092036247253,
+      "eval_mean_token_accuracy": 0.8262251099843657,
+      "eval_num_tokens": 7366217.0,
+      "eval_runtime": 87.9635,
+      "eval_samples_per_second": 16.143,
+      "eval_steps_per_second": 2.024,
+      "step": 3180
+    },
+    {
+      "entropy": 0.11955609126016498,
+      "epoch": 7.712130356065178,
+      "grad_norm": 0.7005976438522339,
+      "learning_rate": 5.705014101934252e-05,
+      "loss": 0.09319776892662049,
+      "mean_token_accuracy": 0.9674513377249241,
+      "num_tokens": 7410887.0,
+      "step": 3200
+    },
+    {
+      "epoch": 7.712130356065178,
+      "eval_entropy": 0.2626291513442993,
+      "eval_loss": 0.91884446144104,
+      "eval_mean_token_accuracy": 0.827563955207889,
+      "eval_num_tokens": 7410887.0,
+      "eval_runtime": 87.8849,
+      "eval_samples_per_second": 16.157,
+      "eval_steps_per_second": 2.025,
+      "step": 3200
+    },
+    {
+      "entropy": 0.11767532899975777,
+      "epoch": 7.76041038020519,
+      "grad_norm": 0.4675362706184387,
+      "learning_rate": 5.479854558149187e-05,
+      "loss": 0.09038087725639343,
+      "mean_token_accuracy": 0.9669680051505566,
+      "num_tokens": 7457578.0,
+      "step": 3220
+    },
+    {
+      "epoch": 7.76041038020519,
+      "eval_entropy": 0.26581037203582486,
+      "eval_loss": 0.9146883487701416,
+      "eval_mean_token_accuracy": 0.8270345812433222,
+      "eval_num_tokens": 7457578.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 3220
+    },
+    {
+      "entropy": 0.11383061073720455,
+      "epoch": 7.808690404345202,
+      "grad_norm": 0.541247546672821,
+      "learning_rate": 5.258468515101918e-05,
+      "loss": 0.08942405581474304,
+      "mean_token_accuracy": 0.968398705124855,
+      "num_tokens": 7505761.0,
+      "step": 3240
+    },
+    {
+      "epoch": 7.808690404345202,
+      "eval_entropy": 0.26687998960861997,
+      "eval_loss": 0.9037085175514221,
+      "eval_mean_token_accuracy": 0.8274977863504646,
+      "eval_num_tokens": 7505761.0,
+      "eval_runtime": 88.5163,
+      "eval_samples_per_second": 16.042,
+      "eval_steps_per_second": 2.011,
+      "step": 3240
+    },
+    {
+      "entropy": 0.12004058873280883,
+      "epoch": 7.856970428485214,
+      "grad_norm": 0.5289552211761475,
+      "learning_rate": 5.040918622423899e-05,
+      "loss": 0.09064805507659912,
+      "mean_token_accuracy": 0.9673330299556255,
+      "num_tokens": 7552557.0,
+      "step": 3260
+    },
+    {
+      "epoch": 7.856970428485214,
+      "eval_entropy": 0.26288379251622085,
+      "eval_loss": 0.9146178364753723,
+      "eval_mean_token_accuracy": 0.8276662448149049,
+      "eval_num_tokens": 7552557.0,
+      "eval_runtime": 88.0723,
+      "eval_samples_per_second": 16.123,
+      "eval_steps_per_second": 2.021,
+      "step": 3260
+    },
+    {
+      "entropy": 0.1180689456872642,
+      "epoch": 7.905250452625227,
+      "grad_norm": 0.4754861891269684,
+      "learning_rate": 4.827266444161377e-05,
+      "loss": 0.09402655363082886,
+      "mean_token_accuracy": 0.9663361787796021,
+      "num_tokens": 7596998.0,
+      "step": 3280
+    },
+    {
+      "epoch": 7.905250452625227,
+      "eval_entropy": 0.261690996922134,
+      "eval_loss": 0.9120186567306519,
+      "eval_mean_token_accuracy": 0.8278995015647974,
+      "eval_num_tokens": 7596998.0,
+      "eval_runtime": 88.6573,
+      "eval_samples_per_second": 16.017,
+      "eval_steps_per_second": 2.008,
+      "step": 3280
+    },
+    {
+      "entropy": 0.1162164019420743,
+      "epoch": 7.953530476765239,
+      "grad_norm": 0.6203398108482361,
+      "learning_rate": 4.61757244135355e-05,
+      "loss": 0.08832768201828003,
+      "mean_token_accuracy": 0.9694220684468746,
+      "num_tokens": 7643288.0,
+      "step": 3300
+    },
+    {
+      "epoch": 7.953530476765239,
+      "eval_entropy": 0.26158678129817664,
+      "eval_loss": 0.9301361441612244,
+      "eval_mean_token_accuracy": 0.8269603885291668,
+      "eval_num_tokens": 7643288.0,
+      "eval_runtime": 88.7064,
+      "eval_samples_per_second": 16.008,
+      "eval_steps_per_second": 2.007,
+      "step": 3300
+    },
+    {
+      "entropy": 0.10877202204489089,
+      "epoch": 8.0,
+      "grad_norm": 3.058673143386841,
+      "learning_rate": 4.411895954922763e-05,
+      "loss": 0.09159615635871887,
+      "mean_token_accuracy": 0.9686878834451947,
+      "num_tokens": 7688608.0,
+      "step": 3320
+    },
+    {
+      "epoch": 8.0,
+      "eval_entropy": 0.2590438057197614,
+      "eval_loss": 0.9262065887451172,
+      "eval_mean_token_accuracy": 0.8277404546067956,
+      "eval_num_tokens": 7688608.0,
+      "eval_runtime": 87.7557,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 3320
+    },
+    {
+      "entropy": 0.09969092034734786,
+      "epoch": 8.048280024140013,
+      "grad_norm": 0.2561911642551422,
+      "learning_rate": 4.2102951888817644e-05,
+      "loss": 0.07228735685348511,
+      "mean_token_accuracy": 0.9743556551635265,
+      "num_tokens": 7732856.0,
+      "step": 3340
+    },
+    {
+      "epoch": 8.048280024140013,
+      "eval_entropy": 0.2462534454933713,
+      "eval_loss": 0.9755983352661133,
+      "eval_mean_token_accuracy": 0.827510209565752,
+      "eval_num_tokens": 7732856.0,
+      "eval_runtime": 88.0297,
+      "eval_samples_per_second": 16.131,
+      "eval_steps_per_second": 2.022,
+      "step": 3340
+    },
+    {
+      "entropy": 0.09477624730207027,
+      "epoch": 8.096560048280024,
+      "grad_norm": 0.3670356571674347,
+      "learning_rate": 4.012827193862633e-05,
+      "loss": 0.0686036467552185,
+      "mean_token_accuracy": 0.9743670664727688,
+      "num_tokens": 7778585.0,
+      "step": 3360
+    },
+    {
+      "epoch": 8.096560048280024,
+      "eval_entropy": 0.2440882951356052,
+      "eval_loss": 1.0118423700332642,
+      "eval_mean_token_accuracy": 0.8271241070849172,
+      "eval_num_tokens": 7778585.0,
+      "eval_runtime": 88.1364,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 3360
+    },
+    {
+      "entropy": 0.09417980071157217,
+      "epoch": 8.144840072420036,
+      "grad_norm": 0.5785111784934998,
+      "learning_rate": 3.819547850972133e-05,
+      "loss": 0.07100698947906495,
+      "mean_token_accuracy": 0.9737804099917412,
+      "num_tokens": 7824287.0,
+      "step": 3380
+    },
+    {
+      "epoch": 8.144840072420036,
+      "eval_entropy": 0.24603172860453637,
+      "eval_loss": 1.0021991729736328,
+      "eval_mean_token_accuracy": 0.827162122793412,
+      "eval_num_tokens": 7824287.0,
+      "eval_runtime": 88.4534,
+      "eval_samples_per_second": 16.054,
+      "eval_steps_per_second": 2.012,
+      "step": 3380
+    },
+    {
+      "entropy": 0.087222247896716,
+      "epoch": 8.193120096560047,
+      "grad_norm": 0.44848504662513733,
+      "learning_rate": 3.6305118559780393e-05,
+      "loss": 0.06325234174728393,
+      "mean_token_accuracy": 0.9769335232675076,
+      "num_tokens": 7874784.0,
+      "step": 3400
+    },
+    {
+      "epoch": 8.193120096560047,
+      "eval_entropy": 0.24544863905129807,
+      "eval_loss": 1.005008339881897,
+      "eval_mean_token_accuracy": 0.8273035240976998,
+      "eval_num_tokens": 7874784.0,
+      "eval_runtime": 88.6737,
+      "eval_samples_per_second": 16.014,
+      "eval_steps_per_second": 2.007,
+      "step": 3400
+    },
+    {
+      "entropy": 0.09037275062873959,
+      "epoch": 8.24140012070006,
+      "grad_norm": 0.23529155552387238,
+      "learning_rate": 3.4457727038308435e-05,
+      "loss": 0.06589569449424744,
+      "mean_token_accuracy": 0.9751470290124417,
+      "num_tokens": 7925240.0,
+      "step": 3420
+    },
+    {
+      "epoch": 8.24140012070006,
+      "eval_entropy": 0.24203423494368456,
+      "eval_loss": 1.0052824020385742,
+      "eval_mean_token_accuracy": 0.8275575503874361,
+      "eval_num_tokens": 7925240.0,
+      "eval_runtime": 88.3486,
+      "eval_samples_per_second": 16.073,
+      "eval_steps_per_second": 2.015,
+      "step": 3420
+    },
+    {
+      "entropy": 0.10373208015225828,
+      "epoch": 8.289680144840073,
+      "grad_norm": 0.24796408414840698,
+      "learning_rate": 3.2653826735253594e-05,
+      "loss": 0.07714771628379821,
+      "mean_token_accuracy": 0.9722891598939896,
+      "num_tokens": 7966685.0,
+      "step": 3440
+    },
+    {
+      "epoch": 8.289680144840073,
+      "eval_entropy": 0.23461157050025597,
+      "eval_loss": 1.012108325958252,
+      "eval_mean_token_accuracy": 0.82825900564033,
+      "eval_num_tokens": 7966685.0,
+      "eval_runtime": 88.1018,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 3440
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.366551133274399e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3460/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3460/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3460/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3460/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3460/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3460/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3460/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3460/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3460/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3460/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3460/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3460/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3460/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3460/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..57cfa3bf8dd6a82e8e6e9931763bb156692d5b9d
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3460/trainer_state.json
@@ -0,0 +1,3667 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 8.337960168980084,
+  "eval_steps": 20,
+  "global_step": 3460,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    },
+    {
+      "entropy": 0.1326121488586068,
+      "epoch": 7.181050090525045,
+      "grad_norm": 0.4843117296695709,
+      "learning_rate": 8.411870386507724e-05,
+      "loss": 0.09537842869758606,
+      "mean_token_accuracy": 0.9645151920616627,
+      "num_tokens": 6898444.0,
+      "step": 2980
+    },
+    {
+      "epoch": 7.181050090525045,
+      "eval_entropy": 0.27999477783280813,
+      "eval_loss": 0.9031037092208862,
+      "eval_mean_token_accuracy": 0.8261882204018282,
+      "eval_num_tokens": 6898444.0,
+      "eval_runtime": 88.5602,
+      "eval_samples_per_second": 16.034,
+      "eval_steps_per_second": 2.01,
+      "step": 2980
+    },
+    {
+      "entropy": 0.1156751093454659,
+      "epoch": 7.229330114665057,
+      "grad_norm": 0.5297922492027283,
+      "learning_rate": 8.149629351430326e-05,
+      "loss": 0.09203402996063233,
+      "mean_token_accuracy": 0.9689733698964119,
+      "num_tokens": 6947106.0,
+      "step": 3000
+    },
+    {
+      "epoch": 7.229330114665057,
+      "eval_entropy": 0.2740225845508361,
+      "eval_loss": 0.8876440525054932,
+      "eval_mean_token_accuracy": 0.8265116857678703,
+      "eval_num_tokens": 6947106.0,
+      "eval_runtime": 87.7971,
+      "eval_samples_per_second": 16.174,
+      "eval_steps_per_second": 2.027,
+      "step": 3000
+    },
+    {
+      "entropy": 0.12116114404052496,
+      "epoch": 7.27761013880507,
+      "grad_norm": 0.5467398166656494,
+      "learning_rate": 7.890406302404618e-05,
+      "loss": 0.0933056652545929,
+      "mean_token_accuracy": 0.9676455594599247,
+      "num_tokens": 6992784.0,
+      "step": 3020
+    },
+    {
+      "epoch": 7.27761013880507,
+      "eval_entropy": 0.27238342793804876,
+      "eval_loss": 0.9166117310523987,
+      "eval_mean_token_accuracy": 0.825499015242866,
+      "eval_num_tokens": 6992784.0,
+      "eval_runtime": 88.3276,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 3020
+    },
+    {
+      "entropy": 0.11715119928121567,
+      "epoch": 7.325890162945082,
+      "grad_norm": 0.4976702332496643,
+      "learning_rate": 7.634274596487176e-05,
+      "loss": 0.09449659585952759,
+      "mean_token_accuracy": 0.966506478190422,
+      "num_tokens": 7040482.0,
+      "step": 3040
+    },
+    {
+      "epoch": 7.325890162945082,
+      "eval_entropy": 0.2788975556077582,
+      "eval_loss": 0.8833065629005432,
+      "eval_mean_token_accuracy": 0.8269920596915684,
+      "eval_num_tokens": 7040482.0,
+      "eval_runtime": 88.9016,
+      "eval_samples_per_second": 15.973,
+      "eval_steps_per_second": 2.002,
+      "step": 3040
+    },
+    {
+      "entropy": 0.12565911412239075,
+      "epoch": 7.3741701870850935,
+      "grad_norm": 0.5728679895401001,
+      "learning_rate": 7.381306715921025e-05,
+      "loss": 0.09626244902610778,
+      "mean_token_accuracy": 0.9661444149911403,
+      "num_tokens": 7086261.0,
+      "step": 3060
+    },
+    {
+      "epoch": 7.3741701870850935,
+      "eval_entropy": 0.2762492011939542,
+      "eval_loss": 0.8961160778999329,
+      "eval_mean_token_accuracy": 0.826369910092836,
+      "eval_num_tokens": 7086261.0,
+      "eval_runtime": 88.5204,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 3060
+    },
+    {
+      "entropy": 0.12258567921817302,
+      "epoch": 7.422450211225105,
+      "grad_norm": 0.468985915184021,
+      "learning_rate": 7.131574247624031e-05,
+      "loss": 0.09120266437530518,
+      "mean_token_accuracy": 0.9674938447773457,
+      "num_tokens": 7133247.0,
+      "step": 3080
+    },
+    {
+      "epoch": 7.422450211225105,
+      "eval_entropy": 0.2688512086366,
+      "eval_loss": 0.9048654437065125,
+      "eval_mean_token_accuracy": 0.8265582302982888,
+      "eval_num_tokens": 7133247.0,
+      "eval_runtime": 88.164,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 3080
+    },
+    {
+      "entropy": 0.12070536343380808,
+      "epoch": 7.470730235365117,
+      "grad_norm": 0.5310841798782349,
+      "learning_rate": 6.885147862930688e-05,
+      "loss": 0.09386247992515565,
+      "mean_token_accuracy": 0.96664222702384,
+      "num_tokens": 7177810.0,
+      "step": 3100
+    },
+    {
+      "epoch": 7.470730235365117,
+      "eval_entropy": 0.26289632874593305,
+      "eval_loss": 0.902837336063385,
+      "eval_mean_token_accuracy": 0.828052184554968,
+      "eval_num_tokens": 7177810.0,
+      "eval_runtime": 88.1201,
+      "eval_samples_per_second": 16.114,
+      "eval_steps_per_second": 2.02,
+      "step": 3100
+    },
+    {
+      "entropy": 0.11322665181942285,
+      "epoch": 7.51901025950513,
+      "grad_norm": 0.6846067905426025,
+      "learning_rate": 6.642097297592908e-05,
+      "loss": 0.08772538304328918,
+      "mean_token_accuracy": 0.9691362135112286,
+      "num_tokens": 7225302.0,
+      "step": 3120
+    },
+    {
+      "epoch": 7.51901025950513,
+      "eval_entropy": 0.2670605613776807,
+      "eval_loss": 0.9130566120147705,
+      "eval_mean_token_accuracy": 0.826366903741708,
+      "eval_num_tokens": 7225302.0,
+      "eval_runtime": 88.6035,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3120
+    },
+    {
+      "entropy": 0.10985926580615342,
+      "epoch": 7.567290283645142,
+      "grad_norm": 0.4799748361110687,
+      "learning_rate": 6.402491332045683e-05,
+      "loss": 0.08672296404838561,
+      "mean_token_accuracy": 0.9698824048042297,
+      "num_tokens": 7274881.0,
+      "step": 3140
+    },
+    {
+      "epoch": 7.567290283645142,
+      "eval_entropy": 0.26657010866015146,
+      "eval_loss": 0.9056380987167358,
+      "eval_mean_token_accuracy": 0.8272326461384806,
+      "eval_num_tokens": 7274881.0,
+      "eval_runtime": 88.2967,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3140
+    },
+    {
+      "entropy": 0.11890473761595785,
+      "epoch": 7.615570307785154,
+      "grad_norm": 0.4081912934780121,
+      "learning_rate": 6.166397771942987e-05,
+      "loss": 0.09396827220916748,
+      "mean_token_accuracy": 0.967512384802103,
+      "num_tokens": 7321163.0,
+      "step": 3160
+    },
+    {
+      "epoch": 7.615570307785154,
+      "eval_entropy": 0.26364776575833226,
+      "eval_loss": 0.9204733967781067,
+      "eval_mean_token_accuracy": 0.827419729714983,
+      "eval_num_tokens": 7321163.0,
+      "eval_runtime": 88.3441,
+      "eval_samples_per_second": 16.074,
+      "eval_steps_per_second": 2.015,
+      "step": 3160
+    },
+    {
+      "entropy": 0.12209134344011545,
+      "epoch": 7.663850331925166,
+      "grad_norm": 0.5159774422645569,
+      "learning_rate": 5.93388342896962e-05,
+      "loss": 0.09292978644371033,
+      "mean_token_accuracy": 0.9658421002328396,
+      "num_tokens": 7366217.0,
+      "step": 3180
+    },
+    {
+      "epoch": 7.663850331925166,
+      "eval_entropy": 0.2714110068725736,
+      "eval_loss": 0.9143092036247253,
+      "eval_mean_token_accuracy": 0.8262251099843657,
+      "eval_num_tokens": 7366217.0,
+      "eval_runtime": 87.9635,
+      "eval_samples_per_second": 16.143,
+      "eval_steps_per_second": 2.024,
+      "step": 3180
+    },
+    {
+      "entropy": 0.11955609126016498,
+      "epoch": 7.712130356065178,
+      "grad_norm": 0.7005976438522339,
+      "learning_rate": 5.705014101934252e-05,
+      "loss": 0.09319776892662049,
+      "mean_token_accuracy": 0.9674513377249241,
+      "num_tokens": 7410887.0,
+      "step": 3200
+    },
+    {
+      "epoch": 7.712130356065178,
+      "eval_entropy": 0.2626291513442993,
+      "eval_loss": 0.91884446144104,
+      "eval_mean_token_accuracy": 0.827563955207889,
+      "eval_num_tokens": 7410887.0,
+      "eval_runtime": 87.8849,
+      "eval_samples_per_second": 16.157,
+      "eval_steps_per_second": 2.025,
+      "step": 3200
+    },
+    {
+      "entropy": 0.11767532899975777,
+      "epoch": 7.76041038020519,
+      "grad_norm": 0.4675362706184387,
+      "learning_rate": 5.479854558149187e-05,
+      "loss": 0.09038087725639343,
+      "mean_token_accuracy": 0.9669680051505566,
+      "num_tokens": 7457578.0,
+      "step": 3220
+    },
+    {
+      "epoch": 7.76041038020519,
+      "eval_entropy": 0.26581037203582486,
+      "eval_loss": 0.9146883487701416,
+      "eval_mean_token_accuracy": 0.8270345812433222,
+      "eval_num_tokens": 7457578.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 3220
+    },
+    {
+      "entropy": 0.11383061073720455,
+      "epoch": 7.808690404345202,
+      "grad_norm": 0.541247546672821,
+      "learning_rate": 5.258468515101918e-05,
+      "loss": 0.08942405581474304,
+      "mean_token_accuracy": 0.968398705124855,
+      "num_tokens": 7505761.0,
+      "step": 3240
+    },
+    {
+      "epoch": 7.808690404345202,
+      "eval_entropy": 0.26687998960861997,
+      "eval_loss": 0.9037085175514221,
+      "eval_mean_token_accuracy": 0.8274977863504646,
+      "eval_num_tokens": 7505761.0,
+      "eval_runtime": 88.5163,
+      "eval_samples_per_second": 16.042,
+      "eval_steps_per_second": 2.011,
+      "step": 3240
+    },
+    {
+      "entropy": 0.12004058873280883,
+      "epoch": 7.856970428485214,
+      "grad_norm": 0.5289552211761475,
+      "learning_rate": 5.040918622423899e-05,
+      "loss": 0.09064805507659912,
+      "mean_token_accuracy": 0.9673330299556255,
+      "num_tokens": 7552557.0,
+      "step": 3260
+    },
+    {
+      "epoch": 7.856970428485214,
+      "eval_entropy": 0.26288379251622085,
+      "eval_loss": 0.9146178364753723,
+      "eval_mean_token_accuracy": 0.8276662448149049,
+      "eval_num_tokens": 7552557.0,
+      "eval_runtime": 88.0723,
+      "eval_samples_per_second": 16.123,
+      "eval_steps_per_second": 2.021,
+      "step": 3260
+    },
+    {
+      "entropy": 0.1180689456872642,
+      "epoch": 7.905250452625227,
+      "grad_norm": 0.4754861891269684,
+      "learning_rate": 4.827266444161377e-05,
+      "loss": 0.09402655363082886,
+      "mean_token_accuracy": 0.9663361787796021,
+      "num_tokens": 7596998.0,
+      "step": 3280
+    },
+    {
+      "epoch": 7.905250452625227,
+      "eval_entropy": 0.261690996922134,
+      "eval_loss": 0.9120186567306519,
+      "eval_mean_token_accuracy": 0.8278995015647974,
+      "eval_num_tokens": 7596998.0,
+      "eval_runtime": 88.6573,
+      "eval_samples_per_second": 16.017,
+      "eval_steps_per_second": 2.008,
+      "step": 3280
+    },
+    {
+      "entropy": 0.1162164019420743,
+      "epoch": 7.953530476765239,
+      "grad_norm": 0.6203398108482361,
+      "learning_rate": 4.61757244135355e-05,
+      "loss": 0.08832768201828003,
+      "mean_token_accuracy": 0.9694220684468746,
+      "num_tokens": 7643288.0,
+      "step": 3300
+    },
+    {
+      "epoch": 7.953530476765239,
+      "eval_entropy": 0.26158678129817664,
+      "eval_loss": 0.9301361441612244,
+      "eval_mean_token_accuracy": 0.8269603885291668,
+      "eval_num_tokens": 7643288.0,
+      "eval_runtime": 88.7064,
+      "eval_samples_per_second": 16.008,
+      "eval_steps_per_second": 2.007,
+      "step": 3300
+    },
+    {
+      "entropy": 0.10877202204489089,
+      "epoch": 8.0,
+      "grad_norm": 3.058673143386841,
+      "learning_rate": 4.411895954922763e-05,
+      "loss": 0.09159615635871887,
+      "mean_token_accuracy": 0.9686878834451947,
+      "num_tokens": 7688608.0,
+      "step": 3320
+    },
+    {
+      "epoch": 8.0,
+      "eval_entropy": 0.2590438057197614,
+      "eval_loss": 0.9262065887451172,
+      "eval_mean_token_accuracy": 0.8277404546067956,
+      "eval_num_tokens": 7688608.0,
+      "eval_runtime": 87.7557,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 3320
+    },
+    {
+      "entropy": 0.09969092034734786,
+      "epoch": 8.048280024140013,
+      "grad_norm": 0.2561911642551422,
+      "learning_rate": 4.2102951888817644e-05,
+      "loss": 0.07228735685348511,
+      "mean_token_accuracy": 0.9743556551635265,
+      "num_tokens": 7732856.0,
+      "step": 3340
+    },
+    {
+      "epoch": 8.048280024140013,
+      "eval_entropy": 0.2462534454933713,
+      "eval_loss": 0.9755983352661133,
+      "eval_mean_token_accuracy": 0.827510209565752,
+      "eval_num_tokens": 7732856.0,
+      "eval_runtime": 88.0297,
+      "eval_samples_per_second": 16.131,
+      "eval_steps_per_second": 2.022,
+      "step": 3340
+    },
+    {
+      "entropy": 0.09477624730207027,
+      "epoch": 8.096560048280024,
+      "grad_norm": 0.3670356571674347,
+      "learning_rate": 4.012827193862633e-05,
+      "loss": 0.0686036467552185,
+      "mean_token_accuracy": 0.9743670664727688,
+      "num_tokens": 7778585.0,
+      "step": 3360
+    },
+    {
+      "epoch": 8.096560048280024,
+      "eval_entropy": 0.2440882951356052,
+      "eval_loss": 1.0118423700332642,
+      "eval_mean_token_accuracy": 0.8271241070849172,
+      "eval_num_tokens": 7778585.0,
+      "eval_runtime": 88.1364,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 3360
+    },
+    {
+      "entropy": 0.09417980071157217,
+      "epoch": 8.144840072420036,
+      "grad_norm": 0.5785111784934998,
+      "learning_rate": 3.819547850972133e-05,
+      "loss": 0.07100698947906495,
+      "mean_token_accuracy": 0.9737804099917412,
+      "num_tokens": 7824287.0,
+      "step": 3380
+    },
+    {
+      "epoch": 8.144840072420036,
+      "eval_entropy": 0.24603172860453637,
+      "eval_loss": 1.0021991729736328,
+      "eval_mean_token_accuracy": 0.827162122793412,
+      "eval_num_tokens": 7824287.0,
+      "eval_runtime": 88.4534,
+      "eval_samples_per_second": 16.054,
+      "eval_steps_per_second": 2.012,
+      "step": 3380
+    },
+    {
+      "entropy": 0.087222247896716,
+      "epoch": 8.193120096560047,
+      "grad_norm": 0.44848504662513733,
+      "learning_rate": 3.6305118559780393e-05,
+      "loss": 0.06325234174728393,
+      "mean_token_accuracy": 0.9769335232675076,
+      "num_tokens": 7874784.0,
+      "step": 3400
+    },
+    {
+      "epoch": 8.193120096560047,
+      "eval_entropy": 0.24544863905129807,
+      "eval_loss": 1.005008339881897,
+      "eval_mean_token_accuracy": 0.8273035240976998,
+      "eval_num_tokens": 7874784.0,
+      "eval_runtime": 88.6737,
+      "eval_samples_per_second": 16.014,
+      "eval_steps_per_second": 2.007,
+      "step": 3400
+    },
+    {
+      "entropy": 0.09037275062873959,
+      "epoch": 8.24140012070006,
+      "grad_norm": 0.23529155552387238,
+      "learning_rate": 3.4457727038308435e-05,
+      "loss": 0.06589569449424744,
+      "mean_token_accuracy": 0.9751470290124417,
+      "num_tokens": 7925240.0,
+      "step": 3420
+    },
+    {
+      "epoch": 8.24140012070006,
+      "eval_entropy": 0.24203423494368456,
+      "eval_loss": 1.0052824020385742,
+      "eval_mean_token_accuracy": 0.8275575503874361,
+      "eval_num_tokens": 7925240.0,
+      "eval_runtime": 88.3486,
+      "eval_samples_per_second": 16.073,
+      "eval_steps_per_second": 2.015,
+      "step": 3420
+    },
+    {
+      "entropy": 0.10373208015225828,
+      "epoch": 8.289680144840073,
+      "grad_norm": 0.24796408414840698,
+      "learning_rate": 3.2653826735253594e-05,
+      "loss": 0.07714771628379821,
+      "mean_token_accuracy": 0.9722891598939896,
+      "num_tokens": 7966685.0,
+      "step": 3440
+    },
+    {
+      "epoch": 8.289680144840073,
+      "eval_entropy": 0.23461157050025597,
+      "eval_loss": 1.012108325958252,
+      "eval_mean_token_accuracy": 0.82825900564033,
+      "eval_num_tokens": 7966685.0,
+      "eval_runtime": 88.1018,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 3440
+    },
+    {
+      "entropy": 0.08967954274266958,
+      "epoch": 8.337960168980084,
+      "grad_norm": 0.5314674973487854,
+      "learning_rate": 3.08939281330635e-05,
+      "loss": 0.06746266484260559,
+      "mean_token_accuracy": 0.9753133699297905,
+      "num_tokens": 8013656.0,
+      "step": 3460
+    },
+    {
+      "epoch": 8.337960168980084,
+      "eval_entropy": 0.23781998589467468,
+      "eval_loss": 1.0119913816452026,
+      "eval_mean_token_accuracy": 0.8273287157664139,
+      "eval_num_tokens": 8013656.0,
+      "eval_runtime": 88.6038,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3460
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.386422272848732e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3480/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3480/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3480/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3480/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3480/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3480/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3480/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3480/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3480/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3480/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3480/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3480/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3480/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3480/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..b56321eace4e4aeb99d65166959aae545d8c066b
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3480/trainer_state.json
@@ -0,0 +1,3688 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 8.386240193120097,
+  "eval_steps": 20,
+  "global_step": 3480,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    },
+    {
+      "entropy": 0.1326121488586068,
+      "epoch": 7.181050090525045,
+      "grad_norm": 0.4843117296695709,
+      "learning_rate": 8.411870386507724e-05,
+      "loss": 0.09537842869758606,
+      "mean_token_accuracy": 0.9645151920616627,
+      "num_tokens": 6898444.0,
+      "step": 2980
+    },
+    {
+      "epoch": 7.181050090525045,
+      "eval_entropy": 0.27999477783280813,
+      "eval_loss": 0.9031037092208862,
+      "eval_mean_token_accuracy": 0.8261882204018282,
+      "eval_num_tokens": 6898444.0,
+      "eval_runtime": 88.5602,
+      "eval_samples_per_second": 16.034,
+      "eval_steps_per_second": 2.01,
+      "step": 2980
+    },
+    {
+      "entropy": 0.1156751093454659,
+      "epoch": 7.229330114665057,
+      "grad_norm": 0.5297922492027283,
+      "learning_rate": 8.149629351430326e-05,
+      "loss": 0.09203402996063233,
+      "mean_token_accuracy": 0.9689733698964119,
+      "num_tokens": 6947106.0,
+      "step": 3000
+    },
+    {
+      "epoch": 7.229330114665057,
+      "eval_entropy": 0.2740225845508361,
+      "eval_loss": 0.8876440525054932,
+      "eval_mean_token_accuracy": 0.8265116857678703,
+      "eval_num_tokens": 6947106.0,
+      "eval_runtime": 87.7971,
+      "eval_samples_per_second": 16.174,
+      "eval_steps_per_second": 2.027,
+      "step": 3000
+    },
+    {
+      "entropy": 0.12116114404052496,
+      "epoch": 7.27761013880507,
+      "grad_norm": 0.5467398166656494,
+      "learning_rate": 7.890406302404618e-05,
+      "loss": 0.0933056652545929,
+      "mean_token_accuracy": 0.9676455594599247,
+      "num_tokens": 6992784.0,
+      "step": 3020
+    },
+    {
+      "epoch": 7.27761013880507,
+      "eval_entropy": 0.27238342793804876,
+      "eval_loss": 0.9166117310523987,
+      "eval_mean_token_accuracy": 0.825499015242866,
+      "eval_num_tokens": 6992784.0,
+      "eval_runtime": 88.3276,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 3020
+    },
+    {
+      "entropy": 0.11715119928121567,
+      "epoch": 7.325890162945082,
+      "grad_norm": 0.4976702332496643,
+      "learning_rate": 7.634274596487176e-05,
+      "loss": 0.09449659585952759,
+      "mean_token_accuracy": 0.966506478190422,
+      "num_tokens": 7040482.0,
+      "step": 3040
+    },
+    {
+      "epoch": 7.325890162945082,
+      "eval_entropy": 0.2788975556077582,
+      "eval_loss": 0.8833065629005432,
+      "eval_mean_token_accuracy": 0.8269920596915684,
+      "eval_num_tokens": 7040482.0,
+      "eval_runtime": 88.9016,
+      "eval_samples_per_second": 15.973,
+      "eval_steps_per_second": 2.002,
+      "step": 3040
+    },
+    {
+      "entropy": 0.12565911412239075,
+      "epoch": 7.3741701870850935,
+      "grad_norm": 0.5728679895401001,
+      "learning_rate": 7.381306715921025e-05,
+      "loss": 0.09626244902610778,
+      "mean_token_accuracy": 0.9661444149911403,
+      "num_tokens": 7086261.0,
+      "step": 3060
+    },
+    {
+      "epoch": 7.3741701870850935,
+      "eval_entropy": 0.2762492011939542,
+      "eval_loss": 0.8961160778999329,
+      "eval_mean_token_accuracy": 0.826369910092836,
+      "eval_num_tokens": 7086261.0,
+      "eval_runtime": 88.5204,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 3060
+    },
+    {
+      "entropy": 0.12258567921817302,
+      "epoch": 7.422450211225105,
+      "grad_norm": 0.468985915184021,
+      "learning_rate": 7.131574247624031e-05,
+      "loss": 0.09120266437530518,
+      "mean_token_accuracy": 0.9674938447773457,
+      "num_tokens": 7133247.0,
+      "step": 3080
+    },
+    {
+      "epoch": 7.422450211225105,
+      "eval_entropy": 0.2688512086366,
+      "eval_loss": 0.9048654437065125,
+      "eval_mean_token_accuracy": 0.8265582302982888,
+      "eval_num_tokens": 7133247.0,
+      "eval_runtime": 88.164,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 3080
+    },
+    {
+      "entropy": 0.12070536343380808,
+      "epoch": 7.470730235365117,
+      "grad_norm": 0.5310841798782349,
+      "learning_rate": 6.885147862930688e-05,
+      "loss": 0.09386247992515565,
+      "mean_token_accuracy": 0.96664222702384,
+      "num_tokens": 7177810.0,
+      "step": 3100
+    },
+    {
+      "epoch": 7.470730235365117,
+      "eval_entropy": 0.26289632874593305,
+      "eval_loss": 0.902837336063385,
+      "eval_mean_token_accuracy": 0.828052184554968,
+      "eval_num_tokens": 7177810.0,
+      "eval_runtime": 88.1201,
+      "eval_samples_per_second": 16.114,
+      "eval_steps_per_second": 2.02,
+      "step": 3100
+    },
+    {
+      "entropy": 0.11322665181942285,
+      "epoch": 7.51901025950513,
+      "grad_norm": 0.6846067905426025,
+      "learning_rate": 6.642097297592908e-05,
+      "loss": 0.08772538304328918,
+      "mean_token_accuracy": 0.9691362135112286,
+      "num_tokens": 7225302.0,
+      "step": 3120
+    },
+    {
+      "epoch": 7.51901025950513,
+      "eval_entropy": 0.2670605613776807,
+      "eval_loss": 0.9130566120147705,
+      "eval_mean_token_accuracy": 0.826366903741708,
+      "eval_num_tokens": 7225302.0,
+      "eval_runtime": 88.6035,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3120
+    },
+    {
+      "entropy": 0.10985926580615342,
+      "epoch": 7.567290283645142,
+      "grad_norm": 0.4799748361110687,
+      "learning_rate": 6.402491332045683e-05,
+      "loss": 0.08672296404838561,
+      "mean_token_accuracy": 0.9698824048042297,
+      "num_tokens": 7274881.0,
+      "step": 3140
+    },
+    {
+      "epoch": 7.567290283645142,
+      "eval_entropy": 0.26657010866015146,
+      "eval_loss": 0.9056380987167358,
+      "eval_mean_token_accuracy": 0.8272326461384806,
+      "eval_num_tokens": 7274881.0,
+      "eval_runtime": 88.2967,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3140
+    },
+    {
+      "entropy": 0.11890473761595785,
+      "epoch": 7.615570307785154,
+      "grad_norm": 0.4081912934780121,
+      "learning_rate": 6.166397771942987e-05,
+      "loss": 0.09396827220916748,
+      "mean_token_accuracy": 0.967512384802103,
+      "num_tokens": 7321163.0,
+      "step": 3160
+    },
+    {
+      "epoch": 7.615570307785154,
+      "eval_entropy": 0.26364776575833226,
+      "eval_loss": 0.9204733967781067,
+      "eval_mean_token_accuracy": 0.827419729714983,
+      "eval_num_tokens": 7321163.0,
+      "eval_runtime": 88.3441,
+      "eval_samples_per_second": 16.074,
+      "eval_steps_per_second": 2.015,
+      "step": 3160
+    },
+    {
+      "entropy": 0.12209134344011545,
+      "epoch": 7.663850331925166,
+      "grad_norm": 0.5159774422645569,
+      "learning_rate": 5.93388342896962e-05,
+      "loss": 0.09292978644371033,
+      "mean_token_accuracy": 0.9658421002328396,
+      "num_tokens": 7366217.0,
+      "step": 3180
+    },
+    {
+      "epoch": 7.663850331925166,
+      "eval_entropy": 0.2714110068725736,
+      "eval_loss": 0.9143092036247253,
+      "eval_mean_token_accuracy": 0.8262251099843657,
+      "eval_num_tokens": 7366217.0,
+      "eval_runtime": 87.9635,
+      "eval_samples_per_second": 16.143,
+      "eval_steps_per_second": 2.024,
+      "step": 3180
+    },
+    {
+      "entropy": 0.11955609126016498,
+      "epoch": 7.712130356065178,
+      "grad_norm": 0.7005976438522339,
+      "learning_rate": 5.705014101934252e-05,
+      "loss": 0.09319776892662049,
+      "mean_token_accuracy": 0.9674513377249241,
+      "num_tokens": 7410887.0,
+      "step": 3200
+    },
+    {
+      "epoch": 7.712130356065178,
+      "eval_entropy": 0.2626291513442993,
+      "eval_loss": 0.91884446144104,
+      "eval_mean_token_accuracy": 0.827563955207889,
+      "eval_num_tokens": 7410887.0,
+      "eval_runtime": 87.8849,
+      "eval_samples_per_second": 16.157,
+      "eval_steps_per_second": 2.025,
+      "step": 3200
+    },
+    {
+      "entropy": 0.11767532899975777,
+      "epoch": 7.76041038020519,
+      "grad_norm": 0.4675362706184387,
+      "learning_rate": 5.479854558149187e-05,
+      "loss": 0.09038087725639343,
+      "mean_token_accuracy": 0.9669680051505566,
+      "num_tokens": 7457578.0,
+      "step": 3220
+    },
+    {
+      "epoch": 7.76041038020519,
+      "eval_entropy": 0.26581037203582486,
+      "eval_loss": 0.9146883487701416,
+      "eval_mean_token_accuracy": 0.8270345812433222,
+      "eval_num_tokens": 7457578.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 3220
+    },
+    {
+      "entropy": 0.11383061073720455,
+      "epoch": 7.808690404345202,
+      "grad_norm": 0.541247546672821,
+      "learning_rate": 5.258468515101918e-05,
+      "loss": 0.08942405581474304,
+      "mean_token_accuracy": 0.968398705124855,
+      "num_tokens": 7505761.0,
+      "step": 3240
+    },
+    {
+      "epoch": 7.808690404345202,
+      "eval_entropy": 0.26687998960861997,
+      "eval_loss": 0.9037085175514221,
+      "eval_mean_token_accuracy": 0.8274977863504646,
+      "eval_num_tokens": 7505761.0,
+      "eval_runtime": 88.5163,
+      "eval_samples_per_second": 16.042,
+      "eval_steps_per_second": 2.011,
+      "step": 3240
+    },
+    {
+      "entropy": 0.12004058873280883,
+      "epoch": 7.856970428485214,
+      "grad_norm": 0.5289552211761475,
+      "learning_rate": 5.040918622423899e-05,
+      "loss": 0.09064805507659912,
+      "mean_token_accuracy": 0.9673330299556255,
+      "num_tokens": 7552557.0,
+      "step": 3260
+    },
+    {
+      "epoch": 7.856970428485214,
+      "eval_entropy": 0.26288379251622085,
+      "eval_loss": 0.9146178364753723,
+      "eval_mean_token_accuracy": 0.8276662448149049,
+      "eval_num_tokens": 7552557.0,
+      "eval_runtime": 88.0723,
+      "eval_samples_per_second": 16.123,
+      "eval_steps_per_second": 2.021,
+      "step": 3260
+    },
+    {
+      "entropy": 0.1180689456872642,
+      "epoch": 7.905250452625227,
+      "grad_norm": 0.4754861891269684,
+      "learning_rate": 4.827266444161377e-05,
+      "loss": 0.09402655363082886,
+      "mean_token_accuracy": 0.9663361787796021,
+      "num_tokens": 7596998.0,
+      "step": 3280
+    },
+    {
+      "epoch": 7.905250452625227,
+      "eval_entropy": 0.261690996922134,
+      "eval_loss": 0.9120186567306519,
+      "eval_mean_token_accuracy": 0.8278995015647974,
+      "eval_num_tokens": 7596998.0,
+      "eval_runtime": 88.6573,
+      "eval_samples_per_second": 16.017,
+      "eval_steps_per_second": 2.008,
+      "step": 3280
+    },
+    {
+      "entropy": 0.1162164019420743,
+      "epoch": 7.953530476765239,
+      "grad_norm": 0.6203398108482361,
+      "learning_rate": 4.61757244135355e-05,
+      "loss": 0.08832768201828003,
+      "mean_token_accuracy": 0.9694220684468746,
+      "num_tokens": 7643288.0,
+      "step": 3300
+    },
+    {
+      "epoch": 7.953530476765239,
+      "eval_entropy": 0.26158678129817664,
+      "eval_loss": 0.9301361441612244,
+      "eval_mean_token_accuracy": 0.8269603885291668,
+      "eval_num_tokens": 7643288.0,
+      "eval_runtime": 88.7064,
+      "eval_samples_per_second": 16.008,
+      "eval_steps_per_second": 2.007,
+      "step": 3300
+    },
+    {
+      "entropy": 0.10877202204489089,
+      "epoch": 8.0,
+      "grad_norm": 3.058673143386841,
+      "learning_rate": 4.411895954922763e-05,
+      "loss": 0.09159615635871887,
+      "mean_token_accuracy": 0.9686878834451947,
+      "num_tokens": 7688608.0,
+      "step": 3320
+    },
+    {
+      "epoch": 8.0,
+      "eval_entropy": 0.2590438057197614,
+      "eval_loss": 0.9262065887451172,
+      "eval_mean_token_accuracy": 0.8277404546067956,
+      "eval_num_tokens": 7688608.0,
+      "eval_runtime": 87.7557,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 3320
+    },
+    {
+      "entropy": 0.09969092034734786,
+      "epoch": 8.048280024140013,
+      "grad_norm": 0.2561911642551422,
+      "learning_rate": 4.2102951888817644e-05,
+      "loss": 0.07228735685348511,
+      "mean_token_accuracy": 0.9743556551635265,
+      "num_tokens": 7732856.0,
+      "step": 3340
+    },
+    {
+      "epoch": 8.048280024140013,
+      "eval_entropy": 0.2462534454933713,
+      "eval_loss": 0.9755983352661133,
+      "eval_mean_token_accuracy": 0.827510209565752,
+      "eval_num_tokens": 7732856.0,
+      "eval_runtime": 88.0297,
+      "eval_samples_per_second": 16.131,
+      "eval_steps_per_second": 2.022,
+      "step": 3340
+    },
+    {
+      "entropy": 0.09477624730207027,
+      "epoch": 8.096560048280024,
+      "grad_norm": 0.3670356571674347,
+      "learning_rate": 4.012827193862633e-05,
+      "loss": 0.0686036467552185,
+      "mean_token_accuracy": 0.9743670664727688,
+      "num_tokens": 7778585.0,
+      "step": 3360
+    },
+    {
+      "epoch": 8.096560048280024,
+      "eval_entropy": 0.2440882951356052,
+      "eval_loss": 1.0118423700332642,
+      "eval_mean_token_accuracy": 0.8271241070849172,
+      "eval_num_tokens": 7778585.0,
+      "eval_runtime": 88.1364,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 3360
+    },
+    {
+      "entropy": 0.09417980071157217,
+      "epoch": 8.144840072420036,
+      "grad_norm": 0.5785111784934998,
+      "learning_rate": 3.819547850972133e-05,
+      "loss": 0.07100698947906495,
+      "mean_token_accuracy": 0.9737804099917412,
+      "num_tokens": 7824287.0,
+      "step": 3380
+    },
+    {
+      "epoch": 8.144840072420036,
+      "eval_entropy": 0.24603172860453637,
+      "eval_loss": 1.0021991729736328,
+      "eval_mean_token_accuracy": 0.827162122793412,
+      "eval_num_tokens": 7824287.0,
+      "eval_runtime": 88.4534,
+      "eval_samples_per_second": 16.054,
+      "eval_steps_per_second": 2.012,
+      "step": 3380
+    },
+    {
+      "entropy": 0.087222247896716,
+      "epoch": 8.193120096560047,
+      "grad_norm": 0.44848504662513733,
+      "learning_rate": 3.6305118559780393e-05,
+      "loss": 0.06325234174728393,
+      "mean_token_accuracy": 0.9769335232675076,
+      "num_tokens": 7874784.0,
+      "step": 3400
+    },
+    {
+      "epoch": 8.193120096560047,
+      "eval_entropy": 0.24544863905129807,
+      "eval_loss": 1.005008339881897,
+      "eval_mean_token_accuracy": 0.8273035240976998,
+      "eval_num_tokens": 7874784.0,
+      "eval_runtime": 88.6737,
+      "eval_samples_per_second": 16.014,
+      "eval_steps_per_second": 2.007,
+      "step": 3400
+    },
+    {
+      "entropy": 0.09037275062873959,
+      "epoch": 8.24140012070006,
+      "grad_norm": 0.23529155552387238,
+      "learning_rate": 3.4457727038308435e-05,
+      "loss": 0.06589569449424744,
+      "mean_token_accuracy": 0.9751470290124417,
+      "num_tokens": 7925240.0,
+      "step": 3420
+    },
+    {
+      "epoch": 8.24140012070006,
+      "eval_entropy": 0.24203423494368456,
+      "eval_loss": 1.0052824020385742,
+      "eval_mean_token_accuracy": 0.8275575503874361,
+      "eval_num_tokens": 7925240.0,
+      "eval_runtime": 88.3486,
+      "eval_samples_per_second": 16.073,
+      "eval_steps_per_second": 2.015,
+      "step": 3420
+    },
+    {
+      "entropy": 0.10373208015225828,
+      "epoch": 8.289680144840073,
+      "grad_norm": 0.24796408414840698,
+      "learning_rate": 3.2653826735253594e-05,
+      "loss": 0.07714771628379821,
+      "mean_token_accuracy": 0.9722891598939896,
+      "num_tokens": 7966685.0,
+      "step": 3440
+    },
+    {
+      "epoch": 8.289680144840073,
+      "eval_entropy": 0.23461157050025597,
+      "eval_loss": 1.012108325958252,
+      "eval_mean_token_accuracy": 0.82825900564033,
+      "eval_num_tokens": 7966685.0,
+      "eval_runtime": 88.1018,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 3440
+    },
+    {
+      "entropy": 0.08967954274266958,
+      "epoch": 8.337960168980084,
+      "grad_norm": 0.5314674973487854,
+      "learning_rate": 3.08939281330635e-05,
+      "loss": 0.06746266484260559,
+      "mean_token_accuracy": 0.9753133699297905,
+      "num_tokens": 8013656.0,
+      "step": 3460
+    },
+    {
+      "epoch": 8.337960168980084,
+      "eval_entropy": 0.23781998589467468,
+      "eval_loss": 1.0119913816452026,
+      "eval_mean_token_accuracy": 0.8273287157664139,
+      "eval_num_tokens": 8013656.0,
+      "eval_runtime": 88.6038,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3460
+    },
+    {
+      "entropy": 0.09280927330255509,
+      "epoch": 8.386240193120097,
+      "grad_norm": 0.30658936500549316,
+      "learning_rate": 2.9178529262224946e-05,
+      "loss": 0.06936421990394592,
+      "mean_token_accuracy": 0.9742105580866337,
+      "num_tokens": 8061108.0,
+      "step": 3480
+    },
+    {
+      "epoch": 8.386240193120097,
+      "eval_entropy": 0.23672426734747512,
+      "eval_loss": 1.0162041187286377,
+      "eval_mean_token_accuracy": 0.8278394281194451,
+      "eval_num_tokens": 8061108.0,
+      "eval_runtime": 88.6015,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 3480
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.406589904948388e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3500/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3500/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3500/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3500/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3500/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3500/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3500/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3500/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3500/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3500/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3500/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3500/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3500/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3500/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..b43026a2f0b10b68026ae8a3c32b567280605fb2
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3500/trainer_state.json
@@ -0,0 +1,3709 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 8.43452021726011,
+  "eval_steps": 20,
+  "global_step": 3500,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    },
+    {
+      "entropy": 0.1326121488586068,
+      "epoch": 7.181050090525045,
+      "grad_norm": 0.4843117296695709,
+      "learning_rate": 8.411870386507724e-05,
+      "loss": 0.09537842869758606,
+      "mean_token_accuracy": 0.9645151920616627,
+      "num_tokens": 6898444.0,
+      "step": 2980
+    },
+    {
+      "epoch": 7.181050090525045,
+      "eval_entropy": 0.27999477783280813,
+      "eval_loss": 0.9031037092208862,
+      "eval_mean_token_accuracy": 0.8261882204018282,
+      "eval_num_tokens": 6898444.0,
+      "eval_runtime": 88.5602,
+      "eval_samples_per_second": 16.034,
+      "eval_steps_per_second": 2.01,
+      "step": 2980
+    },
+    {
+      "entropy": 0.1156751093454659,
+      "epoch": 7.229330114665057,
+      "grad_norm": 0.5297922492027283,
+      "learning_rate": 8.149629351430326e-05,
+      "loss": 0.09203402996063233,
+      "mean_token_accuracy": 0.9689733698964119,
+      "num_tokens": 6947106.0,
+      "step": 3000
+    },
+    {
+      "epoch": 7.229330114665057,
+      "eval_entropy": 0.2740225845508361,
+      "eval_loss": 0.8876440525054932,
+      "eval_mean_token_accuracy": 0.8265116857678703,
+      "eval_num_tokens": 6947106.0,
+      "eval_runtime": 87.7971,
+      "eval_samples_per_second": 16.174,
+      "eval_steps_per_second": 2.027,
+      "step": 3000
+    },
+    {
+      "entropy": 0.12116114404052496,
+      "epoch": 7.27761013880507,
+      "grad_norm": 0.5467398166656494,
+      "learning_rate": 7.890406302404618e-05,
+      "loss": 0.0933056652545929,
+      "mean_token_accuracy": 0.9676455594599247,
+      "num_tokens": 6992784.0,
+      "step": 3020
+    },
+    {
+      "epoch": 7.27761013880507,
+      "eval_entropy": 0.27238342793804876,
+      "eval_loss": 0.9166117310523987,
+      "eval_mean_token_accuracy": 0.825499015242866,
+      "eval_num_tokens": 6992784.0,
+      "eval_runtime": 88.3276,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 3020
+    },
+    {
+      "entropy": 0.11715119928121567,
+      "epoch": 7.325890162945082,
+      "grad_norm": 0.4976702332496643,
+      "learning_rate": 7.634274596487176e-05,
+      "loss": 0.09449659585952759,
+      "mean_token_accuracy": 0.966506478190422,
+      "num_tokens": 7040482.0,
+      "step": 3040
+    },
+    {
+      "epoch": 7.325890162945082,
+      "eval_entropy": 0.2788975556077582,
+      "eval_loss": 0.8833065629005432,
+      "eval_mean_token_accuracy": 0.8269920596915684,
+      "eval_num_tokens": 7040482.0,
+      "eval_runtime": 88.9016,
+      "eval_samples_per_second": 15.973,
+      "eval_steps_per_second": 2.002,
+      "step": 3040
+    },
+    {
+      "entropy": 0.12565911412239075,
+      "epoch": 7.3741701870850935,
+      "grad_norm": 0.5728679895401001,
+      "learning_rate": 7.381306715921025e-05,
+      "loss": 0.09626244902610778,
+      "mean_token_accuracy": 0.9661444149911403,
+      "num_tokens": 7086261.0,
+      "step": 3060
+    },
+    {
+      "epoch": 7.3741701870850935,
+      "eval_entropy": 0.2762492011939542,
+      "eval_loss": 0.8961160778999329,
+      "eval_mean_token_accuracy": 0.826369910092836,
+      "eval_num_tokens": 7086261.0,
+      "eval_runtime": 88.5204,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 3060
+    },
+    {
+      "entropy": 0.12258567921817302,
+      "epoch": 7.422450211225105,
+      "grad_norm": 0.468985915184021,
+      "learning_rate": 7.131574247624031e-05,
+      "loss": 0.09120266437530518,
+      "mean_token_accuracy": 0.9674938447773457,
+      "num_tokens": 7133247.0,
+      "step": 3080
+    },
+    {
+      "epoch": 7.422450211225105,
+      "eval_entropy": 0.2688512086366,
+      "eval_loss": 0.9048654437065125,
+      "eval_mean_token_accuracy": 0.8265582302982888,
+      "eval_num_tokens": 7133247.0,
+      "eval_runtime": 88.164,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 3080
+    },
+    {
+      "entropy": 0.12070536343380808,
+      "epoch": 7.470730235365117,
+      "grad_norm": 0.5310841798782349,
+      "learning_rate": 6.885147862930688e-05,
+      "loss": 0.09386247992515565,
+      "mean_token_accuracy": 0.96664222702384,
+      "num_tokens": 7177810.0,
+      "step": 3100
+    },
+    {
+      "epoch": 7.470730235365117,
+      "eval_entropy": 0.26289632874593305,
+      "eval_loss": 0.902837336063385,
+      "eval_mean_token_accuracy": 0.828052184554968,
+      "eval_num_tokens": 7177810.0,
+      "eval_runtime": 88.1201,
+      "eval_samples_per_second": 16.114,
+      "eval_steps_per_second": 2.02,
+      "step": 3100
+    },
+    {
+      "entropy": 0.11322665181942285,
+      "epoch": 7.51901025950513,
+      "grad_norm": 0.6846067905426025,
+      "learning_rate": 6.642097297592908e-05,
+      "loss": 0.08772538304328918,
+      "mean_token_accuracy": 0.9691362135112286,
+      "num_tokens": 7225302.0,
+      "step": 3120
+    },
+    {
+      "epoch": 7.51901025950513,
+      "eval_entropy": 0.2670605613776807,
+      "eval_loss": 0.9130566120147705,
+      "eval_mean_token_accuracy": 0.826366903741708,
+      "eval_num_tokens": 7225302.0,
+      "eval_runtime": 88.6035,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3120
+    },
+    {
+      "entropy": 0.10985926580615342,
+      "epoch": 7.567290283645142,
+      "grad_norm": 0.4799748361110687,
+      "learning_rate": 6.402491332045683e-05,
+      "loss": 0.08672296404838561,
+      "mean_token_accuracy": 0.9698824048042297,
+      "num_tokens": 7274881.0,
+      "step": 3140
+    },
+    {
+      "epoch": 7.567290283645142,
+      "eval_entropy": 0.26657010866015146,
+      "eval_loss": 0.9056380987167358,
+      "eval_mean_token_accuracy": 0.8272326461384806,
+      "eval_num_tokens": 7274881.0,
+      "eval_runtime": 88.2967,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3140
+    },
+    {
+      "entropy": 0.11890473761595785,
+      "epoch": 7.615570307785154,
+      "grad_norm": 0.4081912934780121,
+      "learning_rate": 6.166397771942987e-05,
+      "loss": 0.09396827220916748,
+      "mean_token_accuracy": 0.967512384802103,
+      "num_tokens": 7321163.0,
+      "step": 3160
+    },
+    {
+      "epoch": 7.615570307785154,
+      "eval_entropy": 0.26364776575833226,
+      "eval_loss": 0.9204733967781067,
+      "eval_mean_token_accuracy": 0.827419729714983,
+      "eval_num_tokens": 7321163.0,
+      "eval_runtime": 88.3441,
+      "eval_samples_per_second": 16.074,
+      "eval_steps_per_second": 2.015,
+      "step": 3160
+    },
+    {
+      "entropy": 0.12209134344011545,
+      "epoch": 7.663850331925166,
+      "grad_norm": 0.5159774422645569,
+      "learning_rate": 5.93388342896962e-05,
+      "loss": 0.09292978644371033,
+      "mean_token_accuracy": 0.9658421002328396,
+      "num_tokens": 7366217.0,
+      "step": 3180
+    },
+    {
+      "epoch": 7.663850331925166,
+      "eval_entropy": 0.2714110068725736,
+      "eval_loss": 0.9143092036247253,
+      "eval_mean_token_accuracy": 0.8262251099843657,
+      "eval_num_tokens": 7366217.0,
+      "eval_runtime": 87.9635,
+      "eval_samples_per_second": 16.143,
+      "eval_steps_per_second": 2.024,
+      "step": 3180
+    },
+    {
+      "entropy": 0.11955609126016498,
+      "epoch": 7.712130356065178,
+      "grad_norm": 0.7005976438522339,
+      "learning_rate": 5.705014101934252e-05,
+      "loss": 0.09319776892662049,
+      "mean_token_accuracy": 0.9674513377249241,
+      "num_tokens": 7410887.0,
+      "step": 3200
+    },
+    {
+      "epoch": 7.712130356065178,
+      "eval_entropy": 0.2626291513442993,
+      "eval_loss": 0.91884446144104,
+      "eval_mean_token_accuracy": 0.827563955207889,
+      "eval_num_tokens": 7410887.0,
+      "eval_runtime": 87.8849,
+      "eval_samples_per_second": 16.157,
+      "eval_steps_per_second": 2.025,
+      "step": 3200
+    },
+    {
+      "entropy": 0.11767532899975777,
+      "epoch": 7.76041038020519,
+      "grad_norm": 0.4675362706184387,
+      "learning_rate": 5.479854558149187e-05,
+      "loss": 0.09038087725639343,
+      "mean_token_accuracy": 0.9669680051505566,
+      "num_tokens": 7457578.0,
+      "step": 3220
+    },
+    {
+      "epoch": 7.76041038020519,
+      "eval_entropy": 0.26581037203582486,
+      "eval_loss": 0.9146883487701416,
+      "eval_mean_token_accuracy": 0.8270345812433222,
+      "eval_num_tokens": 7457578.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 3220
+    },
+    {
+      "entropy": 0.11383061073720455,
+      "epoch": 7.808690404345202,
+      "grad_norm": 0.541247546672821,
+      "learning_rate": 5.258468515101918e-05,
+      "loss": 0.08942405581474304,
+      "mean_token_accuracy": 0.968398705124855,
+      "num_tokens": 7505761.0,
+      "step": 3240
+    },
+    {
+      "epoch": 7.808690404345202,
+      "eval_entropy": 0.26687998960861997,
+      "eval_loss": 0.9037085175514221,
+      "eval_mean_token_accuracy": 0.8274977863504646,
+      "eval_num_tokens": 7505761.0,
+      "eval_runtime": 88.5163,
+      "eval_samples_per_second": 16.042,
+      "eval_steps_per_second": 2.011,
+      "step": 3240
+    },
+    {
+      "entropy": 0.12004058873280883,
+      "epoch": 7.856970428485214,
+      "grad_norm": 0.5289552211761475,
+      "learning_rate": 5.040918622423899e-05,
+      "loss": 0.09064805507659912,
+      "mean_token_accuracy": 0.9673330299556255,
+      "num_tokens": 7552557.0,
+      "step": 3260
+    },
+    {
+      "epoch": 7.856970428485214,
+      "eval_entropy": 0.26288379251622085,
+      "eval_loss": 0.9146178364753723,
+      "eval_mean_token_accuracy": 0.8276662448149049,
+      "eval_num_tokens": 7552557.0,
+      "eval_runtime": 88.0723,
+      "eval_samples_per_second": 16.123,
+      "eval_steps_per_second": 2.021,
+      "step": 3260
+    },
+    {
+      "entropy": 0.1180689456872642,
+      "epoch": 7.905250452625227,
+      "grad_norm": 0.4754861891269684,
+      "learning_rate": 4.827266444161377e-05,
+      "loss": 0.09402655363082886,
+      "mean_token_accuracy": 0.9663361787796021,
+      "num_tokens": 7596998.0,
+      "step": 3280
+    },
+    {
+      "epoch": 7.905250452625227,
+      "eval_entropy": 0.261690996922134,
+      "eval_loss": 0.9120186567306519,
+      "eval_mean_token_accuracy": 0.8278995015647974,
+      "eval_num_tokens": 7596998.0,
+      "eval_runtime": 88.6573,
+      "eval_samples_per_second": 16.017,
+      "eval_steps_per_second": 2.008,
+      "step": 3280
+    },
+    {
+      "entropy": 0.1162164019420743,
+      "epoch": 7.953530476765239,
+      "grad_norm": 0.6203398108482361,
+      "learning_rate": 4.61757244135355e-05,
+      "loss": 0.08832768201828003,
+      "mean_token_accuracy": 0.9694220684468746,
+      "num_tokens": 7643288.0,
+      "step": 3300
+    },
+    {
+      "epoch": 7.953530476765239,
+      "eval_entropy": 0.26158678129817664,
+      "eval_loss": 0.9301361441612244,
+      "eval_mean_token_accuracy": 0.8269603885291668,
+      "eval_num_tokens": 7643288.0,
+      "eval_runtime": 88.7064,
+      "eval_samples_per_second": 16.008,
+      "eval_steps_per_second": 2.007,
+      "step": 3300
+    },
+    {
+      "entropy": 0.10877202204489089,
+      "epoch": 8.0,
+      "grad_norm": 3.058673143386841,
+      "learning_rate": 4.411895954922763e-05,
+      "loss": 0.09159615635871887,
+      "mean_token_accuracy": 0.9686878834451947,
+      "num_tokens": 7688608.0,
+      "step": 3320
+    },
+    {
+      "epoch": 8.0,
+      "eval_entropy": 0.2590438057197614,
+      "eval_loss": 0.9262065887451172,
+      "eval_mean_token_accuracy": 0.8277404546067956,
+      "eval_num_tokens": 7688608.0,
+      "eval_runtime": 87.7557,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 3320
+    },
+    {
+      "entropy": 0.09969092034734786,
+      "epoch": 8.048280024140013,
+      "grad_norm": 0.2561911642551422,
+      "learning_rate": 4.2102951888817644e-05,
+      "loss": 0.07228735685348511,
+      "mean_token_accuracy": 0.9743556551635265,
+      "num_tokens": 7732856.0,
+      "step": 3340
+    },
+    {
+      "epoch": 8.048280024140013,
+      "eval_entropy": 0.2462534454933713,
+      "eval_loss": 0.9755983352661133,
+      "eval_mean_token_accuracy": 0.827510209565752,
+      "eval_num_tokens": 7732856.0,
+      "eval_runtime": 88.0297,
+      "eval_samples_per_second": 16.131,
+      "eval_steps_per_second": 2.022,
+      "step": 3340
+    },
+    {
+      "entropy": 0.09477624730207027,
+      "epoch": 8.096560048280024,
+      "grad_norm": 0.3670356571674347,
+      "learning_rate": 4.012827193862633e-05,
+      "loss": 0.0686036467552185,
+      "mean_token_accuracy": 0.9743670664727688,
+      "num_tokens": 7778585.0,
+      "step": 3360
+    },
+    {
+      "epoch": 8.096560048280024,
+      "eval_entropy": 0.2440882951356052,
+      "eval_loss": 1.0118423700332642,
+      "eval_mean_token_accuracy": 0.8271241070849172,
+      "eval_num_tokens": 7778585.0,
+      "eval_runtime": 88.1364,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 3360
+    },
+    {
+      "entropy": 0.09417980071157217,
+      "epoch": 8.144840072420036,
+      "grad_norm": 0.5785111784934998,
+      "learning_rate": 3.819547850972133e-05,
+      "loss": 0.07100698947906495,
+      "mean_token_accuracy": 0.9737804099917412,
+      "num_tokens": 7824287.0,
+      "step": 3380
+    },
+    {
+      "epoch": 8.144840072420036,
+      "eval_entropy": 0.24603172860453637,
+      "eval_loss": 1.0021991729736328,
+      "eval_mean_token_accuracy": 0.827162122793412,
+      "eval_num_tokens": 7824287.0,
+      "eval_runtime": 88.4534,
+      "eval_samples_per_second": 16.054,
+      "eval_steps_per_second": 2.012,
+      "step": 3380
+    },
+    {
+      "entropy": 0.087222247896716,
+      "epoch": 8.193120096560047,
+      "grad_norm": 0.44848504662513733,
+      "learning_rate": 3.6305118559780393e-05,
+      "loss": 0.06325234174728393,
+      "mean_token_accuracy": 0.9769335232675076,
+      "num_tokens": 7874784.0,
+      "step": 3400
+    },
+    {
+      "epoch": 8.193120096560047,
+      "eval_entropy": 0.24544863905129807,
+      "eval_loss": 1.005008339881897,
+      "eval_mean_token_accuracy": 0.8273035240976998,
+      "eval_num_tokens": 7874784.0,
+      "eval_runtime": 88.6737,
+      "eval_samples_per_second": 16.014,
+      "eval_steps_per_second": 2.007,
+      "step": 3400
+    },
+    {
+      "entropy": 0.09037275062873959,
+      "epoch": 8.24140012070006,
+      "grad_norm": 0.23529155552387238,
+      "learning_rate": 3.4457727038308435e-05,
+      "loss": 0.06589569449424744,
+      "mean_token_accuracy": 0.9751470290124417,
+      "num_tokens": 7925240.0,
+      "step": 3420
+    },
+    {
+      "epoch": 8.24140012070006,
+      "eval_entropy": 0.24203423494368456,
+      "eval_loss": 1.0052824020385742,
+      "eval_mean_token_accuracy": 0.8275575503874361,
+      "eval_num_tokens": 7925240.0,
+      "eval_runtime": 88.3486,
+      "eval_samples_per_second": 16.073,
+      "eval_steps_per_second": 2.015,
+      "step": 3420
+    },
+    {
+      "entropy": 0.10373208015225828,
+      "epoch": 8.289680144840073,
+      "grad_norm": 0.24796408414840698,
+      "learning_rate": 3.2653826735253594e-05,
+      "loss": 0.07714771628379821,
+      "mean_token_accuracy": 0.9722891598939896,
+      "num_tokens": 7966685.0,
+      "step": 3440
+    },
+    {
+      "epoch": 8.289680144840073,
+      "eval_entropy": 0.23461157050025597,
+      "eval_loss": 1.012108325958252,
+      "eval_mean_token_accuracy": 0.82825900564033,
+      "eval_num_tokens": 7966685.0,
+      "eval_runtime": 88.1018,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 3440
+    },
+    {
+      "entropy": 0.08967954274266958,
+      "epoch": 8.337960168980084,
+      "grad_norm": 0.5314674973487854,
+      "learning_rate": 3.08939281330635e-05,
+      "loss": 0.06746266484260559,
+      "mean_token_accuracy": 0.9753133699297905,
+      "num_tokens": 8013656.0,
+      "step": 3460
+    },
+    {
+      "epoch": 8.337960168980084,
+      "eval_entropy": 0.23781998589467468,
+      "eval_loss": 1.0119913816452026,
+      "eval_mean_token_accuracy": 0.8273287157664139,
+      "eval_num_tokens": 8013656.0,
+      "eval_runtime": 88.6038,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3460
+    },
+    {
+      "entropy": 0.09280927330255509,
+      "epoch": 8.386240193120097,
+      "grad_norm": 0.30658936500549316,
+      "learning_rate": 2.9178529262224946e-05,
+      "loss": 0.06936421990394592,
+      "mean_token_accuracy": 0.9742105580866337,
+      "num_tokens": 8061108.0,
+      "step": 3480
+    },
+    {
+      "epoch": 8.386240193120097,
+      "eval_entropy": 0.23672426734747512,
+      "eval_loss": 1.0162041187286377,
+      "eval_mean_token_accuracy": 0.8278394281194451,
+      "eval_num_tokens": 8061108.0,
+      "eval_runtime": 88.6015,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 3480
+    },
+    {
+      "entropy": 0.09195122267119586,
+      "epoch": 8.43452021726011,
+      "grad_norm": 0.24512454867362976,
+      "learning_rate": 2.7508115560327256e-05,
+      "loss": 0.06996064782142639,
+      "mean_token_accuracy": 0.974104854464531,
+      "num_tokens": 8106924.0,
+      "step": 3500
+    },
+    {
+      "epoch": 8.43452021726011,
+      "eval_entropy": 0.23758245828781235,
+      "eval_loss": 1.0157747268676758,
+      "eval_mean_token_accuracy": 0.8278501579600773,
+      "eval_num_tokens": 8106924.0,
+      "eval_runtime": 88.5678,
+      "eval_samples_per_second": 16.033,
+      "eval_steps_per_second": 2.01,
+      "step": 3500
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.426522448300155e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3520/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3520/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3520/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3520/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3520/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3520/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3520/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3520/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3520/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3520/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3520/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3520/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3520/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3520/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..42411db050929ee1d8b9ea04edafa04bef1772b6
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3520/trainer_state.json
@@ -0,0 +1,3730 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 8.48280024140012,
+  "eval_steps": 20,
+  "global_step": 3520,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    },
+    {
+      "entropy": 0.1326121488586068,
+      "epoch": 7.181050090525045,
+      "grad_norm": 0.4843117296695709,
+      "learning_rate": 8.411870386507724e-05,
+      "loss": 0.09537842869758606,
+      "mean_token_accuracy": 0.9645151920616627,
+      "num_tokens": 6898444.0,
+      "step": 2980
+    },
+    {
+      "epoch": 7.181050090525045,
+      "eval_entropy": 0.27999477783280813,
+      "eval_loss": 0.9031037092208862,
+      "eval_mean_token_accuracy": 0.8261882204018282,
+      "eval_num_tokens": 6898444.0,
+      "eval_runtime": 88.5602,
+      "eval_samples_per_second": 16.034,
+      "eval_steps_per_second": 2.01,
+      "step": 2980
+    },
+    {
+      "entropy": 0.1156751093454659,
+      "epoch": 7.229330114665057,
+      "grad_norm": 0.5297922492027283,
+      "learning_rate": 8.149629351430326e-05,
+      "loss": 0.09203402996063233,
+      "mean_token_accuracy": 0.9689733698964119,
+      "num_tokens": 6947106.0,
+      "step": 3000
+    },
+    {
+      "epoch": 7.229330114665057,
+      "eval_entropy": 0.2740225845508361,
+      "eval_loss": 0.8876440525054932,
+      "eval_mean_token_accuracy": 0.8265116857678703,
+      "eval_num_tokens": 6947106.0,
+      "eval_runtime": 87.7971,
+      "eval_samples_per_second": 16.174,
+      "eval_steps_per_second": 2.027,
+      "step": 3000
+    },
+    {
+      "entropy": 0.12116114404052496,
+      "epoch": 7.27761013880507,
+      "grad_norm": 0.5467398166656494,
+      "learning_rate": 7.890406302404618e-05,
+      "loss": 0.0933056652545929,
+      "mean_token_accuracy": 0.9676455594599247,
+      "num_tokens": 6992784.0,
+      "step": 3020
+    },
+    {
+      "epoch": 7.27761013880507,
+      "eval_entropy": 0.27238342793804876,
+      "eval_loss": 0.9166117310523987,
+      "eval_mean_token_accuracy": 0.825499015242866,
+      "eval_num_tokens": 6992784.0,
+      "eval_runtime": 88.3276,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 3020
+    },
+    {
+      "entropy": 0.11715119928121567,
+      "epoch": 7.325890162945082,
+      "grad_norm": 0.4976702332496643,
+      "learning_rate": 7.634274596487176e-05,
+      "loss": 0.09449659585952759,
+      "mean_token_accuracy": 0.966506478190422,
+      "num_tokens": 7040482.0,
+      "step": 3040
+    },
+    {
+      "epoch": 7.325890162945082,
+      "eval_entropy": 0.2788975556077582,
+      "eval_loss": 0.8833065629005432,
+      "eval_mean_token_accuracy": 0.8269920596915684,
+      "eval_num_tokens": 7040482.0,
+      "eval_runtime": 88.9016,
+      "eval_samples_per_second": 15.973,
+      "eval_steps_per_second": 2.002,
+      "step": 3040
+    },
+    {
+      "entropy": 0.12565911412239075,
+      "epoch": 7.3741701870850935,
+      "grad_norm": 0.5728679895401001,
+      "learning_rate": 7.381306715921025e-05,
+      "loss": 0.09626244902610778,
+      "mean_token_accuracy": 0.9661444149911403,
+      "num_tokens": 7086261.0,
+      "step": 3060
+    },
+    {
+      "epoch": 7.3741701870850935,
+      "eval_entropy": 0.2762492011939542,
+      "eval_loss": 0.8961160778999329,
+      "eval_mean_token_accuracy": 0.826369910092836,
+      "eval_num_tokens": 7086261.0,
+      "eval_runtime": 88.5204,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 3060
+    },
+    {
+      "entropy": 0.12258567921817302,
+      "epoch": 7.422450211225105,
+      "grad_norm": 0.468985915184021,
+      "learning_rate": 7.131574247624031e-05,
+      "loss": 0.09120266437530518,
+      "mean_token_accuracy": 0.9674938447773457,
+      "num_tokens": 7133247.0,
+      "step": 3080
+    },
+    {
+      "epoch": 7.422450211225105,
+      "eval_entropy": 0.2688512086366,
+      "eval_loss": 0.9048654437065125,
+      "eval_mean_token_accuracy": 0.8265582302982888,
+      "eval_num_tokens": 7133247.0,
+      "eval_runtime": 88.164,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 3080
+    },
+    {
+      "entropy": 0.12070536343380808,
+      "epoch": 7.470730235365117,
+      "grad_norm": 0.5310841798782349,
+      "learning_rate": 6.885147862930688e-05,
+      "loss": 0.09386247992515565,
+      "mean_token_accuracy": 0.96664222702384,
+      "num_tokens": 7177810.0,
+      "step": 3100
+    },
+    {
+      "epoch": 7.470730235365117,
+      "eval_entropy": 0.26289632874593305,
+      "eval_loss": 0.902837336063385,
+      "eval_mean_token_accuracy": 0.828052184554968,
+      "eval_num_tokens": 7177810.0,
+      "eval_runtime": 88.1201,
+      "eval_samples_per_second": 16.114,
+      "eval_steps_per_second": 2.02,
+      "step": 3100
+    },
+    {
+      "entropy": 0.11322665181942285,
+      "epoch": 7.51901025950513,
+      "grad_norm": 0.6846067905426025,
+      "learning_rate": 6.642097297592908e-05,
+      "loss": 0.08772538304328918,
+      "mean_token_accuracy": 0.9691362135112286,
+      "num_tokens": 7225302.0,
+      "step": 3120
+    },
+    {
+      "epoch": 7.51901025950513,
+      "eval_entropy": 0.2670605613776807,
+      "eval_loss": 0.9130566120147705,
+      "eval_mean_token_accuracy": 0.826366903741708,
+      "eval_num_tokens": 7225302.0,
+      "eval_runtime": 88.6035,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3120
+    },
+    {
+      "entropy": 0.10985926580615342,
+      "epoch": 7.567290283645142,
+      "grad_norm": 0.4799748361110687,
+      "learning_rate": 6.402491332045683e-05,
+      "loss": 0.08672296404838561,
+      "mean_token_accuracy": 0.9698824048042297,
+      "num_tokens": 7274881.0,
+      "step": 3140
+    },
+    {
+      "epoch": 7.567290283645142,
+      "eval_entropy": 0.26657010866015146,
+      "eval_loss": 0.9056380987167358,
+      "eval_mean_token_accuracy": 0.8272326461384806,
+      "eval_num_tokens": 7274881.0,
+      "eval_runtime": 88.2967,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3140
+    },
+    {
+      "entropy": 0.11890473761595785,
+      "epoch": 7.615570307785154,
+      "grad_norm": 0.4081912934780121,
+      "learning_rate": 6.166397771942987e-05,
+      "loss": 0.09396827220916748,
+      "mean_token_accuracy": 0.967512384802103,
+      "num_tokens": 7321163.0,
+      "step": 3160
+    },
+    {
+      "epoch": 7.615570307785154,
+      "eval_entropy": 0.26364776575833226,
+      "eval_loss": 0.9204733967781067,
+      "eval_mean_token_accuracy": 0.827419729714983,
+      "eval_num_tokens": 7321163.0,
+      "eval_runtime": 88.3441,
+      "eval_samples_per_second": 16.074,
+      "eval_steps_per_second": 2.015,
+      "step": 3160
+    },
+    {
+      "entropy": 0.12209134344011545,
+      "epoch": 7.663850331925166,
+      "grad_norm": 0.5159774422645569,
+      "learning_rate": 5.93388342896962e-05,
+      "loss": 0.09292978644371033,
+      "mean_token_accuracy": 0.9658421002328396,
+      "num_tokens": 7366217.0,
+      "step": 3180
+    },
+    {
+      "epoch": 7.663850331925166,
+      "eval_entropy": 0.2714110068725736,
+      "eval_loss": 0.9143092036247253,
+      "eval_mean_token_accuracy": 0.8262251099843657,
+      "eval_num_tokens": 7366217.0,
+      "eval_runtime": 87.9635,
+      "eval_samples_per_second": 16.143,
+      "eval_steps_per_second": 2.024,
+      "step": 3180
+    },
+    {
+      "entropy": 0.11955609126016498,
+      "epoch": 7.712130356065178,
+      "grad_norm": 0.7005976438522339,
+      "learning_rate": 5.705014101934252e-05,
+      "loss": 0.09319776892662049,
+      "mean_token_accuracy": 0.9674513377249241,
+      "num_tokens": 7410887.0,
+      "step": 3200
+    },
+    {
+      "epoch": 7.712130356065178,
+      "eval_entropy": 0.2626291513442993,
+      "eval_loss": 0.91884446144104,
+      "eval_mean_token_accuracy": 0.827563955207889,
+      "eval_num_tokens": 7410887.0,
+      "eval_runtime": 87.8849,
+      "eval_samples_per_second": 16.157,
+      "eval_steps_per_second": 2.025,
+      "step": 3200
+    },
+    {
+      "entropy": 0.11767532899975777,
+      "epoch": 7.76041038020519,
+      "grad_norm": 0.4675362706184387,
+      "learning_rate": 5.479854558149187e-05,
+      "loss": 0.09038087725639343,
+      "mean_token_accuracy": 0.9669680051505566,
+      "num_tokens": 7457578.0,
+      "step": 3220
+    },
+    {
+      "epoch": 7.76041038020519,
+      "eval_entropy": 0.26581037203582486,
+      "eval_loss": 0.9146883487701416,
+      "eval_mean_token_accuracy": 0.8270345812433222,
+      "eval_num_tokens": 7457578.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 3220
+    },
+    {
+      "entropy": 0.11383061073720455,
+      "epoch": 7.808690404345202,
+      "grad_norm": 0.541247546672821,
+      "learning_rate": 5.258468515101918e-05,
+      "loss": 0.08942405581474304,
+      "mean_token_accuracy": 0.968398705124855,
+      "num_tokens": 7505761.0,
+      "step": 3240
+    },
+    {
+      "epoch": 7.808690404345202,
+      "eval_entropy": 0.26687998960861997,
+      "eval_loss": 0.9037085175514221,
+      "eval_mean_token_accuracy": 0.8274977863504646,
+      "eval_num_tokens": 7505761.0,
+      "eval_runtime": 88.5163,
+      "eval_samples_per_second": 16.042,
+      "eval_steps_per_second": 2.011,
+      "step": 3240
+    },
+    {
+      "entropy": 0.12004058873280883,
+      "epoch": 7.856970428485214,
+      "grad_norm": 0.5289552211761475,
+      "learning_rate": 5.040918622423899e-05,
+      "loss": 0.09064805507659912,
+      "mean_token_accuracy": 0.9673330299556255,
+      "num_tokens": 7552557.0,
+      "step": 3260
+    },
+    {
+      "epoch": 7.856970428485214,
+      "eval_entropy": 0.26288379251622085,
+      "eval_loss": 0.9146178364753723,
+      "eval_mean_token_accuracy": 0.8276662448149049,
+      "eval_num_tokens": 7552557.0,
+      "eval_runtime": 88.0723,
+      "eval_samples_per_second": 16.123,
+      "eval_steps_per_second": 2.021,
+      "step": 3260
+    },
+    {
+      "entropy": 0.1180689456872642,
+      "epoch": 7.905250452625227,
+      "grad_norm": 0.4754861891269684,
+      "learning_rate": 4.827266444161377e-05,
+      "loss": 0.09402655363082886,
+      "mean_token_accuracy": 0.9663361787796021,
+      "num_tokens": 7596998.0,
+      "step": 3280
+    },
+    {
+      "epoch": 7.905250452625227,
+      "eval_entropy": 0.261690996922134,
+      "eval_loss": 0.9120186567306519,
+      "eval_mean_token_accuracy": 0.8278995015647974,
+      "eval_num_tokens": 7596998.0,
+      "eval_runtime": 88.6573,
+      "eval_samples_per_second": 16.017,
+      "eval_steps_per_second": 2.008,
+      "step": 3280
+    },
+    {
+      "entropy": 0.1162164019420743,
+      "epoch": 7.953530476765239,
+      "grad_norm": 0.6203398108482361,
+      "learning_rate": 4.61757244135355e-05,
+      "loss": 0.08832768201828003,
+      "mean_token_accuracy": 0.9694220684468746,
+      "num_tokens": 7643288.0,
+      "step": 3300
+    },
+    {
+      "epoch": 7.953530476765239,
+      "eval_entropy": 0.26158678129817664,
+      "eval_loss": 0.9301361441612244,
+      "eval_mean_token_accuracy": 0.8269603885291668,
+      "eval_num_tokens": 7643288.0,
+      "eval_runtime": 88.7064,
+      "eval_samples_per_second": 16.008,
+      "eval_steps_per_second": 2.007,
+      "step": 3300
+    },
+    {
+      "entropy": 0.10877202204489089,
+      "epoch": 8.0,
+      "grad_norm": 3.058673143386841,
+      "learning_rate": 4.411895954922763e-05,
+      "loss": 0.09159615635871887,
+      "mean_token_accuracy": 0.9686878834451947,
+      "num_tokens": 7688608.0,
+      "step": 3320
+    },
+    {
+      "epoch": 8.0,
+      "eval_entropy": 0.2590438057197614,
+      "eval_loss": 0.9262065887451172,
+      "eval_mean_token_accuracy": 0.8277404546067956,
+      "eval_num_tokens": 7688608.0,
+      "eval_runtime": 87.7557,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 3320
+    },
+    {
+      "entropy": 0.09969092034734786,
+      "epoch": 8.048280024140013,
+      "grad_norm": 0.2561911642551422,
+      "learning_rate": 4.2102951888817644e-05,
+      "loss": 0.07228735685348511,
+      "mean_token_accuracy": 0.9743556551635265,
+      "num_tokens": 7732856.0,
+      "step": 3340
+    },
+    {
+      "epoch": 8.048280024140013,
+      "eval_entropy": 0.2462534454933713,
+      "eval_loss": 0.9755983352661133,
+      "eval_mean_token_accuracy": 0.827510209565752,
+      "eval_num_tokens": 7732856.0,
+      "eval_runtime": 88.0297,
+      "eval_samples_per_second": 16.131,
+      "eval_steps_per_second": 2.022,
+      "step": 3340
+    },
+    {
+      "entropy": 0.09477624730207027,
+      "epoch": 8.096560048280024,
+      "grad_norm": 0.3670356571674347,
+      "learning_rate": 4.012827193862633e-05,
+      "loss": 0.0686036467552185,
+      "mean_token_accuracy": 0.9743670664727688,
+      "num_tokens": 7778585.0,
+      "step": 3360
+    },
+    {
+      "epoch": 8.096560048280024,
+      "eval_entropy": 0.2440882951356052,
+      "eval_loss": 1.0118423700332642,
+      "eval_mean_token_accuracy": 0.8271241070849172,
+      "eval_num_tokens": 7778585.0,
+      "eval_runtime": 88.1364,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 3360
+    },
+    {
+      "entropy": 0.09417980071157217,
+      "epoch": 8.144840072420036,
+      "grad_norm": 0.5785111784934998,
+      "learning_rate": 3.819547850972133e-05,
+      "loss": 0.07100698947906495,
+      "mean_token_accuracy": 0.9737804099917412,
+      "num_tokens": 7824287.0,
+      "step": 3380
+    },
+    {
+      "epoch": 8.144840072420036,
+      "eval_entropy": 0.24603172860453637,
+      "eval_loss": 1.0021991729736328,
+      "eval_mean_token_accuracy": 0.827162122793412,
+      "eval_num_tokens": 7824287.0,
+      "eval_runtime": 88.4534,
+      "eval_samples_per_second": 16.054,
+      "eval_steps_per_second": 2.012,
+      "step": 3380
+    },
+    {
+      "entropy": 0.087222247896716,
+      "epoch": 8.193120096560047,
+      "grad_norm": 0.44848504662513733,
+      "learning_rate": 3.6305118559780393e-05,
+      "loss": 0.06325234174728393,
+      "mean_token_accuracy": 0.9769335232675076,
+      "num_tokens": 7874784.0,
+      "step": 3400
+    },
+    {
+      "epoch": 8.193120096560047,
+      "eval_entropy": 0.24544863905129807,
+      "eval_loss": 1.005008339881897,
+      "eval_mean_token_accuracy": 0.8273035240976998,
+      "eval_num_tokens": 7874784.0,
+      "eval_runtime": 88.6737,
+      "eval_samples_per_second": 16.014,
+      "eval_steps_per_second": 2.007,
+      "step": 3400
+    },
+    {
+      "entropy": 0.09037275062873959,
+      "epoch": 8.24140012070006,
+      "grad_norm": 0.23529155552387238,
+      "learning_rate": 3.4457727038308435e-05,
+      "loss": 0.06589569449424744,
+      "mean_token_accuracy": 0.9751470290124417,
+      "num_tokens": 7925240.0,
+      "step": 3420
+    },
+    {
+      "epoch": 8.24140012070006,
+      "eval_entropy": 0.24203423494368456,
+      "eval_loss": 1.0052824020385742,
+      "eval_mean_token_accuracy": 0.8275575503874361,
+      "eval_num_tokens": 7925240.0,
+      "eval_runtime": 88.3486,
+      "eval_samples_per_second": 16.073,
+      "eval_steps_per_second": 2.015,
+      "step": 3420
+    },
+    {
+      "entropy": 0.10373208015225828,
+      "epoch": 8.289680144840073,
+      "grad_norm": 0.24796408414840698,
+      "learning_rate": 3.2653826735253594e-05,
+      "loss": 0.07714771628379821,
+      "mean_token_accuracy": 0.9722891598939896,
+      "num_tokens": 7966685.0,
+      "step": 3440
+    },
+    {
+      "epoch": 8.289680144840073,
+      "eval_entropy": 0.23461157050025597,
+      "eval_loss": 1.012108325958252,
+      "eval_mean_token_accuracy": 0.82825900564033,
+      "eval_num_tokens": 7966685.0,
+      "eval_runtime": 88.1018,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 3440
+    },
+    {
+      "entropy": 0.08967954274266958,
+      "epoch": 8.337960168980084,
+      "grad_norm": 0.5314674973487854,
+      "learning_rate": 3.08939281330635e-05,
+      "loss": 0.06746266484260559,
+      "mean_token_accuracy": 0.9753133699297905,
+      "num_tokens": 8013656.0,
+      "step": 3460
+    },
+    {
+      "epoch": 8.337960168980084,
+      "eval_entropy": 0.23781998589467468,
+      "eval_loss": 1.0119913816452026,
+      "eval_mean_token_accuracy": 0.8273287157664139,
+      "eval_num_tokens": 8013656.0,
+      "eval_runtime": 88.6038,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3460
+    },
+    {
+      "entropy": 0.09280927330255509,
+      "epoch": 8.386240193120097,
+      "grad_norm": 0.30658936500549316,
+      "learning_rate": 2.9178529262224946e-05,
+      "loss": 0.06936421990394592,
+      "mean_token_accuracy": 0.9742105580866337,
+      "num_tokens": 8061108.0,
+      "step": 3480
+    },
+    {
+      "epoch": 8.386240193120097,
+      "eval_entropy": 0.23672426734747512,
+      "eval_loss": 1.0162041187286377,
+      "eval_mean_token_accuracy": 0.8278394281194451,
+      "eval_num_tokens": 8061108.0,
+      "eval_runtime": 88.6015,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 3480
+    },
+    {
+      "entropy": 0.09195122267119586,
+      "epoch": 8.43452021726011,
+      "grad_norm": 0.24512454867362976,
+      "learning_rate": 2.7508115560327256e-05,
+      "loss": 0.06996064782142639,
+      "mean_token_accuracy": 0.974104854464531,
+      "num_tokens": 8106924.0,
+      "step": 3500
+    },
+    {
+      "epoch": 8.43452021726011,
+      "eval_entropy": 0.23758245828781235,
+      "eval_loss": 1.0157747268676758,
+      "eval_mean_token_accuracy": 0.8278501579600773,
+      "eval_num_tokens": 8106924.0,
+      "eval_runtime": 88.5678,
+      "eval_samples_per_second": 16.033,
+      "eval_steps_per_second": 2.01,
+      "step": 3500
+    },
+    {
+      "entropy": 0.09518477590754629,
+      "epoch": 8.48280024140012,
+      "grad_norm": 0.2373228371143341,
+      "learning_rate": 2.5883159734688868e-05,
+      "loss": 0.06832797527313232,
+      "mean_token_accuracy": 0.9743806682527065,
+      "num_tokens": 8153070.0,
+      "step": 3520
+    },
+    {
+      "epoch": 8.48280024140012,
+      "eval_entropy": 0.2379308074210467,
+      "eval_loss": 1.0180530548095703,
+      "eval_mean_token_accuracy": 0.8282477052024241,
+      "eval_num_tokens": 8153070.0,
+      "eval_runtime": 88.4651,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 3520
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.446115516482396e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3540/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3540/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3540/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3540/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3540/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3540/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3540/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3540/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3540/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3540/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3540/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3540/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3540/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3540/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..408c160e6c4bb39c0ea79d206f052bbef4ce7529
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3540/trainer_state.json
@@ -0,0 +1,3751 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 8.531080265540133,
+  "eval_steps": 20,
+  "global_step": 3540,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    },
+    {
+      "entropy": 0.1326121488586068,
+      "epoch": 7.181050090525045,
+      "grad_norm": 0.4843117296695709,
+      "learning_rate": 8.411870386507724e-05,
+      "loss": 0.09537842869758606,
+      "mean_token_accuracy": 0.9645151920616627,
+      "num_tokens": 6898444.0,
+      "step": 2980
+    },
+    {
+      "epoch": 7.181050090525045,
+      "eval_entropy": 0.27999477783280813,
+      "eval_loss": 0.9031037092208862,
+      "eval_mean_token_accuracy": 0.8261882204018282,
+      "eval_num_tokens": 6898444.0,
+      "eval_runtime": 88.5602,
+      "eval_samples_per_second": 16.034,
+      "eval_steps_per_second": 2.01,
+      "step": 2980
+    },
+    {
+      "entropy": 0.1156751093454659,
+      "epoch": 7.229330114665057,
+      "grad_norm": 0.5297922492027283,
+      "learning_rate": 8.149629351430326e-05,
+      "loss": 0.09203402996063233,
+      "mean_token_accuracy": 0.9689733698964119,
+      "num_tokens": 6947106.0,
+      "step": 3000
+    },
+    {
+      "epoch": 7.229330114665057,
+      "eval_entropy": 0.2740225845508361,
+      "eval_loss": 0.8876440525054932,
+      "eval_mean_token_accuracy": 0.8265116857678703,
+      "eval_num_tokens": 6947106.0,
+      "eval_runtime": 87.7971,
+      "eval_samples_per_second": 16.174,
+      "eval_steps_per_second": 2.027,
+      "step": 3000
+    },
+    {
+      "entropy": 0.12116114404052496,
+      "epoch": 7.27761013880507,
+      "grad_norm": 0.5467398166656494,
+      "learning_rate": 7.890406302404618e-05,
+      "loss": 0.0933056652545929,
+      "mean_token_accuracy": 0.9676455594599247,
+      "num_tokens": 6992784.0,
+      "step": 3020
+    },
+    {
+      "epoch": 7.27761013880507,
+      "eval_entropy": 0.27238342793804876,
+      "eval_loss": 0.9166117310523987,
+      "eval_mean_token_accuracy": 0.825499015242866,
+      "eval_num_tokens": 6992784.0,
+      "eval_runtime": 88.3276,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 3020
+    },
+    {
+      "entropy": 0.11715119928121567,
+      "epoch": 7.325890162945082,
+      "grad_norm": 0.4976702332496643,
+      "learning_rate": 7.634274596487176e-05,
+      "loss": 0.09449659585952759,
+      "mean_token_accuracy": 0.966506478190422,
+      "num_tokens": 7040482.0,
+      "step": 3040
+    },
+    {
+      "epoch": 7.325890162945082,
+      "eval_entropy": 0.2788975556077582,
+      "eval_loss": 0.8833065629005432,
+      "eval_mean_token_accuracy": 0.8269920596915684,
+      "eval_num_tokens": 7040482.0,
+      "eval_runtime": 88.9016,
+      "eval_samples_per_second": 15.973,
+      "eval_steps_per_second": 2.002,
+      "step": 3040
+    },
+    {
+      "entropy": 0.12565911412239075,
+      "epoch": 7.3741701870850935,
+      "grad_norm": 0.5728679895401001,
+      "learning_rate": 7.381306715921025e-05,
+      "loss": 0.09626244902610778,
+      "mean_token_accuracy": 0.9661444149911403,
+      "num_tokens": 7086261.0,
+      "step": 3060
+    },
+    {
+      "epoch": 7.3741701870850935,
+      "eval_entropy": 0.2762492011939542,
+      "eval_loss": 0.8961160778999329,
+      "eval_mean_token_accuracy": 0.826369910092836,
+      "eval_num_tokens": 7086261.0,
+      "eval_runtime": 88.5204,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 3060
+    },
+    {
+      "entropy": 0.12258567921817302,
+      "epoch": 7.422450211225105,
+      "grad_norm": 0.468985915184021,
+      "learning_rate": 7.131574247624031e-05,
+      "loss": 0.09120266437530518,
+      "mean_token_accuracy": 0.9674938447773457,
+      "num_tokens": 7133247.0,
+      "step": 3080
+    },
+    {
+      "epoch": 7.422450211225105,
+      "eval_entropy": 0.2688512086366,
+      "eval_loss": 0.9048654437065125,
+      "eval_mean_token_accuracy": 0.8265582302982888,
+      "eval_num_tokens": 7133247.0,
+      "eval_runtime": 88.164,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 3080
+    },
+    {
+      "entropy": 0.12070536343380808,
+      "epoch": 7.470730235365117,
+      "grad_norm": 0.5310841798782349,
+      "learning_rate": 6.885147862930688e-05,
+      "loss": 0.09386247992515565,
+      "mean_token_accuracy": 0.96664222702384,
+      "num_tokens": 7177810.0,
+      "step": 3100
+    },
+    {
+      "epoch": 7.470730235365117,
+      "eval_entropy": 0.26289632874593305,
+      "eval_loss": 0.902837336063385,
+      "eval_mean_token_accuracy": 0.828052184554968,
+      "eval_num_tokens": 7177810.0,
+      "eval_runtime": 88.1201,
+      "eval_samples_per_second": 16.114,
+      "eval_steps_per_second": 2.02,
+      "step": 3100
+    },
+    {
+      "entropy": 0.11322665181942285,
+      "epoch": 7.51901025950513,
+      "grad_norm": 0.6846067905426025,
+      "learning_rate": 6.642097297592908e-05,
+      "loss": 0.08772538304328918,
+      "mean_token_accuracy": 0.9691362135112286,
+      "num_tokens": 7225302.0,
+      "step": 3120
+    },
+    {
+      "epoch": 7.51901025950513,
+      "eval_entropy": 0.2670605613776807,
+      "eval_loss": 0.9130566120147705,
+      "eval_mean_token_accuracy": 0.826366903741708,
+      "eval_num_tokens": 7225302.0,
+      "eval_runtime": 88.6035,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3120
+    },
+    {
+      "entropy": 0.10985926580615342,
+      "epoch": 7.567290283645142,
+      "grad_norm": 0.4799748361110687,
+      "learning_rate": 6.402491332045683e-05,
+      "loss": 0.08672296404838561,
+      "mean_token_accuracy": 0.9698824048042297,
+      "num_tokens": 7274881.0,
+      "step": 3140
+    },
+    {
+      "epoch": 7.567290283645142,
+      "eval_entropy": 0.26657010866015146,
+      "eval_loss": 0.9056380987167358,
+      "eval_mean_token_accuracy": 0.8272326461384806,
+      "eval_num_tokens": 7274881.0,
+      "eval_runtime": 88.2967,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3140
+    },
+    {
+      "entropy": 0.11890473761595785,
+      "epoch": 7.615570307785154,
+      "grad_norm": 0.4081912934780121,
+      "learning_rate": 6.166397771942987e-05,
+      "loss": 0.09396827220916748,
+      "mean_token_accuracy": 0.967512384802103,
+      "num_tokens": 7321163.0,
+      "step": 3160
+    },
+    {
+      "epoch": 7.615570307785154,
+      "eval_entropy": 0.26364776575833226,
+      "eval_loss": 0.9204733967781067,
+      "eval_mean_token_accuracy": 0.827419729714983,
+      "eval_num_tokens": 7321163.0,
+      "eval_runtime": 88.3441,
+      "eval_samples_per_second": 16.074,
+      "eval_steps_per_second": 2.015,
+      "step": 3160
+    },
+    {
+      "entropy": 0.12209134344011545,
+      "epoch": 7.663850331925166,
+      "grad_norm": 0.5159774422645569,
+      "learning_rate": 5.93388342896962e-05,
+      "loss": 0.09292978644371033,
+      "mean_token_accuracy": 0.9658421002328396,
+      "num_tokens": 7366217.0,
+      "step": 3180
+    },
+    {
+      "epoch": 7.663850331925166,
+      "eval_entropy": 0.2714110068725736,
+      "eval_loss": 0.9143092036247253,
+      "eval_mean_token_accuracy": 0.8262251099843657,
+      "eval_num_tokens": 7366217.0,
+      "eval_runtime": 87.9635,
+      "eval_samples_per_second": 16.143,
+      "eval_steps_per_second": 2.024,
+      "step": 3180
+    },
+    {
+      "entropy": 0.11955609126016498,
+      "epoch": 7.712130356065178,
+      "grad_norm": 0.7005976438522339,
+      "learning_rate": 5.705014101934252e-05,
+      "loss": 0.09319776892662049,
+      "mean_token_accuracy": 0.9674513377249241,
+      "num_tokens": 7410887.0,
+      "step": 3200
+    },
+    {
+      "epoch": 7.712130356065178,
+      "eval_entropy": 0.2626291513442993,
+      "eval_loss": 0.91884446144104,
+      "eval_mean_token_accuracy": 0.827563955207889,
+      "eval_num_tokens": 7410887.0,
+      "eval_runtime": 87.8849,
+      "eval_samples_per_second": 16.157,
+      "eval_steps_per_second": 2.025,
+      "step": 3200
+    },
+    {
+      "entropy": 0.11767532899975777,
+      "epoch": 7.76041038020519,
+      "grad_norm": 0.4675362706184387,
+      "learning_rate": 5.479854558149187e-05,
+      "loss": 0.09038087725639343,
+      "mean_token_accuracy": 0.9669680051505566,
+      "num_tokens": 7457578.0,
+      "step": 3220
+    },
+    {
+      "epoch": 7.76041038020519,
+      "eval_entropy": 0.26581037203582486,
+      "eval_loss": 0.9146883487701416,
+      "eval_mean_token_accuracy": 0.8270345812433222,
+      "eval_num_tokens": 7457578.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 3220
+    },
+    {
+      "entropy": 0.11383061073720455,
+      "epoch": 7.808690404345202,
+      "grad_norm": 0.541247546672821,
+      "learning_rate": 5.258468515101918e-05,
+      "loss": 0.08942405581474304,
+      "mean_token_accuracy": 0.968398705124855,
+      "num_tokens": 7505761.0,
+      "step": 3240
+    },
+    {
+      "epoch": 7.808690404345202,
+      "eval_entropy": 0.26687998960861997,
+      "eval_loss": 0.9037085175514221,
+      "eval_mean_token_accuracy": 0.8274977863504646,
+      "eval_num_tokens": 7505761.0,
+      "eval_runtime": 88.5163,
+      "eval_samples_per_second": 16.042,
+      "eval_steps_per_second": 2.011,
+      "step": 3240
+    },
+    {
+      "entropy": 0.12004058873280883,
+      "epoch": 7.856970428485214,
+      "grad_norm": 0.5289552211761475,
+      "learning_rate": 5.040918622423899e-05,
+      "loss": 0.09064805507659912,
+      "mean_token_accuracy": 0.9673330299556255,
+      "num_tokens": 7552557.0,
+      "step": 3260
+    },
+    {
+      "epoch": 7.856970428485214,
+      "eval_entropy": 0.26288379251622085,
+      "eval_loss": 0.9146178364753723,
+      "eval_mean_token_accuracy": 0.8276662448149049,
+      "eval_num_tokens": 7552557.0,
+      "eval_runtime": 88.0723,
+      "eval_samples_per_second": 16.123,
+      "eval_steps_per_second": 2.021,
+      "step": 3260
+    },
+    {
+      "entropy": 0.1180689456872642,
+      "epoch": 7.905250452625227,
+      "grad_norm": 0.4754861891269684,
+      "learning_rate": 4.827266444161377e-05,
+      "loss": 0.09402655363082886,
+      "mean_token_accuracy": 0.9663361787796021,
+      "num_tokens": 7596998.0,
+      "step": 3280
+    },
+    {
+      "epoch": 7.905250452625227,
+      "eval_entropy": 0.261690996922134,
+      "eval_loss": 0.9120186567306519,
+      "eval_mean_token_accuracy": 0.8278995015647974,
+      "eval_num_tokens": 7596998.0,
+      "eval_runtime": 88.6573,
+      "eval_samples_per_second": 16.017,
+      "eval_steps_per_second": 2.008,
+      "step": 3280
+    },
+    {
+      "entropy": 0.1162164019420743,
+      "epoch": 7.953530476765239,
+      "grad_norm": 0.6203398108482361,
+      "learning_rate": 4.61757244135355e-05,
+      "loss": 0.08832768201828003,
+      "mean_token_accuracy": 0.9694220684468746,
+      "num_tokens": 7643288.0,
+      "step": 3300
+    },
+    {
+      "epoch": 7.953530476765239,
+      "eval_entropy": 0.26158678129817664,
+      "eval_loss": 0.9301361441612244,
+      "eval_mean_token_accuracy": 0.8269603885291668,
+      "eval_num_tokens": 7643288.0,
+      "eval_runtime": 88.7064,
+      "eval_samples_per_second": 16.008,
+      "eval_steps_per_second": 2.007,
+      "step": 3300
+    },
+    {
+      "entropy": 0.10877202204489089,
+      "epoch": 8.0,
+      "grad_norm": 3.058673143386841,
+      "learning_rate": 4.411895954922763e-05,
+      "loss": 0.09159615635871887,
+      "mean_token_accuracy": 0.9686878834451947,
+      "num_tokens": 7688608.0,
+      "step": 3320
+    },
+    {
+      "epoch": 8.0,
+      "eval_entropy": 0.2590438057197614,
+      "eval_loss": 0.9262065887451172,
+      "eval_mean_token_accuracy": 0.8277404546067956,
+      "eval_num_tokens": 7688608.0,
+      "eval_runtime": 87.7557,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 3320
+    },
+    {
+      "entropy": 0.09969092034734786,
+      "epoch": 8.048280024140013,
+      "grad_norm": 0.2561911642551422,
+      "learning_rate": 4.2102951888817644e-05,
+      "loss": 0.07228735685348511,
+      "mean_token_accuracy": 0.9743556551635265,
+      "num_tokens": 7732856.0,
+      "step": 3340
+    },
+    {
+      "epoch": 8.048280024140013,
+      "eval_entropy": 0.2462534454933713,
+      "eval_loss": 0.9755983352661133,
+      "eval_mean_token_accuracy": 0.827510209565752,
+      "eval_num_tokens": 7732856.0,
+      "eval_runtime": 88.0297,
+      "eval_samples_per_second": 16.131,
+      "eval_steps_per_second": 2.022,
+      "step": 3340
+    },
+    {
+      "entropy": 0.09477624730207027,
+      "epoch": 8.096560048280024,
+      "grad_norm": 0.3670356571674347,
+      "learning_rate": 4.012827193862633e-05,
+      "loss": 0.0686036467552185,
+      "mean_token_accuracy": 0.9743670664727688,
+      "num_tokens": 7778585.0,
+      "step": 3360
+    },
+    {
+      "epoch": 8.096560048280024,
+      "eval_entropy": 0.2440882951356052,
+      "eval_loss": 1.0118423700332642,
+      "eval_mean_token_accuracy": 0.8271241070849172,
+      "eval_num_tokens": 7778585.0,
+      "eval_runtime": 88.1364,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 3360
+    },
+    {
+      "entropy": 0.09417980071157217,
+      "epoch": 8.144840072420036,
+      "grad_norm": 0.5785111784934998,
+      "learning_rate": 3.819547850972133e-05,
+      "loss": 0.07100698947906495,
+      "mean_token_accuracy": 0.9737804099917412,
+      "num_tokens": 7824287.0,
+      "step": 3380
+    },
+    {
+      "epoch": 8.144840072420036,
+      "eval_entropy": 0.24603172860453637,
+      "eval_loss": 1.0021991729736328,
+      "eval_mean_token_accuracy": 0.827162122793412,
+      "eval_num_tokens": 7824287.0,
+      "eval_runtime": 88.4534,
+      "eval_samples_per_second": 16.054,
+      "eval_steps_per_second": 2.012,
+      "step": 3380
+    },
+    {
+      "entropy": 0.087222247896716,
+      "epoch": 8.193120096560047,
+      "grad_norm": 0.44848504662513733,
+      "learning_rate": 3.6305118559780393e-05,
+      "loss": 0.06325234174728393,
+      "mean_token_accuracy": 0.9769335232675076,
+      "num_tokens": 7874784.0,
+      "step": 3400
+    },
+    {
+      "epoch": 8.193120096560047,
+      "eval_entropy": 0.24544863905129807,
+      "eval_loss": 1.005008339881897,
+      "eval_mean_token_accuracy": 0.8273035240976998,
+      "eval_num_tokens": 7874784.0,
+      "eval_runtime": 88.6737,
+      "eval_samples_per_second": 16.014,
+      "eval_steps_per_second": 2.007,
+      "step": 3400
+    },
+    {
+      "entropy": 0.09037275062873959,
+      "epoch": 8.24140012070006,
+      "grad_norm": 0.23529155552387238,
+      "learning_rate": 3.4457727038308435e-05,
+      "loss": 0.06589569449424744,
+      "mean_token_accuracy": 0.9751470290124417,
+      "num_tokens": 7925240.0,
+      "step": 3420
+    },
+    {
+      "epoch": 8.24140012070006,
+      "eval_entropy": 0.24203423494368456,
+      "eval_loss": 1.0052824020385742,
+      "eval_mean_token_accuracy": 0.8275575503874361,
+      "eval_num_tokens": 7925240.0,
+      "eval_runtime": 88.3486,
+      "eval_samples_per_second": 16.073,
+      "eval_steps_per_second": 2.015,
+      "step": 3420
+    },
+    {
+      "entropy": 0.10373208015225828,
+      "epoch": 8.289680144840073,
+      "grad_norm": 0.24796408414840698,
+      "learning_rate": 3.2653826735253594e-05,
+      "loss": 0.07714771628379821,
+      "mean_token_accuracy": 0.9722891598939896,
+      "num_tokens": 7966685.0,
+      "step": 3440
+    },
+    {
+      "epoch": 8.289680144840073,
+      "eval_entropy": 0.23461157050025597,
+      "eval_loss": 1.012108325958252,
+      "eval_mean_token_accuracy": 0.82825900564033,
+      "eval_num_tokens": 7966685.0,
+      "eval_runtime": 88.1018,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 3440
+    },
+    {
+      "entropy": 0.08967954274266958,
+      "epoch": 8.337960168980084,
+      "grad_norm": 0.5314674973487854,
+      "learning_rate": 3.08939281330635e-05,
+      "loss": 0.06746266484260559,
+      "mean_token_accuracy": 0.9753133699297905,
+      "num_tokens": 8013656.0,
+      "step": 3460
+    },
+    {
+      "epoch": 8.337960168980084,
+      "eval_entropy": 0.23781998589467468,
+      "eval_loss": 1.0119913816452026,
+      "eval_mean_token_accuracy": 0.8273287157664139,
+      "eval_num_tokens": 8013656.0,
+      "eval_runtime": 88.6038,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3460
+    },
+    {
+      "entropy": 0.09280927330255509,
+      "epoch": 8.386240193120097,
+      "grad_norm": 0.30658936500549316,
+      "learning_rate": 2.9178529262224946e-05,
+      "loss": 0.06936421990394592,
+      "mean_token_accuracy": 0.9742105580866337,
+      "num_tokens": 8061108.0,
+      "step": 3480
+    },
+    {
+      "epoch": 8.386240193120097,
+      "eval_entropy": 0.23672426734747512,
+      "eval_loss": 1.0162041187286377,
+      "eval_mean_token_accuracy": 0.8278394281194451,
+      "eval_num_tokens": 8061108.0,
+      "eval_runtime": 88.6015,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 3480
+    },
+    {
+      "entropy": 0.09195122267119586,
+      "epoch": 8.43452021726011,
+      "grad_norm": 0.24512454867362976,
+      "learning_rate": 2.7508115560327256e-05,
+      "loss": 0.06996064782142639,
+      "mean_token_accuracy": 0.974104854464531,
+      "num_tokens": 8106924.0,
+      "step": 3500
+    },
+    {
+      "epoch": 8.43452021726011,
+      "eval_entropy": 0.23758245828781235,
+      "eval_loss": 1.0157747268676758,
+      "eval_mean_token_accuracy": 0.8278501579600773,
+      "eval_num_tokens": 8106924.0,
+      "eval_runtime": 88.5678,
+      "eval_samples_per_second": 16.033,
+      "eval_steps_per_second": 2.01,
+      "step": 3500
+    },
+    {
+      "entropy": 0.09518477590754629,
+      "epoch": 8.48280024140012,
+      "grad_norm": 0.2373228371143341,
+      "learning_rate": 2.5883159734688868e-05,
+      "loss": 0.06832797527313232,
+      "mean_token_accuracy": 0.9743806682527065,
+      "num_tokens": 8153070.0,
+      "step": 3520
+    },
+    {
+      "epoch": 8.48280024140012,
+      "eval_entropy": 0.2379308074210467,
+      "eval_loss": 1.0180530548095703,
+      "eval_mean_token_accuracy": 0.8282477052024241,
+      "eval_num_tokens": 8153070.0,
+      "eval_runtime": 88.4651,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 3520
+    },
+    {
+      "entropy": 0.0906269280705601,
+      "epoch": 8.531080265540133,
+      "grad_norm": 0.37602657079696655,
+      "learning_rate": 2.4304121628587173e-05,
+      "loss": 0.06921588778495788,
+      "mean_token_accuracy": 0.9745571479201317,
+      "num_tokens": 8200066.0,
+      "step": 3540
+    },
+    {
+      "epoch": 8.531080265540133,
+      "eval_entropy": 0.2357490769263064,
+      "eval_loss": 1.017139196395874,
+      "eval_mean_token_accuracy": 0.828713178299786,
+      "eval_num_tokens": 8200066.0,
+      "eval_runtime": 88.6424,
+      "eval_samples_per_second": 16.019,
+      "eval_steps_per_second": 2.008,
+      "step": 3540
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.466217358820516e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3560/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3560/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3560/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3560/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3560/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3560/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3560/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3560/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3560/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3560/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3560/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3560/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3560/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3560/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..c9ab7258db236e437a7d49026c4805445b161a50
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3560/trainer_state.json
@@ -0,0 +1,3772 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 8.579360289680144,
+  "eval_steps": 20,
+  "global_step": 3560,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    },
+    {
+      "entropy": 0.1326121488586068,
+      "epoch": 7.181050090525045,
+      "grad_norm": 0.4843117296695709,
+      "learning_rate": 8.411870386507724e-05,
+      "loss": 0.09537842869758606,
+      "mean_token_accuracy": 0.9645151920616627,
+      "num_tokens": 6898444.0,
+      "step": 2980
+    },
+    {
+      "epoch": 7.181050090525045,
+      "eval_entropy": 0.27999477783280813,
+      "eval_loss": 0.9031037092208862,
+      "eval_mean_token_accuracy": 0.8261882204018282,
+      "eval_num_tokens": 6898444.0,
+      "eval_runtime": 88.5602,
+      "eval_samples_per_second": 16.034,
+      "eval_steps_per_second": 2.01,
+      "step": 2980
+    },
+    {
+      "entropy": 0.1156751093454659,
+      "epoch": 7.229330114665057,
+      "grad_norm": 0.5297922492027283,
+      "learning_rate": 8.149629351430326e-05,
+      "loss": 0.09203402996063233,
+      "mean_token_accuracy": 0.9689733698964119,
+      "num_tokens": 6947106.0,
+      "step": 3000
+    },
+    {
+      "epoch": 7.229330114665057,
+      "eval_entropy": 0.2740225845508361,
+      "eval_loss": 0.8876440525054932,
+      "eval_mean_token_accuracy": 0.8265116857678703,
+      "eval_num_tokens": 6947106.0,
+      "eval_runtime": 87.7971,
+      "eval_samples_per_second": 16.174,
+      "eval_steps_per_second": 2.027,
+      "step": 3000
+    },
+    {
+      "entropy": 0.12116114404052496,
+      "epoch": 7.27761013880507,
+      "grad_norm": 0.5467398166656494,
+      "learning_rate": 7.890406302404618e-05,
+      "loss": 0.0933056652545929,
+      "mean_token_accuracy": 0.9676455594599247,
+      "num_tokens": 6992784.0,
+      "step": 3020
+    },
+    {
+      "epoch": 7.27761013880507,
+      "eval_entropy": 0.27238342793804876,
+      "eval_loss": 0.9166117310523987,
+      "eval_mean_token_accuracy": 0.825499015242866,
+      "eval_num_tokens": 6992784.0,
+      "eval_runtime": 88.3276,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 3020
+    },
+    {
+      "entropy": 0.11715119928121567,
+      "epoch": 7.325890162945082,
+      "grad_norm": 0.4976702332496643,
+      "learning_rate": 7.634274596487176e-05,
+      "loss": 0.09449659585952759,
+      "mean_token_accuracy": 0.966506478190422,
+      "num_tokens": 7040482.0,
+      "step": 3040
+    },
+    {
+      "epoch": 7.325890162945082,
+      "eval_entropy": 0.2788975556077582,
+      "eval_loss": 0.8833065629005432,
+      "eval_mean_token_accuracy": 0.8269920596915684,
+      "eval_num_tokens": 7040482.0,
+      "eval_runtime": 88.9016,
+      "eval_samples_per_second": 15.973,
+      "eval_steps_per_second": 2.002,
+      "step": 3040
+    },
+    {
+      "entropy": 0.12565911412239075,
+      "epoch": 7.3741701870850935,
+      "grad_norm": 0.5728679895401001,
+      "learning_rate": 7.381306715921025e-05,
+      "loss": 0.09626244902610778,
+      "mean_token_accuracy": 0.9661444149911403,
+      "num_tokens": 7086261.0,
+      "step": 3060
+    },
+    {
+      "epoch": 7.3741701870850935,
+      "eval_entropy": 0.2762492011939542,
+      "eval_loss": 0.8961160778999329,
+      "eval_mean_token_accuracy": 0.826369910092836,
+      "eval_num_tokens": 7086261.0,
+      "eval_runtime": 88.5204,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 3060
+    },
+    {
+      "entropy": 0.12258567921817302,
+      "epoch": 7.422450211225105,
+      "grad_norm": 0.468985915184021,
+      "learning_rate": 7.131574247624031e-05,
+      "loss": 0.09120266437530518,
+      "mean_token_accuracy": 0.9674938447773457,
+      "num_tokens": 7133247.0,
+      "step": 3080
+    },
+    {
+      "epoch": 7.422450211225105,
+      "eval_entropy": 0.2688512086366,
+      "eval_loss": 0.9048654437065125,
+      "eval_mean_token_accuracy": 0.8265582302982888,
+      "eval_num_tokens": 7133247.0,
+      "eval_runtime": 88.164,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 3080
+    },
+    {
+      "entropy": 0.12070536343380808,
+      "epoch": 7.470730235365117,
+      "grad_norm": 0.5310841798782349,
+      "learning_rate": 6.885147862930688e-05,
+      "loss": 0.09386247992515565,
+      "mean_token_accuracy": 0.96664222702384,
+      "num_tokens": 7177810.0,
+      "step": 3100
+    },
+    {
+      "epoch": 7.470730235365117,
+      "eval_entropy": 0.26289632874593305,
+      "eval_loss": 0.902837336063385,
+      "eval_mean_token_accuracy": 0.828052184554968,
+      "eval_num_tokens": 7177810.0,
+      "eval_runtime": 88.1201,
+      "eval_samples_per_second": 16.114,
+      "eval_steps_per_second": 2.02,
+      "step": 3100
+    },
+    {
+      "entropy": 0.11322665181942285,
+      "epoch": 7.51901025950513,
+      "grad_norm": 0.6846067905426025,
+      "learning_rate": 6.642097297592908e-05,
+      "loss": 0.08772538304328918,
+      "mean_token_accuracy": 0.9691362135112286,
+      "num_tokens": 7225302.0,
+      "step": 3120
+    },
+    {
+      "epoch": 7.51901025950513,
+      "eval_entropy": 0.2670605613776807,
+      "eval_loss": 0.9130566120147705,
+      "eval_mean_token_accuracy": 0.826366903741708,
+      "eval_num_tokens": 7225302.0,
+      "eval_runtime": 88.6035,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3120
+    },
+    {
+      "entropy": 0.10985926580615342,
+      "epoch": 7.567290283645142,
+      "grad_norm": 0.4799748361110687,
+      "learning_rate": 6.402491332045683e-05,
+      "loss": 0.08672296404838561,
+      "mean_token_accuracy": 0.9698824048042297,
+      "num_tokens": 7274881.0,
+      "step": 3140
+    },
+    {
+      "epoch": 7.567290283645142,
+      "eval_entropy": 0.26657010866015146,
+      "eval_loss": 0.9056380987167358,
+      "eval_mean_token_accuracy": 0.8272326461384806,
+      "eval_num_tokens": 7274881.0,
+      "eval_runtime": 88.2967,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3140
+    },
+    {
+      "entropy": 0.11890473761595785,
+      "epoch": 7.615570307785154,
+      "grad_norm": 0.4081912934780121,
+      "learning_rate": 6.166397771942987e-05,
+      "loss": 0.09396827220916748,
+      "mean_token_accuracy": 0.967512384802103,
+      "num_tokens": 7321163.0,
+      "step": 3160
+    },
+    {
+      "epoch": 7.615570307785154,
+      "eval_entropy": 0.26364776575833226,
+      "eval_loss": 0.9204733967781067,
+      "eval_mean_token_accuracy": 0.827419729714983,
+      "eval_num_tokens": 7321163.0,
+      "eval_runtime": 88.3441,
+      "eval_samples_per_second": 16.074,
+      "eval_steps_per_second": 2.015,
+      "step": 3160
+    },
+    {
+      "entropy": 0.12209134344011545,
+      "epoch": 7.663850331925166,
+      "grad_norm": 0.5159774422645569,
+      "learning_rate": 5.93388342896962e-05,
+      "loss": 0.09292978644371033,
+      "mean_token_accuracy": 0.9658421002328396,
+      "num_tokens": 7366217.0,
+      "step": 3180
+    },
+    {
+      "epoch": 7.663850331925166,
+      "eval_entropy": 0.2714110068725736,
+      "eval_loss": 0.9143092036247253,
+      "eval_mean_token_accuracy": 0.8262251099843657,
+      "eval_num_tokens": 7366217.0,
+      "eval_runtime": 87.9635,
+      "eval_samples_per_second": 16.143,
+      "eval_steps_per_second": 2.024,
+      "step": 3180
+    },
+    {
+      "entropy": 0.11955609126016498,
+      "epoch": 7.712130356065178,
+      "grad_norm": 0.7005976438522339,
+      "learning_rate": 5.705014101934252e-05,
+      "loss": 0.09319776892662049,
+      "mean_token_accuracy": 0.9674513377249241,
+      "num_tokens": 7410887.0,
+      "step": 3200
+    },
+    {
+      "epoch": 7.712130356065178,
+      "eval_entropy": 0.2626291513442993,
+      "eval_loss": 0.91884446144104,
+      "eval_mean_token_accuracy": 0.827563955207889,
+      "eval_num_tokens": 7410887.0,
+      "eval_runtime": 87.8849,
+      "eval_samples_per_second": 16.157,
+      "eval_steps_per_second": 2.025,
+      "step": 3200
+    },
+    {
+      "entropy": 0.11767532899975777,
+      "epoch": 7.76041038020519,
+      "grad_norm": 0.4675362706184387,
+      "learning_rate": 5.479854558149187e-05,
+      "loss": 0.09038087725639343,
+      "mean_token_accuracy": 0.9669680051505566,
+      "num_tokens": 7457578.0,
+      "step": 3220
+    },
+    {
+      "epoch": 7.76041038020519,
+      "eval_entropy": 0.26581037203582486,
+      "eval_loss": 0.9146883487701416,
+      "eval_mean_token_accuracy": 0.8270345812433222,
+      "eval_num_tokens": 7457578.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 3220
+    },
+    {
+      "entropy": 0.11383061073720455,
+      "epoch": 7.808690404345202,
+      "grad_norm": 0.541247546672821,
+      "learning_rate": 5.258468515101918e-05,
+      "loss": 0.08942405581474304,
+      "mean_token_accuracy": 0.968398705124855,
+      "num_tokens": 7505761.0,
+      "step": 3240
+    },
+    {
+      "epoch": 7.808690404345202,
+      "eval_entropy": 0.26687998960861997,
+      "eval_loss": 0.9037085175514221,
+      "eval_mean_token_accuracy": 0.8274977863504646,
+      "eval_num_tokens": 7505761.0,
+      "eval_runtime": 88.5163,
+      "eval_samples_per_second": 16.042,
+      "eval_steps_per_second": 2.011,
+      "step": 3240
+    },
+    {
+      "entropy": 0.12004058873280883,
+      "epoch": 7.856970428485214,
+      "grad_norm": 0.5289552211761475,
+      "learning_rate": 5.040918622423899e-05,
+      "loss": 0.09064805507659912,
+      "mean_token_accuracy": 0.9673330299556255,
+      "num_tokens": 7552557.0,
+      "step": 3260
+    },
+    {
+      "epoch": 7.856970428485214,
+      "eval_entropy": 0.26288379251622085,
+      "eval_loss": 0.9146178364753723,
+      "eval_mean_token_accuracy": 0.8276662448149049,
+      "eval_num_tokens": 7552557.0,
+      "eval_runtime": 88.0723,
+      "eval_samples_per_second": 16.123,
+      "eval_steps_per_second": 2.021,
+      "step": 3260
+    },
+    {
+      "entropy": 0.1180689456872642,
+      "epoch": 7.905250452625227,
+      "grad_norm": 0.4754861891269684,
+      "learning_rate": 4.827266444161377e-05,
+      "loss": 0.09402655363082886,
+      "mean_token_accuracy": 0.9663361787796021,
+      "num_tokens": 7596998.0,
+      "step": 3280
+    },
+    {
+      "epoch": 7.905250452625227,
+      "eval_entropy": 0.261690996922134,
+      "eval_loss": 0.9120186567306519,
+      "eval_mean_token_accuracy": 0.8278995015647974,
+      "eval_num_tokens": 7596998.0,
+      "eval_runtime": 88.6573,
+      "eval_samples_per_second": 16.017,
+      "eval_steps_per_second": 2.008,
+      "step": 3280
+    },
+    {
+      "entropy": 0.1162164019420743,
+      "epoch": 7.953530476765239,
+      "grad_norm": 0.6203398108482361,
+      "learning_rate": 4.61757244135355e-05,
+      "loss": 0.08832768201828003,
+      "mean_token_accuracy": 0.9694220684468746,
+      "num_tokens": 7643288.0,
+      "step": 3300
+    },
+    {
+      "epoch": 7.953530476765239,
+      "eval_entropy": 0.26158678129817664,
+      "eval_loss": 0.9301361441612244,
+      "eval_mean_token_accuracy": 0.8269603885291668,
+      "eval_num_tokens": 7643288.0,
+      "eval_runtime": 88.7064,
+      "eval_samples_per_second": 16.008,
+      "eval_steps_per_second": 2.007,
+      "step": 3300
+    },
+    {
+      "entropy": 0.10877202204489089,
+      "epoch": 8.0,
+      "grad_norm": 3.058673143386841,
+      "learning_rate": 4.411895954922763e-05,
+      "loss": 0.09159615635871887,
+      "mean_token_accuracy": 0.9686878834451947,
+      "num_tokens": 7688608.0,
+      "step": 3320
+    },
+    {
+      "epoch": 8.0,
+      "eval_entropy": 0.2590438057197614,
+      "eval_loss": 0.9262065887451172,
+      "eval_mean_token_accuracy": 0.8277404546067956,
+      "eval_num_tokens": 7688608.0,
+      "eval_runtime": 87.7557,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 3320
+    },
+    {
+      "entropy": 0.09969092034734786,
+      "epoch": 8.048280024140013,
+      "grad_norm": 0.2561911642551422,
+      "learning_rate": 4.2102951888817644e-05,
+      "loss": 0.07228735685348511,
+      "mean_token_accuracy": 0.9743556551635265,
+      "num_tokens": 7732856.0,
+      "step": 3340
+    },
+    {
+      "epoch": 8.048280024140013,
+      "eval_entropy": 0.2462534454933713,
+      "eval_loss": 0.9755983352661133,
+      "eval_mean_token_accuracy": 0.827510209565752,
+      "eval_num_tokens": 7732856.0,
+      "eval_runtime": 88.0297,
+      "eval_samples_per_second": 16.131,
+      "eval_steps_per_second": 2.022,
+      "step": 3340
+    },
+    {
+      "entropy": 0.09477624730207027,
+      "epoch": 8.096560048280024,
+      "grad_norm": 0.3670356571674347,
+      "learning_rate": 4.012827193862633e-05,
+      "loss": 0.0686036467552185,
+      "mean_token_accuracy": 0.9743670664727688,
+      "num_tokens": 7778585.0,
+      "step": 3360
+    },
+    {
+      "epoch": 8.096560048280024,
+      "eval_entropy": 0.2440882951356052,
+      "eval_loss": 1.0118423700332642,
+      "eval_mean_token_accuracy": 0.8271241070849172,
+      "eval_num_tokens": 7778585.0,
+      "eval_runtime": 88.1364,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 3360
+    },
+    {
+      "entropy": 0.09417980071157217,
+      "epoch": 8.144840072420036,
+      "grad_norm": 0.5785111784934998,
+      "learning_rate": 3.819547850972133e-05,
+      "loss": 0.07100698947906495,
+      "mean_token_accuracy": 0.9737804099917412,
+      "num_tokens": 7824287.0,
+      "step": 3380
+    },
+    {
+      "epoch": 8.144840072420036,
+      "eval_entropy": 0.24603172860453637,
+      "eval_loss": 1.0021991729736328,
+      "eval_mean_token_accuracy": 0.827162122793412,
+      "eval_num_tokens": 7824287.0,
+      "eval_runtime": 88.4534,
+      "eval_samples_per_second": 16.054,
+      "eval_steps_per_second": 2.012,
+      "step": 3380
+    },
+    {
+      "entropy": 0.087222247896716,
+      "epoch": 8.193120096560047,
+      "grad_norm": 0.44848504662513733,
+      "learning_rate": 3.6305118559780393e-05,
+      "loss": 0.06325234174728393,
+      "mean_token_accuracy": 0.9769335232675076,
+      "num_tokens": 7874784.0,
+      "step": 3400
+    },
+    {
+      "epoch": 8.193120096560047,
+      "eval_entropy": 0.24544863905129807,
+      "eval_loss": 1.005008339881897,
+      "eval_mean_token_accuracy": 0.8273035240976998,
+      "eval_num_tokens": 7874784.0,
+      "eval_runtime": 88.6737,
+      "eval_samples_per_second": 16.014,
+      "eval_steps_per_second": 2.007,
+      "step": 3400
+    },
+    {
+      "entropy": 0.09037275062873959,
+      "epoch": 8.24140012070006,
+      "grad_norm": 0.23529155552387238,
+      "learning_rate": 3.4457727038308435e-05,
+      "loss": 0.06589569449424744,
+      "mean_token_accuracy": 0.9751470290124417,
+      "num_tokens": 7925240.0,
+      "step": 3420
+    },
+    {
+      "epoch": 8.24140012070006,
+      "eval_entropy": 0.24203423494368456,
+      "eval_loss": 1.0052824020385742,
+      "eval_mean_token_accuracy": 0.8275575503874361,
+      "eval_num_tokens": 7925240.0,
+      "eval_runtime": 88.3486,
+      "eval_samples_per_second": 16.073,
+      "eval_steps_per_second": 2.015,
+      "step": 3420
+    },
+    {
+      "entropy": 0.10373208015225828,
+      "epoch": 8.289680144840073,
+      "grad_norm": 0.24796408414840698,
+      "learning_rate": 3.2653826735253594e-05,
+      "loss": 0.07714771628379821,
+      "mean_token_accuracy": 0.9722891598939896,
+      "num_tokens": 7966685.0,
+      "step": 3440
+    },
+    {
+      "epoch": 8.289680144840073,
+      "eval_entropy": 0.23461157050025597,
+      "eval_loss": 1.012108325958252,
+      "eval_mean_token_accuracy": 0.82825900564033,
+      "eval_num_tokens": 7966685.0,
+      "eval_runtime": 88.1018,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 3440
+    },
+    {
+      "entropy": 0.08967954274266958,
+      "epoch": 8.337960168980084,
+      "grad_norm": 0.5314674973487854,
+      "learning_rate": 3.08939281330635e-05,
+      "loss": 0.06746266484260559,
+      "mean_token_accuracy": 0.9753133699297905,
+      "num_tokens": 8013656.0,
+      "step": 3460
+    },
+    {
+      "epoch": 8.337960168980084,
+      "eval_entropy": 0.23781998589467468,
+      "eval_loss": 1.0119913816452026,
+      "eval_mean_token_accuracy": 0.8273287157664139,
+      "eval_num_tokens": 8013656.0,
+      "eval_runtime": 88.6038,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3460
+    },
+    {
+      "entropy": 0.09280927330255509,
+      "epoch": 8.386240193120097,
+      "grad_norm": 0.30658936500549316,
+      "learning_rate": 2.9178529262224946e-05,
+      "loss": 0.06936421990394592,
+      "mean_token_accuracy": 0.9742105580866337,
+      "num_tokens": 8061108.0,
+      "step": 3480
+    },
+    {
+      "epoch": 8.386240193120097,
+      "eval_entropy": 0.23672426734747512,
+      "eval_loss": 1.0162041187286377,
+      "eval_mean_token_accuracy": 0.8278394281194451,
+      "eval_num_tokens": 8061108.0,
+      "eval_runtime": 88.6015,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 3480
+    },
+    {
+      "entropy": 0.09195122267119586,
+      "epoch": 8.43452021726011,
+      "grad_norm": 0.24512454867362976,
+      "learning_rate": 2.7508115560327256e-05,
+      "loss": 0.06996064782142639,
+      "mean_token_accuracy": 0.974104854464531,
+      "num_tokens": 8106924.0,
+      "step": 3500
+    },
+    {
+      "epoch": 8.43452021726011,
+      "eval_entropy": 0.23758245828781235,
+      "eval_loss": 1.0157747268676758,
+      "eval_mean_token_accuracy": 0.8278501579600773,
+      "eval_num_tokens": 8106924.0,
+      "eval_runtime": 88.5678,
+      "eval_samples_per_second": 16.033,
+      "eval_steps_per_second": 2.01,
+      "step": 3500
+    },
+    {
+      "entropy": 0.09518477590754629,
+      "epoch": 8.48280024140012,
+      "grad_norm": 0.2373228371143341,
+      "learning_rate": 2.5883159734688868e-05,
+      "loss": 0.06832797527313232,
+      "mean_token_accuracy": 0.9743806682527065,
+      "num_tokens": 8153070.0,
+      "step": 3520
+    },
+    {
+      "epoch": 8.48280024140012,
+      "eval_entropy": 0.2379308074210467,
+      "eval_loss": 1.0180530548095703,
+      "eval_mean_token_accuracy": 0.8282477052024241,
+      "eval_num_tokens": 8153070.0,
+      "eval_runtime": 88.4651,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 3520
+    },
+    {
+      "entropy": 0.0906269280705601,
+      "epoch": 8.531080265540133,
+      "grad_norm": 0.37602657079696655,
+      "learning_rate": 2.4304121628587173e-05,
+      "loss": 0.06921588778495788,
+      "mean_token_accuracy": 0.9745571479201317,
+      "num_tokens": 8200066.0,
+      "step": 3540
+    },
+    {
+      "epoch": 8.531080265540133,
+      "eval_entropy": 0.2357490769263064,
+      "eval_loss": 1.017139196395874,
+      "eval_mean_token_accuracy": 0.828713178299786,
+      "eval_num_tokens": 8200066.0,
+      "eval_runtime": 88.6424,
+      "eval_samples_per_second": 16.019,
+      "eval_steps_per_second": 2.008,
+      "step": 3540
+    },
+    {
+      "entropy": 0.09273334196768701,
+      "epoch": 8.579360289680144,
+      "grad_norm": 0.2978838086128235,
+      "learning_rate": 2.277144809112794e-05,
+      "loss": 0.06970887780189514,
+      "mean_token_accuracy": 0.9751374177634716,
+      "num_tokens": 8246157.0,
+      "step": 3560
+    },
+    {
+      "epoch": 8.579360289680144,
+      "eval_entropy": 0.23590933138065123,
+      "eval_loss": 1.019231915473938,
+      "eval_mean_token_accuracy": 0.8279163412163767,
+      "eval_num_tokens": 8246157.0,
+      "eval_runtime": 88.2691,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 3560
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.4858621816151654e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3580/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3580/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3580/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3580/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3580/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3580/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3580/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3580/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3580/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3580/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3580/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3580/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3580/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3580/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..bce0bdf22b40c7a2adba658254ce46eebe4bf201
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3580/trainer_state.json
@@ -0,0 +1,3793 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 8.627640313820157,
+  "eval_steps": 20,
+  "global_step": 3580,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    },
+    {
+      "entropy": 0.1326121488586068,
+      "epoch": 7.181050090525045,
+      "grad_norm": 0.4843117296695709,
+      "learning_rate": 8.411870386507724e-05,
+      "loss": 0.09537842869758606,
+      "mean_token_accuracy": 0.9645151920616627,
+      "num_tokens": 6898444.0,
+      "step": 2980
+    },
+    {
+      "epoch": 7.181050090525045,
+      "eval_entropy": 0.27999477783280813,
+      "eval_loss": 0.9031037092208862,
+      "eval_mean_token_accuracy": 0.8261882204018282,
+      "eval_num_tokens": 6898444.0,
+      "eval_runtime": 88.5602,
+      "eval_samples_per_second": 16.034,
+      "eval_steps_per_second": 2.01,
+      "step": 2980
+    },
+    {
+      "entropy": 0.1156751093454659,
+      "epoch": 7.229330114665057,
+      "grad_norm": 0.5297922492027283,
+      "learning_rate": 8.149629351430326e-05,
+      "loss": 0.09203402996063233,
+      "mean_token_accuracy": 0.9689733698964119,
+      "num_tokens": 6947106.0,
+      "step": 3000
+    },
+    {
+      "epoch": 7.229330114665057,
+      "eval_entropy": 0.2740225845508361,
+      "eval_loss": 0.8876440525054932,
+      "eval_mean_token_accuracy": 0.8265116857678703,
+      "eval_num_tokens": 6947106.0,
+      "eval_runtime": 87.7971,
+      "eval_samples_per_second": 16.174,
+      "eval_steps_per_second": 2.027,
+      "step": 3000
+    },
+    {
+      "entropy": 0.12116114404052496,
+      "epoch": 7.27761013880507,
+      "grad_norm": 0.5467398166656494,
+      "learning_rate": 7.890406302404618e-05,
+      "loss": 0.0933056652545929,
+      "mean_token_accuracy": 0.9676455594599247,
+      "num_tokens": 6992784.0,
+      "step": 3020
+    },
+    {
+      "epoch": 7.27761013880507,
+      "eval_entropy": 0.27238342793804876,
+      "eval_loss": 0.9166117310523987,
+      "eval_mean_token_accuracy": 0.825499015242866,
+      "eval_num_tokens": 6992784.0,
+      "eval_runtime": 88.3276,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 3020
+    },
+    {
+      "entropy": 0.11715119928121567,
+      "epoch": 7.325890162945082,
+      "grad_norm": 0.4976702332496643,
+      "learning_rate": 7.634274596487176e-05,
+      "loss": 0.09449659585952759,
+      "mean_token_accuracy": 0.966506478190422,
+      "num_tokens": 7040482.0,
+      "step": 3040
+    },
+    {
+      "epoch": 7.325890162945082,
+      "eval_entropy": 0.2788975556077582,
+      "eval_loss": 0.8833065629005432,
+      "eval_mean_token_accuracy": 0.8269920596915684,
+      "eval_num_tokens": 7040482.0,
+      "eval_runtime": 88.9016,
+      "eval_samples_per_second": 15.973,
+      "eval_steps_per_second": 2.002,
+      "step": 3040
+    },
+    {
+      "entropy": 0.12565911412239075,
+      "epoch": 7.3741701870850935,
+      "grad_norm": 0.5728679895401001,
+      "learning_rate": 7.381306715921025e-05,
+      "loss": 0.09626244902610778,
+      "mean_token_accuracy": 0.9661444149911403,
+      "num_tokens": 7086261.0,
+      "step": 3060
+    },
+    {
+      "epoch": 7.3741701870850935,
+      "eval_entropy": 0.2762492011939542,
+      "eval_loss": 0.8961160778999329,
+      "eval_mean_token_accuracy": 0.826369910092836,
+      "eval_num_tokens": 7086261.0,
+      "eval_runtime": 88.5204,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 3060
+    },
+    {
+      "entropy": 0.12258567921817302,
+      "epoch": 7.422450211225105,
+      "grad_norm": 0.468985915184021,
+      "learning_rate": 7.131574247624031e-05,
+      "loss": 0.09120266437530518,
+      "mean_token_accuracy": 0.9674938447773457,
+      "num_tokens": 7133247.0,
+      "step": 3080
+    },
+    {
+      "epoch": 7.422450211225105,
+      "eval_entropy": 0.2688512086366,
+      "eval_loss": 0.9048654437065125,
+      "eval_mean_token_accuracy": 0.8265582302982888,
+      "eval_num_tokens": 7133247.0,
+      "eval_runtime": 88.164,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 3080
+    },
+    {
+      "entropy": 0.12070536343380808,
+      "epoch": 7.470730235365117,
+      "grad_norm": 0.5310841798782349,
+      "learning_rate": 6.885147862930688e-05,
+      "loss": 0.09386247992515565,
+      "mean_token_accuracy": 0.96664222702384,
+      "num_tokens": 7177810.0,
+      "step": 3100
+    },
+    {
+      "epoch": 7.470730235365117,
+      "eval_entropy": 0.26289632874593305,
+      "eval_loss": 0.902837336063385,
+      "eval_mean_token_accuracy": 0.828052184554968,
+      "eval_num_tokens": 7177810.0,
+      "eval_runtime": 88.1201,
+      "eval_samples_per_second": 16.114,
+      "eval_steps_per_second": 2.02,
+      "step": 3100
+    },
+    {
+      "entropy": 0.11322665181942285,
+      "epoch": 7.51901025950513,
+      "grad_norm": 0.6846067905426025,
+      "learning_rate": 6.642097297592908e-05,
+      "loss": 0.08772538304328918,
+      "mean_token_accuracy": 0.9691362135112286,
+      "num_tokens": 7225302.0,
+      "step": 3120
+    },
+    {
+      "epoch": 7.51901025950513,
+      "eval_entropy": 0.2670605613776807,
+      "eval_loss": 0.9130566120147705,
+      "eval_mean_token_accuracy": 0.826366903741708,
+      "eval_num_tokens": 7225302.0,
+      "eval_runtime": 88.6035,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3120
+    },
+    {
+      "entropy": 0.10985926580615342,
+      "epoch": 7.567290283645142,
+      "grad_norm": 0.4799748361110687,
+      "learning_rate": 6.402491332045683e-05,
+      "loss": 0.08672296404838561,
+      "mean_token_accuracy": 0.9698824048042297,
+      "num_tokens": 7274881.0,
+      "step": 3140
+    },
+    {
+      "epoch": 7.567290283645142,
+      "eval_entropy": 0.26657010866015146,
+      "eval_loss": 0.9056380987167358,
+      "eval_mean_token_accuracy": 0.8272326461384806,
+      "eval_num_tokens": 7274881.0,
+      "eval_runtime": 88.2967,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3140
+    },
+    {
+      "entropy": 0.11890473761595785,
+      "epoch": 7.615570307785154,
+      "grad_norm": 0.4081912934780121,
+      "learning_rate": 6.166397771942987e-05,
+      "loss": 0.09396827220916748,
+      "mean_token_accuracy": 0.967512384802103,
+      "num_tokens": 7321163.0,
+      "step": 3160
+    },
+    {
+      "epoch": 7.615570307785154,
+      "eval_entropy": 0.26364776575833226,
+      "eval_loss": 0.9204733967781067,
+      "eval_mean_token_accuracy": 0.827419729714983,
+      "eval_num_tokens": 7321163.0,
+      "eval_runtime": 88.3441,
+      "eval_samples_per_second": 16.074,
+      "eval_steps_per_second": 2.015,
+      "step": 3160
+    },
+    {
+      "entropy": 0.12209134344011545,
+      "epoch": 7.663850331925166,
+      "grad_norm": 0.5159774422645569,
+      "learning_rate": 5.93388342896962e-05,
+      "loss": 0.09292978644371033,
+      "mean_token_accuracy": 0.9658421002328396,
+      "num_tokens": 7366217.0,
+      "step": 3180
+    },
+    {
+      "epoch": 7.663850331925166,
+      "eval_entropy": 0.2714110068725736,
+      "eval_loss": 0.9143092036247253,
+      "eval_mean_token_accuracy": 0.8262251099843657,
+      "eval_num_tokens": 7366217.0,
+      "eval_runtime": 87.9635,
+      "eval_samples_per_second": 16.143,
+      "eval_steps_per_second": 2.024,
+      "step": 3180
+    },
+    {
+      "entropy": 0.11955609126016498,
+      "epoch": 7.712130356065178,
+      "grad_norm": 0.7005976438522339,
+      "learning_rate": 5.705014101934252e-05,
+      "loss": 0.09319776892662049,
+      "mean_token_accuracy": 0.9674513377249241,
+      "num_tokens": 7410887.0,
+      "step": 3200
+    },
+    {
+      "epoch": 7.712130356065178,
+      "eval_entropy": 0.2626291513442993,
+      "eval_loss": 0.91884446144104,
+      "eval_mean_token_accuracy": 0.827563955207889,
+      "eval_num_tokens": 7410887.0,
+      "eval_runtime": 87.8849,
+      "eval_samples_per_second": 16.157,
+      "eval_steps_per_second": 2.025,
+      "step": 3200
+    },
+    {
+      "entropy": 0.11767532899975777,
+      "epoch": 7.76041038020519,
+      "grad_norm": 0.4675362706184387,
+      "learning_rate": 5.479854558149187e-05,
+      "loss": 0.09038087725639343,
+      "mean_token_accuracy": 0.9669680051505566,
+      "num_tokens": 7457578.0,
+      "step": 3220
+    },
+    {
+      "epoch": 7.76041038020519,
+      "eval_entropy": 0.26581037203582486,
+      "eval_loss": 0.9146883487701416,
+      "eval_mean_token_accuracy": 0.8270345812433222,
+      "eval_num_tokens": 7457578.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 3220
+    },
+    {
+      "entropy": 0.11383061073720455,
+      "epoch": 7.808690404345202,
+      "grad_norm": 0.541247546672821,
+      "learning_rate": 5.258468515101918e-05,
+      "loss": 0.08942405581474304,
+      "mean_token_accuracy": 0.968398705124855,
+      "num_tokens": 7505761.0,
+      "step": 3240
+    },
+    {
+      "epoch": 7.808690404345202,
+      "eval_entropy": 0.26687998960861997,
+      "eval_loss": 0.9037085175514221,
+      "eval_mean_token_accuracy": 0.8274977863504646,
+      "eval_num_tokens": 7505761.0,
+      "eval_runtime": 88.5163,
+      "eval_samples_per_second": 16.042,
+      "eval_steps_per_second": 2.011,
+      "step": 3240
+    },
+    {
+      "entropy": 0.12004058873280883,
+      "epoch": 7.856970428485214,
+      "grad_norm": 0.5289552211761475,
+      "learning_rate": 5.040918622423899e-05,
+      "loss": 0.09064805507659912,
+      "mean_token_accuracy": 0.9673330299556255,
+      "num_tokens": 7552557.0,
+      "step": 3260
+    },
+    {
+      "epoch": 7.856970428485214,
+      "eval_entropy": 0.26288379251622085,
+      "eval_loss": 0.9146178364753723,
+      "eval_mean_token_accuracy": 0.8276662448149049,
+      "eval_num_tokens": 7552557.0,
+      "eval_runtime": 88.0723,
+      "eval_samples_per_second": 16.123,
+      "eval_steps_per_second": 2.021,
+      "step": 3260
+    },
+    {
+      "entropy": 0.1180689456872642,
+      "epoch": 7.905250452625227,
+      "grad_norm": 0.4754861891269684,
+      "learning_rate": 4.827266444161377e-05,
+      "loss": 0.09402655363082886,
+      "mean_token_accuracy": 0.9663361787796021,
+      "num_tokens": 7596998.0,
+      "step": 3280
+    },
+    {
+      "epoch": 7.905250452625227,
+      "eval_entropy": 0.261690996922134,
+      "eval_loss": 0.9120186567306519,
+      "eval_mean_token_accuracy": 0.8278995015647974,
+      "eval_num_tokens": 7596998.0,
+      "eval_runtime": 88.6573,
+      "eval_samples_per_second": 16.017,
+      "eval_steps_per_second": 2.008,
+      "step": 3280
+    },
+    {
+      "entropy": 0.1162164019420743,
+      "epoch": 7.953530476765239,
+      "grad_norm": 0.6203398108482361,
+      "learning_rate": 4.61757244135355e-05,
+      "loss": 0.08832768201828003,
+      "mean_token_accuracy": 0.9694220684468746,
+      "num_tokens": 7643288.0,
+      "step": 3300
+    },
+    {
+      "epoch": 7.953530476765239,
+      "eval_entropy": 0.26158678129817664,
+      "eval_loss": 0.9301361441612244,
+      "eval_mean_token_accuracy": 0.8269603885291668,
+      "eval_num_tokens": 7643288.0,
+      "eval_runtime": 88.7064,
+      "eval_samples_per_second": 16.008,
+      "eval_steps_per_second": 2.007,
+      "step": 3300
+    },
+    {
+      "entropy": 0.10877202204489089,
+      "epoch": 8.0,
+      "grad_norm": 3.058673143386841,
+      "learning_rate": 4.411895954922763e-05,
+      "loss": 0.09159615635871887,
+      "mean_token_accuracy": 0.9686878834451947,
+      "num_tokens": 7688608.0,
+      "step": 3320
+    },
+    {
+      "epoch": 8.0,
+      "eval_entropy": 0.2590438057197614,
+      "eval_loss": 0.9262065887451172,
+      "eval_mean_token_accuracy": 0.8277404546067956,
+      "eval_num_tokens": 7688608.0,
+      "eval_runtime": 87.7557,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 3320
+    },
+    {
+      "entropy": 0.09969092034734786,
+      "epoch": 8.048280024140013,
+      "grad_norm": 0.2561911642551422,
+      "learning_rate": 4.2102951888817644e-05,
+      "loss": 0.07228735685348511,
+      "mean_token_accuracy": 0.9743556551635265,
+      "num_tokens": 7732856.0,
+      "step": 3340
+    },
+    {
+      "epoch": 8.048280024140013,
+      "eval_entropy": 0.2462534454933713,
+      "eval_loss": 0.9755983352661133,
+      "eval_mean_token_accuracy": 0.827510209565752,
+      "eval_num_tokens": 7732856.0,
+      "eval_runtime": 88.0297,
+      "eval_samples_per_second": 16.131,
+      "eval_steps_per_second": 2.022,
+      "step": 3340
+    },
+    {
+      "entropy": 0.09477624730207027,
+      "epoch": 8.096560048280024,
+      "grad_norm": 0.3670356571674347,
+      "learning_rate": 4.012827193862633e-05,
+      "loss": 0.0686036467552185,
+      "mean_token_accuracy": 0.9743670664727688,
+      "num_tokens": 7778585.0,
+      "step": 3360
+    },
+    {
+      "epoch": 8.096560048280024,
+      "eval_entropy": 0.2440882951356052,
+      "eval_loss": 1.0118423700332642,
+      "eval_mean_token_accuracy": 0.8271241070849172,
+      "eval_num_tokens": 7778585.0,
+      "eval_runtime": 88.1364,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 3360
+    },
+    {
+      "entropy": 0.09417980071157217,
+      "epoch": 8.144840072420036,
+      "grad_norm": 0.5785111784934998,
+      "learning_rate": 3.819547850972133e-05,
+      "loss": 0.07100698947906495,
+      "mean_token_accuracy": 0.9737804099917412,
+      "num_tokens": 7824287.0,
+      "step": 3380
+    },
+    {
+      "epoch": 8.144840072420036,
+      "eval_entropy": 0.24603172860453637,
+      "eval_loss": 1.0021991729736328,
+      "eval_mean_token_accuracy": 0.827162122793412,
+      "eval_num_tokens": 7824287.0,
+      "eval_runtime": 88.4534,
+      "eval_samples_per_second": 16.054,
+      "eval_steps_per_second": 2.012,
+      "step": 3380
+    },
+    {
+      "entropy": 0.087222247896716,
+      "epoch": 8.193120096560047,
+      "grad_norm": 0.44848504662513733,
+      "learning_rate": 3.6305118559780393e-05,
+      "loss": 0.06325234174728393,
+      "mean_token_accuracy": 0.9769335232675076,
+      "num_tokens": 7874784.0,
+      "step": 3400
+    },
+    {
+      "epoch": 8.193120096560047,
+      "eval_entropy": 0.24544863905129807,
+      "eval_loss": 1.005008339881897,
+      "eval_mean_token_accuracy": 0.8273035240976998,
+      "eval_num_tokens": 7874784.0,
+      "eval_runtime": 88.6737,
+      "eval_samples_per_second": 16.014,
+      "eval_steps_per_second": 2.007,
+      "step": 3400
+    },
+    {
+      "entropy": 0.09037275062873959,
+      "epoch": 8.24140012070006,
+      "grad_norm": 0.23529155552387238,
+      "learning_rate": 3.4457727038308435e-05,
+      "loss": 0.06589569449424744,
+      "mean_token_accuracy": 0.9751470290124417,
+      "num_tokens": 7925240.0,
+      "step": 3420
+    },
+    {
+      "epoch": 8.24140012070006,
+      "eval_entropy": 0.24203423494368456,
+      "eval_loss": 1.0052824020385742,
+      "eval_mean_token_accuracy": 0.8275575503874361,
+      "eval_num_tokens": 7925240.0,
+      "eval_runtime": 88.3486,
+      "eval_samples_per_second": 16.073,
+      "eval_steps_per_second": 2.015,
+      "step": 3420
+    },
+    {
+      "entropy": 0.10373208015225828,
+      "epoch": 8.289680144840073,
+      "grad_norm": 0.24796408414840698,
+      "learning_rate": 3.2653826735253594e-05,
+      "loss": 0.07714771628379821,
+      "mean_token_accuracy": 0.9722891598939896,
+      "num_tokens": 7966685.0,
+      "step": 3440
+    },
+    {
+      "epoch": 8.289680144840073,
+      "eval_entropy": 0.23461157050025597,
+      "eval_loss": 1.012108325958252,
+      "eval_mean_token_accuracy": 0.82825900564033,
+      "eval_num_tokens": 7966685.0,
+      "eval_runtime": 88.1018,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 3440
+    },
+    {
+      "entropy": 0.08967954274266958,
+      "epoch": 8.337960168980084,
+      "grad_norm": 0.5314674973487854,
+      "learning_rate": 3.08939281330635e-05,
+      "loss": 0.06746266484260559,
+      "mean_token_accuracy": 0.9753133699297905,
+      "num_tokens": 8013656.0,
+      "step": 3460
+    },
+    {
+      "epoch": 8.337960168980084,
+      "eval_entropy": 0.23781998589467468,
+      "eval_loss": 1.0119913816452026,
+      "eval_mean_token_accuracy": 0.8273287157664139,
+      "eval_num_tokens": 8013656.0,
+      "eval_runtime": 88.6038,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3460
+    },
+    {
+      "entropy": 0.09280927330255509,
+      "epoch": 8.386240193120097,
+      "grad_norm": 0.30658936500549316,
+      "learning_rate": 2.9178529262224946e-05,
+      "loss": 0.06936421990394592,
+      "mean_token_accuracy": 0.9742105580866337,
+      "num_tokens": 8061108.0,
+      "step": 3480
+    },
+    {
+      "epoch": 8.386240193120097,
+      "eval_entropy": 0.23672426734747512,
+      "eval_loss": 1.0162041187286377,
+      "eval_mean_token_accuracy": 0.8278394281194451,
+      "eval_num_tokens": 8061108.0,
+      "eval_runtime": 88.6015,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 3480
+    },
+    {
+      "entropy": 0.09195122267119586,
+      "epoch": 8.43452021726011,
+      "grad_norm": 0.24512454867362976,
+      "learning_rate": 2.7508115560327256e-05,
+      "loss": 0.06996064782142639,
+      "mean_token_accuracy": 0.974104854464531,
+      "num_tokens": 8106924.0,
+      "step": 3500
+    },
+    {
+      "epoch": 8.43452021726011,
+      "eval_entropy": 0.23758245828781235,
+      "eval_loss": 1.0157747268676758,
+      "eval_mean_token_accuracy": 0.8278501579600773,
+      "eval_num_tokens": 8106924.0,
+      "eval_runtime": 88.5678,
+      "eval_samples_per_second": 16.033,
+      "eval_steps_per_second": 2.01,
+      "step": 3500
+    },
+    {
+      "entropy": 0.09518477590754629,
+      "epoch": 8.48280024140012,
+      "grad_norm": 0.2373228371143341,
+      "learning_rate": 2.5883159734688868e-05,
+      "loss": 0.06832797527313232,
+      "mean_token_accuracy": 0.9743806682527065,
+      "num_tokens": 8153070.0,
+      "step": 3520
+    },
+    {
+      "epoch": 8.48280024140012,
+      "eval_entropy": 0.2379308074210467,
+      "eval_loss": 1.0180530548095703,
+      "eval_mean_token_accuracy": 0.8282477052024241,
+      "eval_num_tokens": 8153070.0,
+      "eval_runtime": 88.4651,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 3520
+    },
+    {
+      "entropy": 0.0906269280705601,
+      "epoch": 8.531080265540133,
+      "grad_norm": 0.37602657079696655,
+      "learning_rate": 2.4304121628587173e-05,
+      "loss": 0.06921588778495788,
+      "mean_token_accuracy": 0.9745571479201317,
+      "num_tokens": 8200066.0,
+      "step": 3540
+    },
+    {
+      "epoch": 8.531080265540133,
+      "eval_entropy": 0.2357490769263064,
+      "eval_loss": 1.017139196395874,
+      "eval_mean_token_accuracy": 0.828713178299786,
+      "eval_num_tokens": 8200066.0,
+      "eval_runtime": 88.6424,
+      "eval_samples_per_second": 16.019,
+      "eval_steps_per_second": 2.008,
+      "step": 3540
+    },
+    {
+      "entropy": 0.09273334196768701,
+      "epoch": 8.579360289680144,
+      "grad_norm": 0.2978838086128235,
+      "learning_rate": 2.277144809112794e-05,
+      "loss": 0.06970887780189514,
+      "mean_token_accuracy": 0.9751374177634716,
+      "num_tokens": 8246157.0,
+      "step": 3560
+    },
+    {
+      "epoch": 8.579360289680144,
+      "eval_entropy": 0.23590933138065123,
+      "eval_loss": 1.019231915473938,
+      "eval_mean_token_accuracy": 0.8279163412163767,
+      "eval_num_tokens": 8246157.0,
+      "eval_runtime": 88.2691,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 3560
+    },
+    {
+      "entropy": 0.09854201921261847,
+      "epoch": 8.627640313820157,
+      "grad_norm": 0.5381243228912354,
+      "learning_rate": 2.1285572850792413e-05,
+      "loss": 0.07123501300811767,
+      "mean_token_accuracy": 0.9723988175392151,
+      "num_tokens": 8291752.0,
+      "step": 3580
+    },
+    {
+      "epoch": 8.627640313820157,
+      "eval_entropy": 0.23592024949494372,
+      "eval_loss": 1.020250678062439,
+      "eval_mean_token_accuracy": 0.8282466226079491,
+      "eval_num_tokens": 8291752.0,
+      "eval_runtime": 87.8939,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 3580
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.5056596366565786e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-360/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-360/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-360/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-360/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-360/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-360/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-360/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-360/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-360/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-360/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-360/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-360/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-360/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-360/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..aa40e067342ae28e23f5106e938f5704a0d46fc1
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-360/trainer_state.json
@@ -0,0 +1,412 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.8690404345202173,
+  "eval_steps": 20,
+  "global_step": 360,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.5625769909518336e+16,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3600/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3600/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3600/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3600/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3600/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3600/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3600/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3600/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3600/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3600/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3600/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3600/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3600/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3600/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..7a80a14dbbe5bbc20dafdf83628208fd93e65346
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3600/trainer_state.json
@@ -0,0 +1,3814 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 8.67592033796017,
+  "eval_steps": 20,
+  "global_step": 3600,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    },
+    {
+      "entropy": 0.1326121488586068,
+      "epoch": 7.181050090525045,
+      "grad_norm": 0.4843117296695709,
+      "learning_rate": 8.411870386507724e-05,
+      "loss": 0.09537842869758606,
+      "mean_token_accuracy": 0.9645151920616627,
+      "num_tokens": 6898444.0,
+      "step": 2980
+    },
+    {
+      "epoch": 7.181050090525045,
+      "eval_entropy": 0.27999477783280813,
+      "eval_loss": 0.9031037092208862,
+      "eval_mean_token_accuracy": 0.8261882204018282,
+      "eval_num_tokens": 6898444.0,
+      "eval_runtime": 88.5602,
+      "eval_samples_per_second": 16.034,
+      "eval_steps_per_second": 2.01,
+      "step": 2980
+    },
+    {
+      "entropy": 0.1156751093454659,
+      "epoch": 7.229330114665057,
+      "grad_norm": 0.5297922492027283,
+      "learning_rate": 8.149629351430326e-05,
+      "loss": 0.09203402996063233,
+      "mean_token_accuracy": 0.9689733698964119,
+      "num_tokens": 6947106.0,
+      "step": 3000
+    },
+    {
+      "epoch": 7.229330114665057,
+      "eval_entropy": 0.2740225845508361,
+      "eval_loss": 0.8876440525054932,
+      "eval_mean_token_accuracy": 0.8265116857678703,
+      "eval_num_tokens": 6947106.0,
+      "eval_runtime": 87.7971,
+      "eval_samples_per_second": 16.174,
+      "eval_steps_per_second": 2.027,
+      "step": 3000
+    },
+    {
+      "entropy": 0.12116114404052496,
+      "epoch": 7.27761013880507,
+      "grad_norm": 0.5467398166656494,
+      "learning_rate": 7.890406302404618e-05,
+      "loss": 0.0933056652545929,
+      "mean_token_accuracy": 0.9676455594599247,
+      "num_tokens": 6992784.0,
+      "step": 3020
+    },
+    {
+      "epoch": 7.27761013880507,
+      "eval_entropy": 0.27238342793804876,
+      "eval_loss": 0.9166117310523987,
+      "eval_mean_token_accuracy": 0.825499015242866,
+      "eval_num_tokens": 6992784.0,
+      "eval_runtime": 88.3276,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 3020
+    },
+    {
+      "entropy": 0.11715119928121567,
+      "epoch": 7.325890162945082,
+      "grad_norm": 0.4976702332496643,
+      "learning_rate": 7.634274596487176e-05,
+      "loss": 0.09449659585952759,
+      "mean_token_accuracy": 0.966506478190422,
+      "num_tokens": 7040482.0,
+      "step": 3040
+    },
+    {
+      "epoch": 7.325890162945082,
+      "eval_entropy": 0.2788975556077582,
+      "eval_loss": 0.8833065629005432,
+      "eval_mean_token_accuracy": 0.8269920596915684,
+      "eval_num_tokens": 7040482.0,
+      "eval_runtime": 88.9016,
+      "eval_samples_per_second": 15.973,
+      "eval_steps_per_second": 2.002,
+      "step": 3040
+    },
+    {
+      "entropy": 0.12565911412239075,
+      "epoch": 7.3741701870850935,
+      "grad_norm": 0.5728679895401001,
+      "learning_rate": 7.381306715921025e-05,
+      "loss": 0.09626244902610778,
+      "mean_token_accuracy": 0.9661444149911403,
+      "num_tokens": 7086261.0,
+      "step": 3060
+    },
+    {
+      "epoch": 7.3741701870850935,
+      "eval_entropy": 0.2762492011939542,
+      "eval_loss": 0.8961160778999329,
+      "eval_mean_token_accuracy": 0.826369910092836,
+      "eval_num_tokens": 7086261.0,
+      "eval_runtime": 88.5204,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 3060
+    },
+    {
+      "entropy": 0.12258567921817302,
+      "epoch": 7.422450211225105,
+      "grad_norm": 0.468985915184021,
+      "learning_rate": 7.131574247624031e-05,
+      "loss": 0.09120266437530518,
+      "mean_token_accuracy": 0.9674938447773457,
+      "num_tokens": 7133247.0,
+      "step": 3080
+    },
+    {
+      "epoch": 7.422450211225105,
+      "eval_entropy": 0.2688512086366,
+      "eval_loss": 0.9048654437065125,
+      "eval_mean_token_accuracy": 0.8265582302982888,
+      "eval_num_tokens": 7133247.0,
+      "eval_runtime": 88.164,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 3080
+    },
+    {
+      "entropy": 0.12070536343380808,
+      "epoch": 7.470730235365117,
+      "grad_norm": 0.5310841798782349,
+      "learning_rate": 6.885147862930688e-05,
+      "loss": 0.09386247992515565,
+      "mean_token_accuracy": 0.96664222702384,
+      "num_tokens": 7177810.0,
+      "step": 3100
+    },
+    {
+      "epoch": 7.470730235365117,
+      "eval_entropy": 0.26289632874593305,
+      "eval_loss": 0.902837336063385,
+      "eval_mean_token_accuracy": 0.828052184554968,
+      "eval_num_tokens": 7177810.0,
+      "eval_runtime": 88.1201,
+      "eval_samples_per_second": 16.114,
+      "eval_steps_per_second": 2.02,
+      "step": 3100
+    },
+    {
+      "entropy": 0.11322665181942285,
+      "epoch": 7.51901025950513,
+      "grad_norm": 0.6846067905426025,
+      "learning_rate": 6.642097297592908e-05,
+      "loss": 0.08772538304328918,
+      "mean_token_accuracy": 0.9691362135112286,
+      "num_tokens": 7225302.0,
+      "step": 3120
+    },
+    {
+      "epoch": 7.51901025950513,
+      "eval_entropy": 0.2670605613776807,
+      "eval_loss": 0.9130566120147705,
+      "eval_mean_token_accuracy": 0.826366903741708,
+      "eval_num_tokens": 7225302.0,
+      "eval_runtime": 88.6035,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3120
+    },
+    {
+      "entropy": 0.10985926580615342,
+      "epoch": 7.567290283645142,
+      "grad_norm": 0.4799748361110687,
+      "learning_rate": 6.402491332045683e-05,
+      "loss": 0.08672296404838561,
+      "mean_token_accuracy": 0.9698824048042297,
+      "num_tokens": 7274881.0,
+      "step": 3140
+    },
+    {
+      "epoch": 7.567290283645142,
+      "eval_entropy": 0.26657010866015146,
+      "eval_loss": 0.9056380987167358,
+      "eval_mean_token_accuracy": 0.8272326461384806,
+      "eval_num_tokens": 7274881.0,
+      "eval_runtime": 88.2967,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3140
+    },
+    {
+      "entropy": 0.11890473761595785,
+      "epoch": 7.615570307785154,
+      "grad_norm": 0.4081912934780121,
+      "learning_rate": 6.166397771942987e-05,
+      "loss": 0.09396827220916748,
+      "mean_token_accuracy": 0.967512384802103,
+      "num_tokens": 7321163.0,
+      "step": 3160
+    },
+    {
+      "epoch": 7.615570307785154,
+      "eval_entropy": 0.26364776575833226,
+      "eval_loss": 0.9204733967781067,
+      "eval_mean_token_accuracy": 0.827419729714983,
+      "eval_num_tokens": 7321163.0,
+      "eval_runtime": 88.3441,
+      "eval_samples_per_second": 16.074,
+      "eval_steps_per_second": 2.015,
+      "step": 3160
+    },
+    {
+      "entropy": 0.12209134344011545,
+      "epoch": 7.663850331925166,
+      "grad_norm": 0.5159774422645569,
+      "learning_rate": 5.93388342896962e-05,
+      "loss": 0.09292978644371033,
+      "mean_token_accuracy": 0.9658421002328396,
+      "num_tokens": 7366217.0,
+      "step": 3180
+    },
+    {
+      "epoch": 7.663850331925166,
+      "eval_entropy": 0.2714110068725736,
+      "eval_loss": 0.9143092036247253,
+      "eval_mean_token_accuracy": 0.8262251099843657,
+      "eval_num_tokens": 7366217.0,
+      "eval_runtime": 87.9635,
+      "eval_samples_per_second": 16.143,
+      "eval_steps_per_second": 2.024,
+      "step": 3180
+    },
+    {
+      "entropy": 0.11955609126016498,
+      "epoch": 7.712130356065178,
+      "grad_norm": 0.7005976438522339,
+      "learning_rate": 5.705014101934252e-05,
+      "loss": 0.09319776892662049,
+      "mean_token_accuracy": 0.9674513377249241,
+      "num_tokens": 7410887.0,
+      "step": 3200
+    },
+    {
+      "epoch": 7.712130356065178,
+      "eval_entropy": 0.2626291513442993,
+      "eval_loss": 0.91884446144104,
+      "eval_mean_token_accuracy": 0.827563955207889,
+      "eval_num_tokens": 7410887.0,
+      "eval_runtime": 87.8849,
+      "eval_samples_per_second": 16.157,
+      "eval_steps_per_second": 2.025,
+      "step": 3200
+    },
+    {
+      "entropy": 0.11767532899975777,
+      "epoch": 7.76041038020519,
+      "grad_norm": 0.4675362706184387,
+      "learning_rate": 5.479854558149187e-05,
+      "loss": 0.09038087725639343,
+      "mean_token_accuracy": 0.9669680051505566,
+      "num_tokens": 7457578.0,
+      "step": 3220
+    },
+    {
+      "epoch": 7.76041038020519,
+      "eval_entropy": 0.26581037203582486,
+      "eval_loss": 0.9146883487701416,
+      "eval_mean_token_accuracy": 0.8270345812433222,
+      "eval_num_tokens": 7457578.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 3220
+    },
+    {
+      "entropy": 0.11383061073720455,
+      "epoch": 7.808690404345202,
+      "grad_norm": 0.541247546672821,
+      "learning_rate": 5.258468515101918e-05,
+      "loss": 0.08942405581474304,
+      "mean_token_accuracy": 0.968398705124855,
+      "num_tokens": 7505761.0,
+      "step": 3240
+    },
+    {
+      "epoch": 7.808690404345202,
+      "eval_entropy": 0.26687998960861997,
+      "eval_loss": 0.9037085175514221,
+      "eval_mean_token_accuracy": 0.8274977863504646,
+      "eval_num_tokens": 7505761.0,
+      "eval_runtime": 88.5163,
+      "eval_samples_per_second": 16.042,
+      "eval_steps_per_second": 2.011,
+      "step": 3240
+    },
+    {
+      "entropy": 0.12004058873280883,
+      "epoch": 7.856970428485214,
+      "grad_norm": 0.5289552211761475,
+      "learning_rate": 5.040918622423899e-05,
+      "loss": 0.09064805507659912,
+      "mean_token_accuracy": 0.9673330299556255,
+      "num_tokens": 7552557.0,
+      "step": 3260
+    },
+    {
+      "epoch": 7.856970428485214,
+      "eval_entropy": 0.26288379251622085,
+      "eval_loss": 0.9146178364753723,
+      "eval_mean_token_accuracy": 0.8276662448149049,
+      "eval_num_tokens": 7552557.0,
+      "eval_runtime": 88.0723,
+      "eval_samples_per_second": 16.123,
+      "eval_steps_per_second": 2.021,
+      "step": 3260
+    },
+    {
+      "entropy": 0.1180689456872642,
+      "epoch": 7.905250452625227,
+      "grad_norm": 0.4754861891269684,
+      "learning_rate": 4.827266444161377e-05,
+      "loss": 0.09402655363082886,
+      "mean_token_accuracy": 0.9663361787796021,
+      "num_tokens": 7596998.0,
+      "step": 3280
+    },
+    {
+      "epoch": 7.905250452625227,
+      "eval_entropy": 0.261690996922134,
+      "eval_loss": 0.9120186567306519,
+      "eval_mean_token_accuracy": 0.8278995015647974,
+      "eval_num_tokens": 7596998.0,
+      "eval_runtime": 88.6573,
+      "eval_samples_per_second": 16.017,
+      "eval_steps_per_second": 2.008,
+      "step": 3280
+    },
+    {
+      "entropy": 0.1162164019420743,
+      "epoch": 7.953530476765239,
+      "grad_norm": 0.6203398108482361,
+      "learning_rate": 4.61757244135355e-05,
+      "loss": 0.08832768201828003,
+      "mean_token_accuracy": 0.9694220684468746,
+      "num_tokens": 7643288.0,
+      "step": 3300
+    },
+    {
+      "epoch": 7.953530476765239,
+      "eval_entropy": 0.26158678129817664,
+      "eval_loss": 0.9301361441612244,
+      "eval_mean_token_accuracy": 0.8269603885291668,
+      "eval_num_tokens": 7643288.0,
+      "eval_runtime": 88.7064,
+      "eval_samples_per_second": 16.008,
+      "eval_steps_per_second": 2.007,
+      "step": 3300
+    },
+    {
+      "entropy": 0.10877202204489089,
+      "epoch": 8.0,
+      "grad_norm": 3.058673143386841,
+      "learning_rate": 4.411895954922763e-05,
+      "loss": 0.09159615635871887,
+      "mean_token_accuracy": 0.9686878834451947,
+      "num_tokens": 7688608.0,
+      "step": 3320
+    },
+    {
+      "epoch": 8.0,
+      "eval_entropy": 0.2590438057197614,
+      "eval_loss": 0.9262065887451172,
+      "eval_mean_token_accuracy": 0.8277404546067956,
+      "eval_num_tokens": 7688608.0,
+      "eval_runtime": 87.7557,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 3320
+    },
+    {
+      "entropy": 0.09969092034734786,
+      "epoch": 8.048280024140013,
+      "grad_norm": 0.2561911642551422,
+      "learning_rate": 4.2102951888817644e-05,
+      "loss": 0.07228735685348511,
+      "mean_token_accuracy": 0.9743556551635265,
+      "num_tokens": 7732856.0,
+      "step": 3340
+    },
+    {
+      "epoch": 8.048280024140013,
+      "eval_entropy": 0.2462534454933713,
+      "eval_loss": 0.9755983352661133,
+      "eval_mean_token_accuracy": 0.827510209565752,
+      "eval_num_tokens": 7732856.0,
+      "eval_runtime": 88.0297,
+      "eval_samples_per_second": 16.131,
+      "eval_steps_per_second": 2.022,
+      "step": 3340
+    },
+    {
+      "entropy": 0.09477624730207027,
+      "epoch": 8.096560048280024,
+      "grad_norm": 0.3670356571674347,
+      "learning_rate": 4.012827193862633e-05,
+      "loss": 0.0686036467552185,
+      "mean_token_accuracy": 0.9743670664727688,
+      "num_tokens": 7778585.0,
+      "step": 3360
+    },
+    {
+      "epoch": 8.096560048280024,
+      "eval_entropy": 0.2440882951356052,
+      "eval_loss": 1.0118423700332642,
+      "eval_mean_token_accuracy": 0.8271241070849172,
+      "eval_num_tokens": 7778585.0,
+      "eval_runtime": 88.1364,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 3360
+    },
+    {
+      "entropy": 0.09417980071157217,
+      "epoch": 8.144840072420036,
+      "grad_norm": 0.5785111784934998,
+      "learning_rate": 3.819547850972133e-05,
+      "loss": 0.07100698947906495,
+      "mean_token_accuracy": 0.9737804099917412,
+      "num_tokens": 7824287.0,
+      "step": 3380
+    },
+    {
+      "epoch": 8.144840072420036,
+      "eval_entropy": 0.24603172860453637,
+      "eval_loss": 1.0021991729736328,
+      "eval_mean_token_accuracy": 0.827162122793412,
+      "eval_num_tokens": 7824287.0,
+      "eval_runtime": 88.4534,
+      "eval_samples_per_second": 16.054,
+      "eval_steps_per_second": 2.012,
+      "step": 3380
+    },
+    {
+      "entropy": 0.087222247896716,
+      "epoch": 8.193120096560047,
+      "grad_norm": 0.44848504662513733,
+      "learning_rate": 3.6305118559780393e-05,
+      "loss": 0.06325234174728393,
+      "mean_token_accuracy": 0.9769335232675076,
+      "num_tokens": 7874784.0,
+      "step": 3400
+    },
+    {
+      "epoch": 8.193120096560047,
+      "eval_entropy": 0.24544863905129807,
+      "eval_loss": 1.005008339881897,
+      "eval_mean_token_accuracy": 0.8273035240976998,
+      "eval_num_tokens": 7874784.0,
+      "eval_runtime": 88.6737,
+      "eval_samples_per_second": 16.014,
+      "eval_steps_per_second": 2.007,
+      "step": 3400
+    },
+    {
+      "entropy": 0.09037275062873959,
+      "epoch": 8.24140012070006,
+      "grad_norm": 0.23529155552387238,
+      "learning_rate": 3.4457727038308435e-05,
+      "loss": 0.06589569449424744,
+      "mean_token_accuracy": 0.9751470290124417,
+      "num_tokens": 7925240.0,
+      "step": 3420
+    },
+    {
+      "epoch": 8.24140012070006,
+      "eval_entropy": 0.24203423494368456,
+      "eval_loss": 1.0052824020385742,
+      "eval_mean_token_accuracy": 0.8275575503874361,
+      "eval_num_tokens": 7925240.0,
+      "eval_runtime": 88.3486,
+      "eval_samples_per_second": 16.073,
+      "eval_steps_per_second": 2.015,
+      "step": 3420
+    },
+    {
+      "entropy": 0.10373208015225828,
+      "epoch": 8.289680144840073,
+      "grad_norm": 0.24796408414840698,
+      "learning_rate": 3.2653826735253594e-05,
+      "loss": 0.07714771628379821,
+      "mean_token_accuracy": 0.9722891598939896,
+      "num_tokens": 7966685.0,
+      "step": 3440
+    },
+    {
+      "epoch": 8.289680144840073,
+      "eval_entropy": 0.23461157050025597,
+      "eval_loss": 1.012108325958252,
+      "eval_mean_token_accuracy": 0.82825900564033,
+      "eval_num_tokens": 7966685.0,
+      "eval_runtime": 88.1018,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 3440
+    },
+    {
+      "entropy": 0.08967954274266958,
+      "epoch": 8.337960168980084,
+      "grad_norm": 0.5314674973487854,
+      "learning_rate": 3.08939281330635e-05,
+      "loss": 0.06746266484260559,
+      "mean_token_accuracy": 0.9753133699297905,
+      "num_tokens": 8013656.0,
+      "step": 3460
+    },
+    {
+      "epoch": 8.337960168980084,
+      "eval_entropy": 0.23781998589467468,
+      "eval_loss": 1.0119913816452026,
+      "eval_mean_token_accuracy": 0.8273287157664139,
+      "eval_num_tokens": 8013656.0,
+      "eval_runtime": 88.6038,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3460
+    },
+    {
+      "entropy": 0.09280927330255509,
+      "epoch": 8.386240193120097,
+      "grad_norm": 0.30658936500549316,
+      "learning_rate": 2.9178529262224946e-05,
+      "loss": 0.06936421990394592,
+      "mean_token_accuracy": 0.9742105580866337,
+      "num_tokens": 8061108.0,
+      "step": 3480
+    },
+    {
+      "epoch": 8.386240193120097,
+      "eval_entropy": 0.23672426734747512,
+      "eval_loss": 1.0162041187286377,
+      "eval_mean_token_accuracy": 0.8278394281194451,
+      "eval_num_tokens": 8061108.0,
+      "eval_runtime": 88.6015,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 3480
+    },
+    {
+      "entropy": 0.09195122267119586,
+      "epoch": 8.43452021726011,
+      "grad_norm": 0.24512454867362976,
+      "learning_rate": 2.7508115560327256e-05,
+      "loss": 0.06996064782142639,
+      "mean_token_accuracy": 0.974104854464531,
+      "num_tokens": 8106924.0,
+      "step": 3500
+    },
+    {
+      "epoch": 8.43452021726011,
+      "eval_entropy": 0.23758245828781235,
+      "eval_loss": 1.0157747268676758,
+      "eval_mean_token_accuracy": 0.8278501579600773,
+      "eval_num_tokens": 8106924.0,
+      "eval_runtime": 88.5678,
+      "eval_samples_per_second": 16.033,
+      "eval_steps_per_second": 2.01,
+      "step": 3500
+    },
+    {
+      "entropy": 0.09518477590754629,
+      "epoch": 8.48280024140012,
+      "grad_norm": 0.2373228371143341,
+      "learning_rate": 2.5883159734688868e-05,
+      "loss": 0.06832797527313232,
+      "mean_token_accuracy": 0.9743806682527065,
+      "num_tokens": 8153070.0,
+      "step": 3520
+    },
+    {
+      "epoch": 8.48280024140012,
+      "eval_entropy": 0.2379308074210467,
+      "eval_loss": 1.0180530548095703,
+      "eval_mean_token_accuracy": 0.8282477052024241,
+      "eval_num_tokens": 8153070.0,
+      "eval_runtime": 88.4651,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 3520
+    },
+    {
+      "entropy": 0.0906269280705601,
+      "epoch": 8.531080265540133,
+      "grad_norm": 0.37602657079696655,
+      "learning_rate": 2.4304121628587173e-05,
+      "loss": 0.06921588778495788,
+      "mean_token_accuracy": 0.9745571479201317,
+      "num_tokens": 8200066.0,
+      "step": 3540
+    },
+    {
+      "epoch": 8.531080265540133,
+      "eval_entropy": 0.2357490769263064,
+      "eval_loss": 1.017139196395874,
+      "eval_mean_token_accuracy": 0.828713178299786,
+      "eval_num_tokens": 8200066.0,
+      "eval_runtime": 88.6424,
+      "eval_samples_per_second": 16.019,
+      "eval_steps_per_second": 2.008,
+      "step": 3540
+    },
+    {
+      "entropy": 0.09273334196768701,
+      "epoch": 8.579360289680144,
+      "grad_norm": 0.2978838086128235,
+      "learning_rate": 2.277144809112794e-05,
+      "loss": 0.06970887780189514,
+      "mean_token_accuracy": 0.9751374177634716,
+      "num_tokens": 8246157.0,
+      "step": 3560
+    },
+    {
+      "epoch": 8.579360289680144,
+      "eval_entropy": 0.23590933138065123,
+      "eval_loss": 1.019231915473938,
+      "eval_mean_token_accuracy": 0.8279163412163767,
+      "eval_num_tokens": 8246157.0,
+      "eval_runtime": 88.2691,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 3560
+    },
+    {
+      "entropy": 0.09854201921261847,
+      "epoch": 8.627640313820157,
+      "grad_norm": 0.5381243228912354,
+      "learning_rate": 2.1285572850792413e-05,
+      "loss": 0.07123501300811767,
+      "mean_token_accuracy": 0.9723988175392151,
+      "num_tokens": 8291752.0,
+      "step": 3580
+    },
+    {
+      "epoch": 8.627640313820157,
+      "eval_entropy": 0.23592024949494372,
+      "eval_loss": 1.020250678062439,
+      "eval_mean_token_accuracy": 0.8282466226079491,
+      "eval_num_tokens": 8291752.0,
+      "eval_runtime": 87.8939,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 3580
+    },
+    {
+      "entropy": 0.09092000788077712,
+      "epoch": 8.67592033796017,
+      "grad_norm": 0.29089561104774475,
+      "learning_rate": 1.9846916392697493e-05,
+      "loss": 0.06982720494270325,
+      "mean_token_accuracy": 0.9749995484948158,
+      "num_tokens": 8337651.0,
+      "step": 3600
+    },
+    {
+      "epoch": 8.67592033796017,
+      "eval_entropy": 0.23658913382318583,
+      "eval_loss": 1.019862413406372,
+      "eval_mean_token_accuracy": 0.8280127842104836,
+      "eval_num_tokens": 8337651.0,
+      "eval_runtime": 88.2855,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 3600
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.5255281446404506e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3620/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3620/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3620/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3620/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3620/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3620/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3620/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3620/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3620/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3620/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3620/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3620/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3620/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3620/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..1573ce2909ecc20301e76c24fe61c9ee7b21f0f6
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3620/trainer_state.json
@@ -0,0 +1,3835 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 8.72420036210018,
+  "eval_steps": 20,
+  "global_step": 3620,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    },
+    {
+      "entropy": 0.1326121488586068,
+      "epoch": 7.181050090525045,
+      "grad_norm": 0.4843117296695709,
+      "learning_rate": 8.411870386507724e-05,
+      "loss": 0.09537842869758606,
+      "mean_token_accuracy": 0.9645151920616627,
+      "num_tokens": 6898444.0,
+      "step": 2980
+    },
+    {
+      "epoch": 7.181050090525045,
+      "eval_entropy": 0.27999477783280813,
+      "eval_loss": 0.9031037092208862,
+      "eval_mean_token_accuracy": 0.8261882204018282,
+      "eval_num_tokens": 6898444.0,
+      "eval_runtime": 88.5602,
+      "eval_samples_per_second": 16.034,
+      "eval_steps_per_second": 2.01,
+      "step": 2980
+    },
+    {
+      "entropy": 0.1156751093454659,
+      "epoch": 7.229330114665057,
+      "grad_norm": 0.5297922492027283,
+      "learning_rate": 8.149629351430326e-05,
+      "loss": 0.09203402996063233,
+      "mean_token_accuracy": 0.9689733698964119,
+      "num_tokens": 6947106.0,
+      "step": 3000
+    },
+    {
+      "epoch": 7.229330114665057,
+      "eval_entropy": 0.2740225845508361,
+      "eval_loss": 0.8876440525054932,
+      "eval_mean_token_accuracy": 0.8265116857678703,
+      "eval_num_tokens": 6947106.0,
+      "eval_runtime": 87.7971,
+      "eval_samples_per_second": 16.174,
+      "eval_steps_per_second": 2.027,
+      "step": 3000
+    },
+    {
+      "entropy": 0.12116114404052496,
+      "epoch": 7.27761013880507,
+      "grad_norm": 0.5467398166656494,
+      "learning_rate": 7.890406302404618e-05,
+      "loss": 0.0933056652545929,
+      "mean_token_accuracy": 0.9676455594599247,
+      "num_tokens": 6992784.0,
+      "step": 3020
+    },
+    {
+      "epoch": 7.27761013880507,
+      "eval_entropy": 0.27238342793804876,
+      "eval_loss": 0.9166117310523987,
+      "eval_mean_token_accuracy": 0.825499015242866,
+      "eval_num_tokens": 6992784.0,
+      "eval_runtime": 88.3276,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 3020
+    },
+    {
+      "entropy": 0.11715119928121567,
+      "epoch": 7.325890162945082,
+      "grad_norm": 0.4976702332496643,
+      "learning_rate": 7.634274596487176e-05,
+      "loss": 0.09449659585952759,
+      "mean_token_accuracy": 0.966506478190422,
+      "num_tokens": 7040482.0,
+      "step": 3040
+    },
+    {
+      "epoch": 7.325890162945082,
+      "eval_entropy": 0.2788975556077582,
+      "eval_loss": 0.8833065629005432,
+      "eval_mean_token_accuracy": 0.8269920596915684,
+      "eval_num_tokens": 7040482.0,
+      "eval_runtime": 88.9016,
+      "eval_samples_per_second": 15.973,
+      "eval_steps_per_second": 2.002,
+      "step": 3040
+    },
+    {
+      "entropy": 0.12565911412239075,
+      "epoch": 7.3741701870850935,
+      "grad_norm": 0.5728679895401001,
+      "learning_rate": 7.381306715921025e-05,
+      "loss": 0.09626244902610778,
+      "mean_token_accuracy": 0.9661444149911403,
+      "num_tokens": 7086261.0,
+      "step": 3060
+    },
+    {
+      "epoch": 7.3741701870850935,
+      "eval_entropy": 0.2762492011939542,
+      "eval_loss": 0.8961160778999329,
+      "eval_mean_token_accuracy": 0.826369910092836,
+      "eval_num_tokens": 7086261.0,
+      "eval_runtime": 88.5204,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 3060
+    },
+    {
+      "entropy": 0.12258567921817302,
+      "epoch": 7.422450211225105,
+      "grad_norm": 0.468985915184021,
+      "learning_rate": 7.131574247624031e-05,
+      "loss": 0.09120266437530518,
+      "mean_token_accuracy": 0.9674938447773457,
+      "num_tokens": 7133247.0,
+      "step": 3080
+    },
+    {
+      "epoch": 7.422450211225105,
+      "eval_entropy": 0.2688512086366,
+      "eval_loss": 0.9048654437065125,
+      "eval_mean_token_accuracy": 0.8265582302982888,
+      "eval_num_tokens": 7133247.0,
+      "eval_runtime": 88.164,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 3080
+    },
+    {
+      "entropy": 0.12070536343380808,
+      "epoch": 7.470730235365117,
+      "grad_norm": 0.5310841798782349,
+      "learning_rate": 6.885147862930688e-05,
+      "loss": 0.09386247992515565,
+      "mean_token_accuracy": 0.96664222702384,
+      "num_tokens": 7177810.0,
+      "step": 3100
+    },
+    {
+      "epoch": 7.470730235365117,
+      "eval_entropy": 0.26289632874593305,
+      "eval_loss": 0.902837336063385,
+      "eval_mean_token_accuracy": 0.828052184554968,
+      "eval_num_tokens": 7177810.0,
+      "eval_runtime": 88.1201,
+      "eval_samples_per_second": 16.114,
+      "eval_steps_per_second": 2.02,
+      "step": 3100
+    },
+    {
+      "entropy": 0.11322665181942285,
+      "epoch": 7.51901025950513,
+      "grad_norm": 0.6846067905426025,
+      "learning_rate": 6.642097297592908e-05,
+      "loss": 0.08772538304328918,
+      "mean_token_accuracy": 0.9691362135112286,
+      "num_tokens": 7225302.0,
+      "step": 3120
+    },
+    {
+      "epoch": 7.51901025950513,
+      "eval_entropy": 0.2670605613776807,
+      "eval_loss": 0.9130566120147705,
+      "eval_mean_token_accuracy": 0.826366903741708,
+      "eval_num_tokens": 7225302.0,
+      "eval_runtime": 88.6035,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3120
+    },
+    {
+      "entropy": 0.10985926580615342,
+      "epoch": 7.567290283645142,
+      "grad_norm": 0.4799748361110687,
+      "learning_rate": 6.402491332045683e-05,
+      "loss": 0.08672296404838561,
+      "mean_token_accuracy": 0.9698824048042297,
+      "num_tokens": 7274881.0,
+      "step": 3140
+    },
+    {
+      "epoch": 7.567290283645142,
+      "eval_entropy": 0.26657010866015146,
+      "eval_loss": 0.9056380987167358,
+      "eval_mean_token_accuracy": 0.8272326461384806,
+      "eval_num_tokens": 7274881.0,
+      "eval_runtime": 88.2967,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3140
+    },
+    {
+      "entropy": 0.11890473761595785,
+      "epoch": 7.615570307785154,
+      "grad_norm": 0.4081912934780121,
+      "learning_rate": 6.166397771942987e-05,
+      "loss": 0.09396827220916748,
+      "mean_token_accuracy": 0.967512384802103,
+      "num_tokens": 7321163.0,
+      "step": 3160
+    },
+    {
+      "epoch": 7.615570307785154,
+      "eval_entropy": 0.26364776575833226,
+      "eval_loss": 0.9204733967781067,
+      "eval_mean_token_accuracy": 0.827419729714983,
+      "eval_num_tokens": 7321163.0,
+      "eval_runtime": 88.3441,
+      "eval_samples_per_second": 16.074,
+      "eval_steps_per_second": 2.015,
+      "step": 3160
+    },
+    {
+      "entropy": 0.12209134344011545,
+      "epoch": 7.663850331925166,
+      "grad_norm": 0.5159774422645569,
+      "learning_rate": 5.93388342896962e-05,
+      "loss": 0.09292978644371033,
+      "mean_token_accuracy": 0.9658421002328396,
+      "num_tokens": 7366217.0,
+      "step": 3180
+    },
+    {
+      "epoch": 7.663850331925166,
+      "eval_entropy": 0.2714110068725736,
+      "eval_loss": 0.9143092036247253,
+      "eval_mean_token_accuracy": 0.8262251099843657,
+      "eval_num_tokens": 7366217.0,
+      "eval_runtime": 87.9635,
+      "eval_samples_per_second": 16.143,
+      "eval_steps_per_second": 2.024,
+      "step": 3180
+    },
+    {
+      "entropy": 0.11955609126016498,
+      "epoch": 7.712130356065178,
+      "grad_norm": 0.7005976438522339,
+      "learning_rate": 5.705014101934252e-05,
+      "loss": 0.09319776892662049,
+      "mean_token_accuracy": 0.9674513377249241,
+      "num_tokens": 7410887.0,
+      "step": 3200
+    },
+    {
+      "epoch": 7.712130356065178,
+      "eval_entropy": 0.2626291513442993,
+      "eval_loss": 0.91884446144104,
+      "eval_mean_token_accuracy": 0.827563955207889,
+      "eval_num_tokens": 7410887.0,
+      "eval_runtime": 87.8849,
+      "eval_samples_per_second": 16.157,
+      "eval_steps_per_second": 2.025,
+      "step": 3200
+    },
+    {
+      "entropy": 0.11767532899975777,
+      "epoch": 7.76041038020519,
+      "grad_norm": 0.4675362706184387,
+      "learning_rate": 5.479854558149187e-05,
+      "loss": 0.09038087725639343,
+      "mean_token_accuracy": 0.9669680051505566,
+      "num_tokens": 7457578.0,
+      "step": 3220
+    },
+    {
+      "epoch": 7.76041038020519,
+      "eval_entropy": 0.26581037203582486,
+      "eval_loss": 0.9146883487701416,
+      "eval_mean_token_accuracy": 0.8270345812433222,
+      "eval_num_tokens": 7457578.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 3220
+    },
+    {
+      "entropy": 0.11383061073720455,
+      "epoch": 7.808690404345202,
+      "grad_norm": 0.541247546672821,
+      "learning_rate": 5.258468515101918e-05,
+      "loss": 0.08942405581474304,
+      "mean_token_accuracy": 0.968398705124855,
+      "num_tokens": 7505761.0,
+      "step": 3240
+    },
+    {
+      "epoch": 7.808690404345202,
+      "eval_entropy": 0.26687998960861997,
+      "eval_loss": 0.9037085175514221,
+      "eval_mean_token_accuracy": 0.8274977863504646,
+      "eval_num_tokens": 7505761.0,
+      "eval_runtime": 88.5163,
+      "eval_samples_per_second": 16.042,
+      "eval_steps_per_second": 2.011,
+      "step": 3240
+    },
+    {
+      "entropy": 0.12004058873280883,
+      "epoch": 7.856970428485214,
+      "grad_norm": 0.5289552211761475,
+      "learning_rate": 5.040918622423899e-05,
+      "loss": 0.09064805507659912,
+      "mean_token_accuracy": 0.9673330299556255,
+      "num_tokens": 7552557.0,
+      "step": 3260
+    },
+    {
+      "epoch": 7.856970428485214,
+      "eval_entropy": 0.26288379251622085,
+      "eval_loss": 0.9146178364753723,
+      "eval_mean_token_accuracy": 0.8276662448149049,
+      "eval_num_tokens": 7552557.0,
+      "eval_runtime": 88.0723,
+      "eval_samples_per_second": 16.123,
+      "eval_steps_per_second": 2.021,
+      "step": 3260
+    },
+    {
+      "entropy": 0.1180689456872642,
+      "epoch": 7.905250452625227,
+      "grad_norm": 0.4754861891269684,
+      "learning_rate": 4.827266444161377e-05,
+      "loss": 0.09402655363082886,
+      "mean_token_accuracy": 0.9663361787796021,
+      "num_tokens": 7596998.0,
+      "step": 3280
+    },
+    {
+      "epoch": 7.905250452625227,
+      "eval_entropy": 0.261690996922134,
+      "eval_loss": 0.9120186567306519,
+      "eval_mean_token_accuracy": 0.8278995015647974,
+      "eval_num_tokens": 7596998.0,
+      "eval_runtime": 88.6573,
+      "eval_samples_per_second": 16.017,
+      "eval_steps_per_second": 2.008,
+      "step": 3280
+    },
+    {
+      "entropy": 0.1162164019420743,
+      "epoch": 7.953530476765239,
+      "grad_norm": 0.6203398108482361,
+      "learning_rate": 4.61757244135355e-05,
+      "loss": 0.08832768201828003,
+      "mean_token_accuracy": 0.9694220684468746,
+      "num_tokens": 7643288.0,
+      "step": 3300
+    },
+    {
+      "epoch": 7.953530476765239,
+      "eval_entropy": 0.26158678129817664,
+      "eval_loss": 0.9301361441612244,
+      "eval_mean_token_accuracy": 0.8269603885291668,
+      "eval_num_tokens": 7643288.0,
+      "eval_runtime": 88.7064,
+      "eval_samples_per_second": 16.008,
+      "eval_steps_per_second": 2.007,
+      "step": 3300
+    },
+    {
+      "entropy": 0.10877202204489089,
+      "epoch": 8.0,
+      "grad_norm": 3.058673143386841,
+      "learning_rate": 4.411895954922763e-05,
+      "loss": 0.09159615635871887,
+      "mean_token_accuracy": 0.9686878834451947,
+      "num_tokens": 7688608.0,
+      "step": 3320
+    },
+    {
+      "epoch": 8.0,
+      "eval_entropy": 0.2590438057197614,
+      "eval_loss": 0.9262065887451172,
+      "eval_mean_token_accuracy": 0.8277404546067956,
+      "eval_num_tokens": 7688608.0,
+      "eval_runtime": 87.7557,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 3320
+    },
+    {
+      "entropy": 0.09969092034734786,
+      "epoch": 8.048280024140013,
+      "grad_norm": 0.2561911642551422,
+      "learning_rate": 4.2102951888817644e-05,
+      "loss": 0.07228735685348511,
+      "mean_token_accuracy": 0.9743556551635265,
+      "num_tokens": 7732856.0,
+      "step": 3340
+    },
+    {
+      "epoch": 8.048280024140013,
+      "eval_entropy": 0.2462534454933713,
+      "eval_loss": 0.9755983352661133,
+      "eval_mean_token_accuracy": 0.827510209565752,
+      "eval_num_tokens": 7732856.0,
+      "eval_runtime": 88.0297,
+      "eval_samples_per_second": 16.131,
+      "eval_steps_per_second": 2.022,
+      "step": 3340
+    },
+    {
+      "entropy": 0.09477624730207027,
+      "epoch": 8.096560048280024,
+      "grad_norm": 0.3670356571674347,
+      "learning_rate": 4.012827193862633e-05,
+      "loss": 0.0686036467552185,
+      "mean_token_accuracy": 0.9743670664727688,
+      "num_tokens": 7778585.0,
+      "step": 3360
+    },
+    {
+      "epoch": 8.096560048280024,
+      "eval_entropy": 0.2440882951356052,
+      "eval_loss": 1.0118423700332642,
+      "eval_mean_token_accuracy": 0.8271241070849172,
+      "eval_num_tokens": 7778585.0,
+      "eval_runtime": 88.1364,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 3360
+    },
+    {
+      "entropy": 0.09417980071157217,
+      "epoch": 8.144840072420036,
+      "grad_norm": 0.5785111784934998,
+      "learning_rate": 3.819547850972133e-05,
+      "loss": 0.07100698947906495,
+      "mean_token_accuracy": 0.9737804099917412,
+      "num_tokens": 7824287.0,
+      "step": 3380
+    },
+    {
+      "epoch": 8.144840072420036,
+      "eval_entropy": 0.24603172860453637,
+      "eval_loss": 1.0021991729736328,
+      "eval_mean_token_accuracy": 0.827162122793412,
+      "eval_num_tokens": 7824287.0,
+      "eval_runtime": 88.4534,
+      "eval_samples_per_second": 16.054,
+      "eval_steps_per_second": 2.012,
+      "step": 3380
+    },
+    {
+      "entropy": 0.087222247896716,
+      "epoch": 8.193120096560047,
+      "grad_norm": 0.44848504662513733,
+      "learning_rate": 3.6305118559780393e-05,
+      "loss": 0.06325234174728393,
+      "mean_token_accuracy": 0.9769335232675076,
+      "num_tokens": 7874784.0,
+      "step": 3400
+    },
+    {
+      "epoch": 8.193120096560047,
+      "eval_entropy": 0.24544863905129807,
+      "eval_loss": 1.005008339881897,
+      "eval_mean_token_accuracy": 0.8273035240976998,
+      "eval_num_tokens": 7874784.0,
+      "eval_runtime": 88.6737,
+      "eval_samples_per_second": 16.014,
+      "eval_steps_per_second": 2.007,
+      "step": 3400
+    },
+    {
+      "entropy": 0.09037275062873959,
+      "epoch": 8.24140012070006,
+      "grad_norm": 0.23529155552387238,
+      "learning_rate": 3.4457727038308435e-05,
+      "loss": 0.06589569449424744,
+      "mean_token_accuracy": 0.9751470290124417,
+      "num_tokens": 7925240.0,
+      "step": 3420
+    },
+    {
+      "epoch": 8.24140012070006,
+      "eval_entropy": 0.24203423494368456,
+      "eval_loss": 1.0052824020385742,
+      "eval_mean_token_accuracy": 0.8275575503874361,
+      "eval_num_tokens": 7925240.0,
+      "eval_runtime": 88.3486,
+      "eval_samples_per_second": 16.073,
+      "eval_steps_per_second": 2.015,
+      "step": 3420
+    },
+    {
+      "entropy": 0.10373208015225828,
+      "epoch": 8.289680144840073,
+      "grad_norm": 0.24796408414840698,
+      "learning_rate": 3.2653826735253594e-05,
+      "loss": 0.07714771628379821,
+      "mean_token_accuracy": 0.9722891598939896,
+      "num_tokens": 7966685.0,
+      "step": 3440
+    },
+    {
+      "epoch": 8.289680144840073,
+      "eval_entropy": 0.23461157050025597,
+      "eval_loss": 1.012108325958252,
+      "eval_mean_token_accuracy": 0.82825900564033,
+      "eval_num_tokens": 7966685.0,
+      "eval_runtime": 88.1018,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 3440
+    },
+    {
+      "entropy": 0.08967954274266958,
+      "epoch": 8.337960168980084,
+      "grad_norm": 0.5314674973487854,
+      "learning_rate": 3.08939281330635e-05,
+      "loss": 0.06746266484260559,
+      "mean_token_accuracy": 0.9753133699297905,
+      "num_tokens": 8013656.0,
+      "step": 3460
+    },
+    {
+      "epoch": 8.337960168980084,
+      "eval_entropy": 0.23781998589467468,
+      "eval_loss": 1.0119913816452026,
+      "eval_mean_token_accuracy": 0.8273287157664139,
+      "eval_num_tokens": 8013656.0,
+      "eval_runtime": 88.6038,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3460
+    },
+    {
+      "entropy": 0.09280927330255509,
+      "epoch": 8.386240193120097,
+      "grad_norm": 0.30658936500549316,
+      "learning_rate": 2.9178529262224946e-05,
+      "loss": 0.06936421990394592,
+      "mean_token_accuracy": 0.9742105580866337,
+      "num_tokens": 8061108.0,
+      "step": 3480
+    },
+    {
+      "epoch": 8.386240193120097,
+      "eval_entropy": 0.23672426734747512,
+      "eval_loss": 1.0162041187286377,
+      "eval_mean_token_accuracy": 0.8278394281194451,
+      "eval_num_tokens": 8061108.0,
+      "eval_runtime": 88.6015,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 3480
+    },
+    {
+      "entropy": 0.09195122267119586,
+      "epoch": 8.43452021726011,
+      "grad_norm": 0.24512454867362976,
+      "learning_rate": 2.7508115560327256e-05,
+      "loss": 0.06996064782142639,
+      "mean_token_accuracy": 0.974104854464531,
+      "num_tokens": 8106924.0,
+      "step": 3500
+    },
+    {
+      "epoch": 8.43452021726011,
+      "eval_entropy": 0.23758245828781235,
+      "eval_loss": 1.0157747268676758,
+      "eval_mean_token_accuracy": 0.8278501579600773,
+      "eval_num_tokens": 8106924.0,
+      "eval_runtime": 88.5678,
+      "eval_samples_per_second": 16.033,
+      "eval_steps_per_second": 2.01,
+      "step": 3500
+    },
+    {
+      "entropy": 0.09518477590754629,
+      "epoch": 8.48280024140012,
+      "grad_norm": 0.2373228371143341,
+      "learning_rate": 2.5883159734688868e-05,
+      "loss": 0.06832797527313232,
+      "mean_token_accuracy": 0.9743806682527065,
+      "num_tokens": 8153070.0,
+      "step": 3520
+    },
+    {
+      "epoch": 8.48280024140012,
+      "eval_entropy": 0.2379308074210467,
+      "eval_loss": 1.0180530548095703,
+      "eval_mean_token_accuracy": 0.8282477052024241,
+      "eval_num_tokens": 8153070.0,
+      "eval_runtime": 88.4651,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 3520
+    },
+    {
+      "entropy": 0.0906269280705601,
+      "epoch": 8.531080265540133,
+      "grad_norm": 0.37602657079696655,
+      "learning_rate": 2.4304121628587173e-05,
+      "loss": 0.06921588778495788,
+      "mean_token_accuracy": 0.9745571479201317,
+      "num_tokens": 8200066.0,
+      "step": 3540
+    },
+    {
+      "epoch": 8.531080265540133,
+      "eval_entropy": 0.2357490769263064,
+      "eval_loss": 1.017139196395874,
+      "eval_mean_token_accuracy": 0.828713178299786,
+      "eval_num_tokens": 8200066.0,
+      "eval_runtime": 88.6424,
+      "eval_samples_per_second": 16.019,
+      "eval_steps_per_second": 2.008,
+      "step": 3540
+    },
+    {
+      "entropy": 0.09273334196768701,
+      "epoch": 8.579360289680144,
+      "grad_norm": 0.2978838086128235,
+      "learning_rate": 2.277144809112794e-05,
+      "loss": 0.06970887780189514,
+      "mean_token_accuracy": 0.9751374177634716,
+      "num_tokens": 8246157.0,
+      "step": 3560
+    },
+    {
+      "epoch": 8.579360289680144,
+      "eval_entropy": 0.23590933138065123,
+      "eval_loss": 1.019231915473938,
+      "eval_mean_token_accuracy": 0.8279163412163767,
+      "eval_num_tokens": 8246157.0,
+      "eval_runtime": 88.2691,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 3560
+    },
+    {
+      "entropy": 0.09854201921261847,
+      "epoch": 8.627640313820157,
+      "grad_norm": 0.5381243228912354,
+      "learning_rate": 2.1285572850792413e-05,
+      "loss": 0.07123501300811767,
+      "mean_token_accuracy": 0.9723988175392151,
+      "num_tokens": 8291752.0,
+      "step": 3580
+    },
+    {
+      "epoch": 8.627640313820157,
+      "eval_entropy": 0.23592024949494372,
+      "eval_loss": 1.020250678062439,
+      "eval_mean_token_accuracy": 0.8282466226079491,
+      "eval_num_tokens": 8291752.0,
+      "eval_runtime": 87.8939,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 3580
+    },
+    {
+      "entropy": 0.09092000788077712,
+      "epoch": 8.67592033796017,
+      "grad_norm": 0.29089561104774475,
+      "learning_rate": 1.9846916392697493e-05,
+      "loss": 0.06982720494270325,
+      "mean_token_accuracy": 0.9749995484948158,
+      "num_tokens": 8337651.0,
+      "step": 3600
+    },
+    {
+      "epoch": 8.67592033796017,
+      "eval_entropy": 0.23658913382318583,
+      "eval_loss": 1.019862413406372,
+      "eval_mean_token_accuracy": 0.8280127842104836,
+      "eval_num_tokens": 8337651.0,
+      "eval_runtime": 88.2855,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 3600
+    },
+    {
+      "entropy": 0.08708424470387399,
+      "epoch": 8.72420036210018,
+      "grad_norm": 0.29904839396476746,
+      "learning_rate": 1.8455885839603106e-05,
+      "loss": 0.06801370978355407,
+      "mean_token_accuracy": 0.9755801670253277,
+      "num_tokens": 8385330.0,
+      "step": 3620
+    },
+    {
+      "epoch": 8.72420036210018,
+      "eval_entropy": 0.23627035771862845,
+      "eval_loss": 1.0204569101333618,
+      "eval_mean_token_accuracy": 0.8279967894045155,
+      "eval_num_tokens": 8385330.0,
+      "eval_runtime": 88.1753,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 3620
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.5458370054282035e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3640/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3640/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3640/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3640/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3640/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3640/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3640/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3640/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3640/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3640/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3640/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3640/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3640/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3640/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..291f00c7d5f67f5da42aeb5da6b0715cad8093ee
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3640/trainer_state.json
@@ -0,0 +1,3856 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 8.772480386240193,
+  "eval_steps": 20,
+  "global_step": 3640,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    },
+    {
+      "entropy": 0.1326121488586068,
+      "epoch": 7.181050090525045,
+      "grad_norm": 0.4843117296695709,
+      "learning_rate": 8.411870386507724e-05,
+      "loss": 0.09537842869758606,
+      "mean_token_accuracy": 0.9645151920616627,
+      "num_tokens": 6898444.0,
+      "step": 2980
+    },
+    {
+      "epoch": 7.181050090525045,
+      "eval_entropy": 0.27999477783280813,
+      "eval_loss": 0.9031037092208862,
+      "eval_mean_token_accuracy": 0.8261882204018282,
+      "eval_num_tokens": 6898444.0,
+      "eval_runtime": 88.5602,
+      "eval_samples_per_second": 16.034,
+      "eval_steps_per_second": 2.01,
+      "step": 2980
+    },
+    {
+      "entropy": 0.1156751093454659,
+      "epoch": 7.229330114665057,
+      "grad_norm": 0.5297922492027283,
+      "learning_rate": 8.149629351430326e-05,
+      "loss": 0.09203402996063233,
+      "mean_token_accuracy": 0.9689733698964119,
+      "num_tokens": 6947106.0,
+      "step": 3000
+    },
+    {
+      "epoch": 7.229330114665057,
+      "eval_entropy": 0.2740225845508361,
+      "eval_loss": 0.8876440525054932,
+      "eval_mean_token_accuracy": 0.8265116857678703,
+      "eval_num_tokens": 6947106.0,
+      "eval_runtime": 87.7971,
+      "eval_samples_per_second": 16.174,
+      "eval_steps_per_second": 2.027,
+      "step": 3000
+    },
+    {
+      "entropy": 0.12116114404052496,
+      "epoch": 7.27761013880507,
+      "grad_norm": 0.5467398166656494,
+      "learning_rate": 7.890406302404618e-05,
+      "loss": 0.0933056652545929,
+      "mean_token_accuracy": 0.9676455594599247,
+      "num_tokens": 6992784.0,
+      "step": 3020
+    },
+    {
+      "epoch": 7.27761013880507,
+      "eval_entropy": 0.27238342793804876,
+      "eval_loss": 0.9166117310523987,
+      "eval_mean_token_accuracy": 0.825499015242866,
+      "eval_num_tokens": 6992784.0,
+      "eval_runtime": 88.3276,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 3020
+    },
+    {
+      "entropy": 0.11715119928121567,
+      "epoch": 7.325890162945082,
+      "grad_norm": 0.4976702332496643,
+      "learning_rate": 7.634274596487176e-05,
+      "loss": 0.09449659585952759,
+      "mean_token_accuracy": 0.966506478190422,
+      "num_tokens": 7040482.0,
+      "step": 3040
+    },
+    {
+      "epoch": 7.325890162945082,
+      "eval_entropy": 0.2788975556077582,
+      "eval_loss": 0.8833065629005432,
+      "eval_mean_token_accuracy": 0.8269920596915684,
+      "eval_num_tokens": 7040482.0,
+      "eval_runtime": 88.9016,
+      "eval_samples_per_second": 15.973,
+      "eval_steps_per_second": 2.002,
+      "step": 3040
+    },
+    {
+      "entropy": 0.12565911412239075,
+      "epoch": 7.3741701870850935,
+      "grad_norm": 0.5728679895401001,
+      "learning_rate": 7.381306715921025e-05,
+      "loss": 0.09626244902610778,
+      "mean_token_accuracy": 0.9661444149911403,
+      "num_tokens": 7086261.0,
+      "step": 3060
+    },
+    {
+      "epoch": 7.3741701870850935,
+      "eval_entropy": 0.2762492011939542,
+      "eval_loss": 0.8961160778999329,
+      "eval_mean_token_accuracy": 0.826369910092836,
+      "eval_num_tokens": 7086261.0,
+      "eval_runtime": 88.5204,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 3060
+    },
+    {
+      "entropy": 0.12258567921817302,
+      "epoch": 7.422450211225105,
+      "grad_norm": 0.468985915184021,
+      "learning_rate": 7.131574247624031e-05,
+      "loss": 0.09120266437530518,
+      "mean_token_accuracy": 0.9674938447773457,
+      "num_tokens": 7133247.0,
+      "step": 3080
+    },
+    {
+      "epoch": 7.422450211225105,
+      "eval_entropy": 0.2688512086366,
+      "eval_loss": 0.9048654437065125,
+      "eval_mean_token_accuracy": 0.8265582302982888,
+      "eval_num_tokens": 7133247.0,
+      "eval_runtime": 88.164,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 3080
+    },
+    {
+      "entropy": 0.12070536343380808,
+      "epoch": 7.470730235365117,
+      "grad_norm": 0.5310841798782349,
+      "learning_rate": 6.885147862930688e-05,
+      "loss": 0.09386247992515565,
+      "mean_token_accuracy": 0.96664222702384,
+      "num_tokens": 7177810.0,
+      "step": 3100
+    },
+    {
+      "epoch": 7.470730235365117,
+      "eval_entropy": 0.26289632874593305,
+      "eval_loss": 0.902837336063385,
+      "eval_mean_token_accuracy": 0.828052184554968,
+      "eval_num_tokens": 7177810.0,
+      "eval_runtime": 88.1201,
+      "eval_samples_per_second": 16.114,
+      "eval_steps_per_second": 2.02,
+      "step": 3100
+    },
+    {
+      "entropy": 0.11322665181942285,
+      "epoch": 7.51901025950513,
+      "grad_norm": 0.6846067905426025,
+      "learning_rate": 6.642097297592908e-05,
+      "loss": 0.08772538304328918,
+      "mean_token_accuracy": 0.9691362135112286,
+      "num_tokens": 7225302.0,
+      "step": 3120
+    },
+    {
+      "epoch": 7.51901025950513,
+      "eval_entropy": 0.2670605613776807,
+      "eval_loss": 0.9130566120147705,
+      "eval_mean_token_accuracy": 0.826366903741708,
+      "eval_num_tokens": 7225302.0,
+      "eval_runtime": 88.6035,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3120
+    },
+    {
+      "entropy": 0.10985926580615342,
+      "epoch": 7.567290283645142,
+      "grad_norm": 0.4799748361110687,
+      "learning_rate": 6.402491332045683e-05,
+      "loss": 0.08672296404838561,
+      "mean_token_accuracy": 0.9698824048042297,
+      "num_tokens": 7274881.0,
+      "step": 3140
+    },
+    {
+      "epoch": 7.567290283645142,
+      "eval_entropy": 0.26657010866015146,
+      "eval_loss": 0.9056380987167358,
+      "eval_mean_token_accuracy": 0.8272326461384806,
+      "eval_num_tokens": 7274881.0,
+      "eval_runtime": 88.2967,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3140
+    },
+    {
+      "entropy": 0.11890473761595785,
+      "epoch": 7.615570307785154,
+      "grad_norm": 0.4081912934780121,
+      "learning_rate": 6.166397771942987e-05,
+      "loss": 0.09396827220916748,
+      "mean_token_accuracy": 0.967512384802103,
+      "num_tokens": 7321163.0,
+      "step": 3160
+    },
+    {
+      "epoch": 7.615570307785154,
+      "eval_entropy": 0.26364776575833226,
+      "eval_loss": 0.9204733967781067,
+      "eval_mean_token_accuracy": 0.827419729714983,
+      "eval_num_tokens": 7321163.0,
+      "eval_runtime": 88.3441,
+      "eval_samples_per_second": 16.074,
+      "eval_steps_per_second": 2.015,
+      "step": 3160
+    },
+    {
+      "entropy": 0.12209134344011545,
+      "epoch": 7.663850331925166,
+      "grad_norm": 0.5159774422645569,
+      "learning_rate": 5.93388342896962e-05,
+      "loss": 0.09292978644371033,
+      "mean_token_accuracy": 0.9658421002328396,
+      "num_tokens": 7366217.0,
+      "step": 3180
+    },
+    {
+      "epoch": 7.663850331925166,
+      "eval_entropy": 0.2714110068725736,
+      "eval_loss": 0.9143092036247253,
+      "eval_mean_token_accuracy": 0.8262251099843657,
+      "eval_num_tokens": 7366217.0,
+      "eval_runtime": 87.9635,
+      "eval_samples_per_second": 16.143,
+      "eval_steps_per_second": 2.024,
+      "step": 3180
+    },
+    {
+      "entropy": 0.11955609126016498,
+      "epoch": 7.712130356065178,
+      "grad_norm": 0.7005976438522339,
+      "learning_rate": 5.705014101934252e-05,
+      "loss": 0.09319776892662049,
+      "mean_token_accuracy": 0.9674513377249241,
+      "num_tokens": 7410887.0,
+      "step": 3200
+    },
+    {
+      "epoch": 7.712130356065178,
+      "eval_entropy": 0.2626291513442993,
+      "eval_loss": 0.91884446144104,
+      "eval_mean_token_accuracy": 0.827563955207889,
+      "eval_num_tokens": 7410887.0,
+      "eval_runtime": 87.8849,
+      "eval_samples_per_second": 16.157,
+      "eval_steps_per_second": 2.025,
+      "step": 3200
+    },
+    {
+      "entropy": 0.11767532899975777,
+      "epoch": 7.76041038020519,
+      "grad_norm": 0.4675362706184387,
+      "learning_rate": 5.479854558149187e-05,
+      "loss": 0.09038087725639343,
+      "mean_token_accuracy": 0.9669680051505566,
+      "num_tokens": 7457578.0,
+      "step": 3220
+    },
+    {
+      "epoch": 7.76041038020519,
+      "eval_entropy": 0.26581037203582486,
+      "eval_loss": 0.9146883487701416,
+      "eval_mean_token_accuracy": 0.8270345812433222,
+      "eval_num_tokens": 7457578.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 3220
+    },
+    {
+      "entropy": 0.11383061073720455,
+      "epoch": 7.808690404345202,
+      "grad_norm": 0.541247546672821,
+      "learning_rate": 5.258468515101918e-05,
+      "loss": 0.08942405581474304,
+      "mean_token_accuracy": 0.968398705124855,
+      "num_tokens": 7505761.0,
+      "step": 3240
+    },
+    {
+      "epoch": 7.808690404345202,
+      "eval_entropy": 0.26687998960861997,
+      "eval_loss": 0.9037085175514221,
+      "eval_mean_token_accuracy": 0.8274977863504646,
+      "eval_num_tokens": 7505761.0,
+      "eval_runtime": 88.5163,
+      "eval_samples_per_second": 16.042,
+      "eval_steps_per_second": 2.011,
+      "step": 3240
+    },
+    {
+      "entropy": 0.12004058873280883,
+      "epoch": 7.856970428485214,
+      "grad_norm": 0.5289552211761475,
+      "learning_rate": 5.040918622423899e-05,
+      "loss": 0.09064805507659912,
+      "mean_token_accuracy": 0.9673330299556255,
+      "num_tokens": 7552557.0,
+      "step": 3260
+    },
+    {
+      "epoch": 7.856970428485214,
+      "eval_entropy": 0.26288379251622085,
+      "eval_loss": 0.9146178364753723,
+      "eval_mean_token_accuracy": 0.8276662448149049,
+      "eval_num_tokens": 7552557.0,
+      "eval_runtime": 88.0723,
+      "eval_samples_per_second": 16.123,
+      "eval_steps_per_second": 2.021,
+      "step": 3260
+    },
+    {
+      "entropy": 0.1180689456872642,
+      "epoch": 7.905250452625227,
+      "grad_norm": 0.4754861891269684,
+      "learning_rate": 4.827266444161377e-05,
+      "loss": 0.09402655363082886,
+      "mean_token_accuracy": 0.9663361787796021,
+      "num_tokens": 7596998.0,
+      "step": 3280
+    },
+    {
+      "epoch": 7.905250452625227,
+      "eval_entropy": 0.261690996922134,
+      "eval_loss": 0.9120186567306519,
+      "eval_mean_token_accuracy": 0.8278995015647974,
+      "eval_num_tokens": 7596998.0,
+      "eval_runtime": 88.6573,
+      "eval_samples_per_second": 16.017,
+      "eval_steps_per_second": 2.008,
+      "step": 3280
+    },
+    {
+      "entropy": 0.1162164019420743,
+      "epoch": 7.953530476765239,
+      "grad_norm": 0.6203398108482361,
+      "learning_rate": 4.61757244135355e-05,
+      "loss": 0.08832768201828003,
+      "mean_token_accuracy": 0.9694220684468746,
+      "num_tokens": 7643288.0,
+      "step": 3300
+    },
+    {
+      "epoch": 7.953530476765239,
+      "eval_entropy": 0.26158678129817664,
+      "eval_loss": 0.9301361441612244,
+      "eval_mean_token_accuracy": 0.8269603885291668,
+      "eval_num_tokens": 7643288.0,
+      "eval_runtime": 88.7064,
+      "eval_samples_per_second": 16.008,
+      "eval_steps_per_second": 2.007,
+      "step": 3300
+    },
+    {
+      "entropy": 0.10877202204489089,
+      "epoch": 8.0,
+      "grad_norm": 3.058673143386841,
+      "learning_rate": 4.411895954922763e-05,
+      "loss": 0.09159615635871887,
+      "mean_token_accuracy": 0.9686878834451947,
+      "num_tokens": 7688608.0,
+      "step": 3320
+    },
+    {
+      "epoch": 8.0,
+      "eval_entropy": 0.2590438057197614,
+      "eval_loss": 0.9262065887451172,
+      "eval_mean_token_accuracy": 0.8277404546067956,
+      "eval_num_tokens": 7688608.0,
+      "eval_runtime": 87.7557,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 3320
+    },
+    {
+      "entropy": 0.09969092034734786,
+      "epoch": 8.048280024140013,
+      "grad_norm": 0.2561911642551422,
+      "learning_rate": 4.2102951888817644e-05,
+      "loss": 0.07228735685348511,
+      "mean_token_accuracy": 0.9743556551635265,
+      "num_tokens": 7732856.0,
+      "step": 3340
+    },
+    {
+      "epoch": 8.048280024140013,
+      "eval_entropy": 0.2462534454933713,
+      "eval_loss": 0.9755983352661133,
+      "eval_mean_token_accuracy": 0.827510209565752,
+      "eval_num_tokens": 7732856.0,
+      "eval_runtime": 88.0297,
+      "eval_samples_per_second": 16.131,
+      "eval_steps_per_second": 2.022,
+      "step": 3340
+    },
+    {
+      "entropy": 0.09477624730207027,
+      "epoch": 8.096560048280024,
+      "grad_norm": 0.3670356571674347,
+      "learning_rate": 4.012827193862633e-05,
+      "loss": 0.0686036467552185,
+      "mean_token_accuracy": 0.9743670664727688,
+      "num_tokens": 7778585.0,
+      "step": 3360
+    },
+    {
+      "epoch": 8.096560048280024,
+      "eval_entropy": 0.2440882951356052,
+      "eval_loss": 1.0118423700332642,
+      "eval_mean_token_accuracy": 0.8271241070849172,
+      "eval_num_tokens": 7778585.0,
+      "eval_runtime": 88.1364,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 3360
+    },
+    {
+      "entropy": 0.09417980071157217,
+      "epoch": 8.144840072420036,
+      "grad_norm": 0.5785111784934998,
+      "learning_rate": 3.819547850972133e-05,
+      "loss": 0.07100698947906495,
+      "mean_token_accuracy": 0.9737804099917412,
+      "num_tokens": 7824287.0,
+      "step": 3380
+    },
+    {
+      "epoch": 8.144840072420036,
+      "eval_entropy": 0.24603172860453637,
+      "eval_loss": 1.0021991729736328,
+      "eval_mean_token_accuracy": 0.827162122793412,
+      "eval_num_tokens": 7824287.0,
+      "eval_runtime": 88.4534,
+      "eval_samples_per_second": 16.054,
+      "eval_steps_per_second": 2.012,
+      "step": 3380
+    },
+    {
+      "entropy": 0.087222247896716,
+      "epoch": 8.193120096560047,
+      "grad_norm": 0.44848504662513733,
+      "learning_rate": 3.6305118559780393e-05,
+      "loss": 0.06325234174728393,
+      "mean_token_accuracy": 0.9769335232675076,
+      "num_tokens": 7874784.0,
+      "step": 3400
+    },
+    {
+      "epoch": 8.193120096560047,
+      "eval_entropy": 0.24544863905129807,
+      "eval_loss": 1.005008339881897,
+      "eval_mean_token_accuracy": 0.8273035240976998,
+      "eval_num_tokens": 7874784.0,
+      "eval_runtime": 88.6737,
+      "eval_samples_per_second": 16.014,
+      "eval_steps_per_second": 2.007,
+      "step": 3400
+    },
+    {
+      "entropy": 0.09037275062873959,
+      "epoch": 8.24140012070006,
+      "grad_norm": 0.23529155552387238,
+      "learning_rate": 3.4457727038308435e-05,
+      "loss": 0.06589569449424744,
+      "mean_token_accuracy": 0.9751470290124417,
+      "num_tokens": 7925240.0,
+      "step": 3420
+    },
+    {
+      "epoch": 8.24140012070006,
+      "eval_entropy": 0.24203423494368456,
+      "eval_loss": 1.0052824020385742,
+      "eval_mean_token_accuracy": 0.8275575503874361,
+      "eval_num_tokens": 7925240.0,
+      "eval_runtime": 88.3486,
+      "eval_samples_per_second": 16.073,
+      "eval_steps_per_second": 2.015,
+      "step": 3420
+    },
+    {
+      "entropy": 0.10373208015225828,
+      "epoch": 8.289680144840073,
+      "grad_norm": 0.24796408414840698,
+      "learning_rate": 3.2653826735253594e-05,
+      "loss": 0.07714771628379821,
+      "mean_token_accuracy": 0.9722891598939896,
+      "num_tokens": 7966685.0,
+      "step": 3440
+    },
+    {
+      "epoch": 8.289680144840073,
+      "eval_entropy": 0.23461157050025597,
+      "eval_loss": 1.012108325958252,
+      "eval_mean_token_accuracy": 0.82825900564033,
+      "eval_num_tokens": 7966685.0,
+      "eval_runtime": 88.1018,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 3440
+    },
+    {
+      "entropy": 0.08967954274266958,
+      "epoch": 8.337960168980084,
+      "grad_norm": 0.5314674973487854,
+      "learning_rate": 3.08939281330635e-05,
+      "loss": 0.06746266484260559,
+      "mean_token_accuracy": 0.9753133699297905,
+      "num_tokens": 8013656.0,
+      "step": 3460
+    },
+    {
+      "epoch": 8.337960168980084,
+      "eval_entropy": 0.23781998589467468,
+      "eval_loss": 1.0119913816452026,
+      "eval_mean_token_accuracy": 0.8273287157664139,
+      "eval_num_tokens": 8013656.0,
+      "eval_runtime": 88.6038,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3460
+    },
+    {
+      "entropy": 0.09280927330255509,
+      "epoch": 8.386240193120097,
+      "grad_norm": 0.30658936500549316,
+      "learning_rate": 2.9178529262224946e-05,
+      "loss": 0.06936421990394592,
+      "mean_token_accuracy": 0.9742105580866337,
+      "num_tokens": 8061108.0,
+      "step": 3480
+    },
+    {
+      "epoch": 8.386240193120097,
+      "eval_entropy": 0.23672426734747512,
+      "eval_loss": 1.0162041187286377,
+      "eval_mean_token_accuracy": 0.8278394281194451,
+      "eval_num_tokens": 8061108.0,
+      "eval_runtime": 88.6015,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 3480
+    },
+    {
+      "entropy": 0.09195122267119586,
+      "epoch": 8.43452021726011,
+      "grad_norm": 0.24512454867362976,
+      "learning_rate": 2.7508115560327256e-05,
+      "loss": 0.06996064782142639,
+      "mean_token_accuracy": 0.974104854464531,
+      "num_tokens": 8106924.0,
+      "step": 3500
+    },
+    {
+      "epoch": 8.43452021726011,
+      "eval_entropy": 0.23758245828781235,
+      "eval_loss": 1.0157747268676758,
+      "eval_mean_token_accuracy": 0.8278501579600773,
+      "eval_num_tokens": 8106924.0,
+      "eval_runtime": 88.5678,
+      "eval_samples_per_second": 16.033,
+      "eval_steps_per_second": 2.01,
+      "step": 3500
+    },
+    {
+      "entropy": 0.09518477590754629,
+      "epoch": 8.48280024140012,
+      "grad_norm": 0.2373228371143341,
+      "learning_rate": 2.5883159734688868e-05,
+      "loss": 0.06832797527313232,
+      "mean_token_accuracy": 0.9743806682527065,
+      "num_tokens": 8153070.0,
+      "step": 3520
+    },
+    {
+      "epoch": 8.48280024140012,
+      "eval_entropy": 0.2379308074210467,
+      "eval_loss": 1.0180530548095703,
+      "eval_mean_token_accuracy": 0.8282477052024241,
+      "eval_num_tokens": 8153070.0,
+      "eval_runtime": 88.4651,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 3520
+    },
+    {
+      "entropy": 0.0906269280705601,
+      "epoch": 8.531080265540133,
+      "grad_norm": 0.37602657079696655,
+      "learning_rate": 2.4304121628587173e-05,
+      "loss": 0.06921588778495788,
+      "mean_token_accuracy": 0.9745571479201317,
+      "num_tokens": 8200066.0,
+      "step": 3540
+    },
+    {
+      "epoch": 8.531080265540133,
+      "eval_entropy": 0.2357490769263064,
+      "eval_loss": 1.017139196395874,
+      "eval_mean_token_accuracy": 0.828713178299786,
+      "eval_num_tokens": 8200066.0,
+      "eval_runtime": 88.6424,
+      "eval_samples_per_second": 16.019,
+      "eval_steps_per_second": 2.008,
+      "step": 3540
+    },
+    {
+      "entropy": 0.09273334196768701,
+      "epoch": 8.579360289680144,
+      "grad_norm": 0.2978838086128235,
+      "learning_rate": 2.277144809112794e-05,
+      "loss": 0.06970887780189514,
+      "mean_token_accuracy": 0.9751374177634716,
+      "num_tokens": 8246157.0,
+      "step": 3560
+    },
+    {
+      "epoch": 8.579360289680144,
+      "eval_entropy": 0.23590933138065123,
+      "eval_loss": 1.019231915473938,
+      "eval_mean_token_accuracy": 0.8279163412163767,
+      "eval_num_tokens": 8246157.0,
+      "eval_runtime": 88.2691,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 3560
+    },
+    {
+      "entropy": 0.09854201921261847,
+      "epoch": 8.627640313820157,
+      "grad_norm": 0.5381243228912354,
+      "learning_rate": 2.1285572850792413e-05,
+      "loss": 0.07123501300811767,
+      "mean_token_accuracy": 0.9723988175392151,
+      "num_tokens": 8291752.0,
+      "step": 3580
+    },
+    {
+      "epoch": 8.627640313820157,
+      "eval_entropy": 0.23592024949494372,
+      "eval_loss": 1.020250678062439,
+      "eval_mean_token_accuracy": 0.8282466226079491,
+      "eval_num_tokens": 8291752.0,
+      "eval_runtime": 87.8939,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 3580
+    },
+    {
+      "entropy": 0.09092000788077712,
+      "epoch": 8.67592033796017,
+      "grad_norm": 0.29089561104774475,
+      "learning_rate": 1.9846916392697493e-05,
+      "loss": 0.06982720494270325,
+      "mean_token_accuracy": 0.9749995484948158,
+      "num_tokens": 8337651.0,
+      "step": 3600
+    },
+    {
+      "epoch": 8.67592033796017,
+      "eval_entropy": 0.23658913382318583,
+      "eval_loss": 1.019862413406372,
+      "eval_mean_token_accuracy": 0.8280127842104836,
+      "eval_num_tokens": 8337651.0,
+      "eval_runtime": 88.2855,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 3600
+    },
+    {
+      "entropy": 0.08708424470387399,
+      "epoch": 8.72420036210018,
+      "grad_norm": 0.29904839396476746,
+      "learning_rate": 1.8455885839603106e-05,
+      "loss": 0.06801370978355407,
+      "mean_token_accuracy": 0.9755801670253277,
+      "num_tokens": 8385330.0,
+      "step": 3620
+    },
+    {
+      "epoch": 8.72420036210018,
+      "eval_entropy": 0.23627035771862845,
+      "eval_loss": 1.0204569101333618,
+      "eval_mean_token_accuracy": 0.8279967894045155,
+      "eval_num_tokens": 8385330.0,
+      "eval_runtime": 88.1753,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 3620
+    },
+    {
+      "entropy": 0.0916255566291511,
+      "epoch": 8.772480386240193,
+      "grad_norm": 0.20571033656597137,
+      "learning_rate": 1.7112874836701502e-05,
+      "loss": 0.0692249894142151,
+      "mean_token_accuracy": 0.9745148062705994,
+      "num_tokens": 8432134.0,
+      "step": 3640
+    },
+    {
+      "epoch": 8.772480386240193,
+      "eval_entropy": 0.2340821772813797,
+      "eval_loss": 1.0212821960449219,
+      "eval_mean_token_accuracy": 0.828373622358515,
+      "eval_num_tokens": 8432134.0,
+      "eval_runtime": 87.2518,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 3640
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.566340603910103e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3660/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3660/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3660/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3660/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3660/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3660/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3660/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3660/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3660/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3660/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3660/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3660/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3660/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3660/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..ecf670b7612b5ded3d0bcd29236d313f48d37a97
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3660/trainer_state.json
@@ -0,0 +1,3877 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 8.820760410380204,
+  "eval_steps": 20,
+  "global_step": 3660,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    },
+    {
+      "entropy": 0.1326121488586068,
+      "epoch": 7.181050090525045,
+      "grad_norm": 0.4843117296695709,
+      "learning_rate": 8.411870386507724e-05,
+      "loss": 0.09537842869758606,
+      "mean_token_accuracy": 0.9645151920616627,
+      "num_tokens": 6898444.0,
+      "step": 2980
+    },
+    {
+      "epoch": 7.181050090525045,
+      "eval_entropy": 0.27999477783280813,
+      "eval_loss": 0.9031037092208862,
+      "eval_mean_token_accuracy": 0.8261882204018282,
+      "eval_num_tokens": 6898444.0,
+      "eval_runtime": 88.5602,
+      "eval_samples_per_second": 16.034,
+      "eval_steps_per_second": 2.01,
+      "step": 2980
+    },
+    {
+      "entropy": 0.1156751093454659,
+      "epoch": 7.229330114665057,
+      "grad_norm": 0.5297922492027283,
+      "learning_rate": 8.149629351430326e-05,
+      "loss": 0.09203402996063233,
+      "mean_token_accuracy": 0.9689733698964119,
+      "num_tokens": 6947106.0,
+      "step": 3000
+    },
+    {
+      "epoch": 7.229330114665057,
+      "eval_entropy": 0.2740225845508361,
+      "eval_loss": 0.8876440525054932,
+      "eval_mean_token_accuracy": 0.8265116857678703,
+      "eval_num_tokens": 6947106.0,
+      "eval_runtime": 87.7971,
+      "eval_samples_per_second": 16.174,
+      "eval_steps_per_second": 2.027,
+      "step": 3000
+    },
+    {
+      "entropy": 0.12116114404052496,
+      "epoch": 7.27761013880507,
+      "grad_norm": 0.5467398166656494,
+      "learning_rate": 7.890406302404618e-05,
+      "loss": 0.0933056652545929,
+      "mean_token_accuracy": 0.9676455594599247,
+      "num_tokens": 6992784.0,
+      "step": 3020
+    },
+    {
+      "epoch": 7.27761013880507,
+      "eval_entropy": 0.27238342793804876,
+      "eval_loss": 0.9166117310523987,
+      "eval_mean_token_accuracy": 0.825499015242866,
+      "eval_num_tokens": 6992784.0,
+      "eval_runtime": 88.3276,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 3020
+    },
+    {
+      "entropy": 0.11715119928121567,
+      "epoch": 7.325890162945082,
+      "grad_norm": 0.4976702332496643,
+      "learning_rate": 7.634274596487176e-05,
+      "loss": 0.09449659585952759,
+      "mean_token_accuracy": 0.966506478190422,
+      "num_tokens": 7040482.0,
+      "step": 3040
+    },
+    {
+      "epoch": 7.325890162945082,
+      "eval_entropy": 0.2788975556077582,
+      "eval_loss": 0.8833065629005432,
+      "eval_mean_token_accuracy": 0.8269920596915684,
+      "eval_num_tokens": 7040482.0,
+      "eval_runtime": 88.9016,
+      "eval_samples_per_second": 15.973,
+      "eval_steps_per_second": 2.002,
+      "step": 3040
+    },
+    {
+      "entropy": 0.12565911412239075,
+      "epoch": 7.3741701870850935,
+      "grad_norm": 0.5728679895401001,
+      "learning_rate": 7.381306715921025e-05,
+      "loss": 0.09626244902610778,
+      "mean_token_accuracy": 0.9661444149911403,
+      "num_tokens": 7086261.0,
+      "step": 3060
+    },
+    {
+      "epoch": 7.3741701870850935,
+      "eval_entropy": 0.2762492011939542,
+      "eval_loss": 0.8961160778999329,
+      "eval_mean_token_accuracy": 0.826369910092836,
+      "eval_num_tokens": 7086261.0,
+      "eval_runtime": 88.5204,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 3060
+    },
+    {
+      "entropy": 0.12258567921817302,
+      "epoch": 7.422450211225105,
+      "grad_norm": 0.468985915184021,
+      "learning_rate": 7.131574247624031e-05,
+      "loss": 0.09120266437530518,
+      "mean_token_accuracy": 0.9674938447773457,
+      "num_tokens": 7133247.0,
+      "step": 3080
+    },
+    {
+      "epoch": 7.422450211225105,
+      "eval_entropy": 0.2688512086366,
+      "eval_loss": 0.9048654437065125,
+      "eval_mean_token_accuracy": 0.8265582302982888,
+      "eval_num_tokens": 7133247.0,
+      "eval_runtime": 88.164,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 3080
+    },
+    {
+      "entropy": 0.12070536343380808,
+      "epoch": 7.470730235365117,
+      "grad_norm": 0.5310841798782349,
+      "learning_rate": 6.885147862930688e-05,
+      "loss": 0.09386247992515565,
+      "mean_token_accuracy": 0.96664222702384,
+      "num_tokens": 7177810.0,
+      "step": 3100
+    },
+    {
+      "epoch": 7.470730235365117,
+      "eval_entropy": 0.26289632874593305,
+      "eval_loss": 0.902837336063385,
+      "eval_mean_token_accuracy": 0.828052184554968,
+      "eval_num_tokens": 7177810.0,
+      "eval_runtime": 88.1201,
+      "eval_samples_per_second": 16.114,
+      "eval_steps_per_second": 2.02,
+      "step": 3100
+    },
+    {
+      "entropy": 0.11322665181942285,
+      "epoch": 7.51901025950513,
+      "grad_norm": 0.6846067905426025,
+      "learning_rate": 6.642097297592908e-05,
+      "loss": 0.08772538304328918,
+      "mean_token_accuracy": 0.9691362135112286,
+      "num_tokens": 7225302.0,
+      "step": 3120
+    },
+    {
+      "epoch": 7.51901025950513,
+      "eval_entropy": 0.2670605613776807,
+      "eval_loss": 0.9130566120147705,
+      "eval_mean_token_accuracy": 0.826366903741708,
+      "eval_num_tokens": 7225302.0,
+      "eval_runtime": 88.6035,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3120
+    },
+    {
+      "entropy": 0.10985926580615342,
+      "epoch": 7.567290283645142,
+      "grad_norm": 0.4799748361110687,
+      "learning_rate": 6.402491332045683e-05,
+      "loss": 0.08672296404838561,
+      "mean_token_accuracy": 0.9698824048042297,
+      "num_tokens": 7274881.0,
+      "step": 3140
+    },
+    {
+      "epoch": 7.567290283645142,
+      "eval_entropy": 0.26657010866015146,
+      "eval_loss": 0.9056380987167358,
+      "eval_mean_token_accuracy": 0.8272326461384806,
+      "eval_num_tokens": 7274881.0,
+      "eval_runtime": 88.2967,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3140
+    },
+    {
+      "entropy": 0.11890473761595785,
+      "epoch": 7.615570307785154,
+      "grad_norm": 0.4081912934780121,
+      "learning_rate": 6.166397771942987e-05,
+      "loss": 0.09396827220916748,
+      "mean_token_accuracy": 0.967512384802103,
+      "num_tokens": 7321163.0,
+      "step": 3160
+    },
+    {
+      "epoch": 7.615570307785154,
+      "eval_entropy": 0.26364776575833226,
+      "eval_loss": 0.9204733967781067,
+      "eval_mean_token_accuracy": 0.827419729714983,
+      "eval_num_tokens": 7321163.0,
+      "eval_runtime": 88.3441,
+      "eval_samples_per_second": 16.074,
+      "eval_steps_per_second": 2.015,
+      "step": 3160
+    },
+    {
+      "entropy": 0.12209134344011545,
+      "epoch": 7.663850331925166,
+      "grad_norm": 0.5159774422645569,
+      "learning_rate": 5.93388342896962e-05,
+      "loss": 0.09292978644371033,
+      "mean_token_accuracy": 0.9658421002328396,
+      "num_tokens": 7366217.0,
+      "step": 3180
+    },
+    {
+      "epoch": 7.663850331925166,
+      "eval_entropy": 0.2714110068725736,
+      "eval_loss": 0.9143092036247253,
+      "eval_mean_token_accuracy": 0.8262251099843657,
+      "eval_num_tokens": 7366217.0,
+      "eval_runtime": 87.9635,
+      "eval_samples_per_second": 16.143,
+      "eval_steps_per_second": 2.024,
+      "step": 3180
+    },
+    {
+      "entropy": 0.11955609126016498,
+      "epoch": 7.712130356065178,
+      "grad_norm": 0.7005976438522339,
+      "learning_rate": 5.705014101934252e-05,
+      "loss": 0.09319776892662049,
+      "mean_token_accuracy": 0.9674513377249241,
+      "num_tokens": 7410887.0,
+      "step": 3200
+    },
+    {
+      "epoch": 7.712130356065178,
+      "eval_entropy": 0.2626291513442993,
+      "eval_loss": 0.91884446144104,
+      "eval_mean_token_accuracy": 0.827563955207889,
+      "eval_num_tokens": 7410887.0,
+      "eval_runtime": 87.8849,
+      "eval_samples_per_second": 16.157,
+      "eval_steps_per_second": 2.025,
+      "step": 3200
+    },
+    {
+      "entropy": 0.11767532899975777,
+      "epoch": 7.76041038020519,
+      "grad_norm": 0.4675362706184387,
+      "learning_rate": 5.479854558149187e-05,
+      "loss": 0.09038087725639343,
+      "mean_token_accuracy": 0.9669680051505566,
+      "num_tokens": 7457578.0,
+      "step": 3220
+    },
+    {
+      "epoch": 7.76041038020519,
+      "eval_entropy": 0.26581037203582486,
+      "eval_loss": 0.9146883487701416,
+      "eval_mean_token_accuracy": 0.8270345812433222,
+      "eval_num_tokens": 7457578.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 3220
+    },
+    {
+      "entropy": 0.11383061073720455,
+      "epoch": 7.808690404345202,
+      "grad_norm": 0.541247546672821,
+      "learning_rate": 5.258468515101918e-05,
+      "loss": 0.08942405581474304,
+      "mean_token_accuracy": 0.968398705124855,
+      "num_tokens": 7505761.0,
+      "step": 3240
+    },
+    {
+      "epoch": 7.808690404345202,
+      "eval_entropy": 0.26687998960861997,
+      "eval_loss": 0.9037085175514221,
+      "eval_mean_token_accuracy": 0.8274977863504646,
+      "eval_num_tokens": 7505761.0,
+      "eval_runtime": 88.5163,
+      "eval_samples_per_second": 16.042,
+      "eval_steps_per_second": 2.011,
+      "step": 3240
+    },
+    {
+      "entropy": 0.12004058873280883,
+      "epoch": 7.856970428485214,
+      "grad_norm": 0.5289552211761475,
+      "learning_rate": 5.040918622423899e-05,
+      "loss": 0.09064805507659912,
+      "mean_token_accuracy": 0.9673330299556255,
+      "num_tokens": 7552557.0,
+      "step": 3260
+    },
+    {
+      "epoch": 7.856970428485214,
+      "eval_entropy": 0.26288379251622085,
+      "eval_loss": 0.9146178364753723,
+      "eval_mean_token_accuracy": 0.8276662448149049,
+      "eval_num_tokens": 7552557.0,
+      "eval_runtime": 88.0723,
+      "eval_samples_per_second": 16.123,
+      "eval_steps_per_second": 2.021,
+      "step": 3260
+    },
+    {
+      "entropy": 0.1180689456872642,
+      "epoch": 7.905250452625227,
+      "grad_norm": 0.4754861891269684,
+      "learning_rate": 4.827266444161377e-05,
+      "loss": 0.09402655363082886,
+      "mean_token_accuracy": 0.9663361787796021,
+      "num_tokens": 7596998.0,
+      "step": 3280
+    },
+    {
+      "epoch": 7.905250452625227,
+      "eval_entropy": 0.261690996922134,
+      "eval_loss": 0.9120186567306519,
+      "eval_mean_token_accuracy": 0.8278995015647974,
+      "eval_num_tokens": 7596998.0,
+      "eval_runtime": 88.6573,
+      "eval_samples_per_second": 16.017,
+      "eval_steps_per_second": 2.008,
+      "step": 3280
+    },
+    {
+      "entropy": 0.1162164019420743,
+      "epoch": 7.953530476765239,
+      "grad_norm": 0.6203398108482361,
+      "learning_rate": 4.61757244135355e-05,
+      "loss": 0.08832768201828003,
+      "mean_token_accuracy": 0.9694220684468746,
+      "num_tokens": 7643288.0,
+      "step": 3300
+    },
+    {
+      "epoch": 7.953530476765239,
+      "eval_entropy": 0.26158678129817664,
+      "eval_loss": 0.9301361441612244,
+      "eval_mean_token_accuracy": 0.8269603885291668,
+      "eval_num_tokens": 7643288.0,
+      "eval_runtime": 88.7064,
+      "eval_samples_per_second": 16.008,
+      "eval_steps_per_second": 2.007,
+      "step": 3300
+    },
+    {
+      "entropy": 0.10877202204489089,
+      "epoch": 8.0,
+      "grad_norm": 3.058673143386841,
+      "learning_rate": 4.411895954922763e-05,
+      "loss": 0.09159615635871887,
+      "mean_token_accuracy": 0.9686878834451947,
+      "num_tokens": 7688608.0,
+      "step": 3320
+    },
+    {
+      "epoch": 8.0,
+      "eval_entropy": 0.2590438057197614,
+      "eval_loss": 0.9262065887451172,
+      "eval_mean_token_accuracy": 0.8277404546067956,
+      "eval_num_tokens": 7688608.0,
+      "eval_runtime": 87.7557,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 3320
+    },
+    {
+      "entropy": 0.09969092034734786,
+      "epoch": 8.048280024140013,
+      "grad_norm": 0.2561911642551422,
+      "learning_rate": 4.2102951888817644e-05,
+      "loss": 0.07228735685348511,
+      "mean_token_accuracy": 0.9743556551635265,
+      "num_tokens": 7732856.0,
+      "step": 3340
+    },
+    {
+      "epoch": 8.048280024140013,
+      "eval_entropy": 0.2462534454933713,
+      "eval_loss": 0.9755983352661133,
+      "eval_mean_token_accuracy": 0.827510209565752,
+      "eval_num_tokens": 7732856.0,
+      "eval_runtime": 88.0297,
+      "eval_samples_per_second": 16.131,
+      "eval_steps_per_second": 2.022,
+      "step": 3340
+    },
+    {
+      "entropy": 0.09477624730207027,
+      "epoch": 8.096560048280024,
+      "grad_norm": 0.3670356571674347,
+      "learning_rate": 4.012827193862633e-05,
+      "loss": 0.0686036467552185,
+      "mean_token_accuracy": 0.9743670664727688,
+      "num_tokens": 7778585.0,
+      "step": 3360
+    },
+    {
+      "epoch": 8.096560048280024,
+      "eval_entropy": 0.2440882951356052,
+      "eval_loss": 1.0118423700332642,
+      "eval_mean_token_accuracy": 0.8271241070849172,
+      "eval_num_tokens": 7778585.0,
+      "eval_runtime": 88.1364,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 3360
+    },
+    {
+      "entropy": 0.09417980071157217,
+      "epoch": 8.144840072420036,
+      "grad_norm": 0.5785111784934998,
+      "learning_rate": 3.819547850972133e-05,
+      "loss": 0.07100698947906495,
+      "mean_token_accuracy": 0.9737804099917412,
+      "num_tokens": 7824287.0,
+      "step": 3380
+    },
+    {
+      "epoch": 8.144840072420036,
+      "eval_entropy": 0.24603172860453637,
+      "eval_loss": 1.0021991729736328,
+      "eval_mean_token_accuracy": 0.827162122793412,
+      "eval_num_tokens": 7824287.0,
+      "eval_runtime": 88.4534,
+      "eval_samples_per_second": 16.054,
+      "eval_steps_per_second": 2.012,
+      "step": 3380
+    },
+    {
+      "entropy": 0.087222247896716,
+      "epoch": 8.193120096560047,
+      "grad_norm": 0.44848504662513733,
+      "learning_rate": 3.6305118559780393e-05,
+      "loss": 0.06325234174728393,
+      "mean_token_accuracy": 0.9769335232675076,
+      "num_tokens": 7874784.0,
+      "step": 3400
+    },
+    {
+      "epoch": 8.193120096560047,
+      "eval_entropy": 0.24544863905129807,
+      "eval_loss": 1.005008339881897,
+      "eval_mean_token_accuracy": 0.8273035240976998,
+      "eval_num_tokens": 7874784.0,
+      "eval_runtime": 88.6737,
+      "eval_samples_per_second": 16.014,
+      "eval_steps_per_second": 2.007,
+      "step": 3400
+    },
+    {
+      "entropy": 0.09037275062873959,
+      "epoch": 8.24140012070006,
+      "grad_norm": 0.23529155552387238,
+      "learning_rate": 3.4457727038308435e-05,
+      "loss": 0.06589569449424744,
+      "mean_token_accuracy": 0.9751470290124417,
+      "num_tokens": 7925240.0,
+      "step": 3420
+    },
+    {
+      "epoch": 8.24140012070006,
+      "eval_entropy": 0.24203423494368456,
+      "eval_loss": 1.0052824020385742,
+      "eval_mean_token_accuracy": 0.8275575503874361,
+      "eval_num_tokens": 7925240.0,
+      "eval_runtime": 88.3486,
+      "eval_samples_per_second": 16.073,
+      "eval_steps_per_second": 2.015,
+      "step": 3420
+    },
+    {
+      "entropy": 0.10373208015225828,
+      "epoch": 8.289680144840073,
+      "grad_norm": 0.24796408414840698,
+      "learning_rate": 3.2653826735253594e-05,
+      "loss": 0.07714771628379821,
+      "mean_token_accuracy": 0.9722891598939896,
+      "num_tokens": 7966685.0,
+      "step": 3440
+    },
+    {
+      "epoch": 8.289680144840073,
+      "eval_entropy": 0.23461157050025597,
+      "eval_loss": 1.012108325958252,
+      "eval_mean_token_accuracy": 0.82825900564033,
+      "eval_num_tokens": 7966685.0,
+      "eval_runtime": 88.1018,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 3440
+    },
+    {
+      "entropy": 0.08967954274266958,
+      "epoch": 8.337960168980084,
+      "grad_norm": 0.5314674973487854,
+      "learning_rate": 3.08939281330635e-05,
+      "loss": 0.06746266484260559,
+      "mean_token_accuracy": 0.9753133699297905,
+      "num_tokens": 8013656.0,
+      "step": 3460
+    },
+    {
+      "epoch": 8.337960168980084,
+      "eval_entropy": 0.23781998589467468,
+      "eval_loss": 1.0119913816452026,
+      "eval_mean_token_accuracy": 0.8273287157664139,
+      "eval_num_tokens": 8013656.0,
+      "eval_runtime": 88.6038,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3460
+    },
+    {
+      "entropy": 0.09280927330255509,
+      "epoch": 8.386240193120097,
+      "grad_norm": 0.30658936500549316,
+      "learning_rate": 2.9178529262224946e-05,
+      "loss": 0.06936421990394592,
+      "mean_token_accuracy": 0.9742105580866337,
+      "num_tokens": 8061108.0,
+      "step": 3480
+    },
+    {
+      "epoch": 8.386240193120097,
+      "eval_entropy": 0.23672426734747512,
+      "eval_loss": 1.0162041187286377,
+      "eval_mean_token_accuracy": 0.8278394281194451,
+      "eval_num_tokens": 8061108.0,
+      "eval_runtime": 88.6015,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 3480
+    },
+    {
+      "entropy": 0.09195122267119586,
+      "epoch": 8.43452021726011,
+      "grad_norm": 0.24512454867362976,
+      "learning_rate": 2.7508115560327256e-05,
+      "loss": 0.06996064782142639,
+      "mean_token_accuracy": 0.974104854464531,
+      "num_tokens": 8106924.0,
+      "step": 3500
+    },
+    {
+      "epoch": 8.43452021726011,
+      "eval_entropy": 0.23758245828781235,
+      "eval_loss": 1.0157747268676758,
+      "eval_mean_token_accuracy": 0.8278501579600773,
+      "eval_num_tokens": 8106924.0,
+      "eval_runtime": 88.5678,
+      "eval_samples_per_second": 16.033,
+      "eval_steps_per_second": 2.01,
+      "step": 3500
+    },
+    {
+      "entropy": 0.09518477590754629,
+      "epoch": 8.48280024140012,
+      "grad_norm": 0.2373228371143341,
+      "learning_rate": 2.5883159734688868e-05,
+      "loss": 0.06832797527313232,
+      "mean_token_accuracy": 0.9743806682527065,
+      "num_tokens": 8153070.0,
+      "step": 3520
+    },
+    {
+      "epoch": 8.48280024140012,
+      "eval_entropy": 0.2379308074210467,
+      "eval_loss": 1.0180530548095703,
+      "eval_mean_token_accuracy": 0.8282477052024241,
+      "eval_num_tokens": 8153070.0,
+      "eval_runtime": 88.4651,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 3520
+    },
+    {
+      "entropy": 0.0906269280705601,
+      "epoch": 8.531080265540133,
+      "grad_norm": 0.37602657079696655,
+      "learning_rate": 2.4304121628587173e-05,
+      "loss": 0.06921588778495788,
+      "mean_token_accuracy": 0.9745571479201317,
+      "num_tokens": 8200066.0,
+      "step": 3540
+    },
+    {
+      "epoch": 8.531080265540133,
+      "eval_entropy": 0.2357490769263064,
+      "eval_loss": 1.017139196395874,
+      "eval_mean_token_accuracy": 0.828713178299786,
+      "eval_num_tokens": 8200066.0,
+      "eval_runtime": 88.6424,
+      "eval_samples_per_second": 16.019,
+      "eval_steps_per_second": 2.008,
+      "step": 3540
+    },
+    {
+      "entropy": 0.09273334196768701,
+      "epoch": 8.579360289680144,
+      "grad_norm": 0.2978838086128235,
+      "learning_rate": 2.277144809112794e-05,
+      "loss": 0.06970887780189514,
+      "mean_token_accuracy": 0.9751374177634716,
+      "num_tokens": 8246157.0,
+      "step": 3560
+    },
+    {
+      "epoch": 8.579360289680144,
+      "eval_entropy": 0.23590933138065123,
+      "eval_loss": 1.019231915473938,
+      "eval_mean_token_accuracy": 0.8279163412163767,
+      "eval_num_tokens": 8246157.0,
+      "eval_runtime": 88.2691,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 3560
+    },
+    {
+      "entropy": 0.09854201921261847,
+      "epoch": 8.627640313820157,
+      "grad_norm": 0.5381243228912354,
+      "learning_rate": 2.1285572850792413e-05,
+      "loss": 0.07123501300811767,
+      "mean_token_accuracy": 0.9723988175392151,
+      "num_tokens": 8291752.0,
+      "step": 3580
+    },
+    {
+      "epoch": 8.627640313820157,
+      "eval_entropy": 0.23592024949494372,
+      "eval_loss": 1.020250678062439,
+      "eval_mean_token_accuracy": 0.8282466226079491,
+      "eval_num_tokens": 8291752.0,
+      "eval_runtime": 87.8939,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 3580
+    },
+    {
+      "entropy": 0.09092000788077712,
+      "epoch": 8.67592033796017,
+      "grad_norm": 0.29089561104774475,
+      "learning_rate": 1.9846916392697493e-05,
+      "loss": 0.06982720494270325,
+      "mean_token_accuracy": 0.9749995484948158,
+      "num_tokens": 8337651.0,
+      "step": 3600
+    },
+    {
+      "epoch": 8.67592033796017,
+      "eval_entropy": 0.23658913382318583,
+      "eval_loss": 1.019862413406372,
+      "eval_mean_token_accuracy": 0.8280127842104836,
+      "eval_num_tokens": 8337651.0,
+      "eval_runtime": 88.2855,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 3600
+    },
+    {
+      "entropy": 0.08708424470387399,
+      "epoch": 8.72420036210018,
+      "grad_norm": 0.29904839396476746,
+      "learning_rate": 1.8455885839603106e-05,
+      "loss": 0.06801370978355407,
+      "mean_token_accuracy": 0.9755801670253277,
+      "num_tokens": 8385330.0,
+      "step": 3620
+    },
+    {
+      "epoch": 8.72420036210018,
+      "eval_entropy": 0.23627035771862845,
+      "eval_loss": 1.0204569101333618,
+      "eval_mean_token_accuracy": 0.8279967894045155,
+      "eval_num_tokens": 8385330.0,
+      "eval_runtime": 88.1753,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 3620
+    },
+    {
+      "entropy": 0.0916255566291511,
+      "epoch": 8.772480386240193,
+      "grad_norm": 0.20571033656597137,
+      "learning_rate": 1.7112874836701502e-05,
+      "loss": 0.0692249894142151,
+      "mean_token_accuracy": 0.9745148062705994,
+      "num_tokens": 8432134.0,
+      "step": 3640
+    },
+    {
+      "epoch": 8.772480386240193,
+      "eval_entropy": 0.2340821772813797,
+      "eval_loss": 1.0212821960449219,
+      "eval_mean_token_accuracy": 0.828373622358515,
+      "eval_num_tokens": 8432134.0,
+      "eval_runtime": 87.2518,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 3640
+    },
+    {
+      "entropy": 0.0926868049427867,
+      "epoch": 8.820760410380204,
+      "grad_norm": 0.2618265450000763,
+      "learning_rate": 1.5818263440220203e-05,
+      "loss": 0.07338647246360779,
+      "mean_token_accuracy": 0.9724708944559097,
+      "num_tokens": 8476758.0,
+      "step": 3660
+    },
+    {
+      "epoch": 8.820760410380204,
+      "eval_entropy": 0.23637074575330433,
+      "eval_loss": 1.0186725854873657,
+      "eval_mean_token_accuracy": 0.8277707605549459,
+      "eval_num_tokens": 8476758.0,
+      "eval_runtime": 88.4695,
+      "eval_samples_per_second": 16.051,
+      "eval_steps_per_second": 2.012,
+      "step": 3660
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.584905597418742e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3680/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3680/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3680/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3680/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3680/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3680/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3680/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3680/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3680/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3680/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3680/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3680/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3680/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3680/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..bc82703251903e46caa7318691725fb801439c9a
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3680/trainer_state.json
@@ -0,0 +1,3898 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 8.869040434520217,
+  "eval_steps": 20,
+  "global_step": 3680,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    },
+    {
+      "entropy": 0.1326121488586068,
+      "epoch": 7.181050090525045,
+      "grad_norm": 0.4843117296695709,
+      "learning_rate": 8.411870386507724e-05,
+      "loss": 0.09537842869758606,
+      "mean_token_accuracy": 0.9645151920616627,
+      "num_tokens": 6898444.0,
+      "step": 2980
+    },
+    {
+      "epoch": 7.181050090525045,
+      "eval_entropy": 0.27999477783280813,
+      "eval_loss": 0.9031037092208862,
+      "eval_mean_token_accuracy": 0.8261882204018282,
+      "eval_num_tokens": 6898444.0,
+      "eval_runtime": 88.5602,
+      "eval_samples_per_second": 16.034,
+      "eval_steps_per_second": 2.01,
+      "step": 2980
+    },
+    {
+      "entropy": 0.1156751093454659,
+      "epoch": 7.229330114665057,
+      "grad_norm": 0.5297922492027283,
+      "learning_rate": 8.149629351430326e-05,
+      "loss": 0.09203402996063233,
+      "mean_token_accuracy": 0.9689733698964119,
+      "num_tokens": 6947106.0,
+      "step": 3000
+    },
+    {
+      "epoch": 7.229330114665057,
+      "eval_entropy": 0.2740225845508361,
+      "eval_loss": 0.8876440525054932,
+      "eval_mean_token_accuracy": 0.8265116857678703,
+      "eval_num_tokens": 6947106.0,
+      "eval_runtime": 87.7971,
+      "eval_samples_per_second": 16.174,
+      "eval_steps_per_second": 2.027,
+      "step": 3000
+    },
+    {
+      "entropy": 0.12116114404052496,
+      "epoch": 7.27761013880507,
+      "grad_norm": 0.5467398166656494,
+      "learning_rate": 7.890406302404618e-05,
+      "loss": 0.0933056652545929,
+      "mean_token_accuracy": 0.9676455594599247,
+      "num_tokens": 6992784.0,
+      "step": 3020
+    },
+    {
+      "epoch": 7.27761013880507,
+      "eval_entropy": 0.27238342793804876,
+      "eval_loss": 0.9166117310523987,
+      "eval_mean_token_accuracy": 0.825499015242866,
+      "eval_num_tokens": 6992784.0,
+      "eval_runtime": 88.3276,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 3020
+    },
+    {
+      "entropy": 0.11715119928121567,
+      "epoch": 7.325890162945082,
+      "grad_norm": 0.4976702332496643,
+      "learning_rate": 7.634274596487176e-05,
+      "loss": 0.09449659585952759,
+      "mean_token_accuracy": 0.966506478190422,
+      "num_tokens": 7040482.0,
+      "step": 3040
+    },
+    {
+      "epoch": 7.325890162945082,
+      "eval_entropy": 0.2788975556077582,
+      "eval_loss": 0.8833065629005432,
+      "eval_mean_token_accuracy": 0.8269920596915684,
+      "eval_num_tokens": 7040482.0,
+      "eval_runtime": 88.9016,
+      "eval_samples_per_second": 15.973,
+      "eval_steps_per_second": 2.002,
+      "step": 3040
+    },
+    {
+      "entropy": 0.12565911412239075,
+      "epoch": 7.3741701870850935,
+      "grad_norm": 0.5728679895401001,
+      "learning_rate": 7.381306715921025e-05,
+      "loss": 0.09626244902610778,
+      "mean_token_accuracy": 0.9661444149911403,
+      "num_tokens": 7086261.0,
+      "step": 3060
+    },
+    {
+      "epoch": 7.3741701870850935,
+      "eval_entropy": 0.2762492011939542,
+      "eval_loss": 0.8961160778999329,
+      "eval_mean_token_accuracy": 0.826369910092836,
+      "eval_num_tokens": 7086261.0,
+      "eval_runtime": 88.5204,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 3060
+    },
+    {
+      "entropy": 0.12258567921817302,
+      "epoch": 7.422450211225105,
+      "grad_norm": 0.468985915184021,
+      "learning_rate": 7.131574247624031e-05,
+      "loss": 0.09120266437530518,
+      "mean_token_accuracy": 0.9674938447773457,
+      "num_tokens": 7133247.0,
+      "step": 3080
+    },
+    {
+      "epoch": 7.422450211225105,
+      "eval_entropy": 0.2688512086366,
+      "eval_loss": 0.9048654437065125,
+      "eval_mean_token_accuracy": 0.8265582302982888,
+      "eval_num_tokens": 7133247.0,
+      "eval_runtime": 88.164,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 3080
+    },
+    {
+      "entropy": 0.12070536343380808,
+      "epoch": 7.470730235365117,
+      "grad_norm": 0.5310841798782349,
+      "learning_rate": 6.885147862930688e-05,
+      "loss": 0.09386247992515565,
+      "mean_token_accuracy": 0.96664222702384,
+      "num_tokens": 7177810.0,
+      "step": 3100
+    },
+    {
+      "epoch": 7.470730235365117,
+      "eval_entropy": 0.26289632874593305,
+      "eval_loss": 0.902837336063385,
+      "eval_mean_token_accuracy": 0.828052184554968,
+      "eval_num_tokens": 7177810.0,
+      "eval_runtime": 88.1201,
+      "eval_samples_per_second": 16.114,
+      "eval_steps_per_second": 2.02,
+      "step": 3100
+    },
+    {
+      "entropy": 0.11322665181942285,
+      "epoch": 7.51901025950513,
+      "grad_norm": 0.6846067905426025,
+      "learning_rate": 6.642097297592908e-05,
+      "loss": 0.08772538304328918,
+      "mean_token_accuracy": 0.9691362135112286,
+      "num_tokens": 7225302.0,
+      "step": 3120
+    },
+    {
+      "epoch": 7.51901025950513,
+      "eval_entropy": 0.2670605613776807,
+      "eval_loss": 0.9130566120147705,
+      "eval_mean_token_accuracy": 0.826366903741708,
+      "eval_num_tokens": 7225302.0,
+      "eval_runtime": 88.6035,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3120
+    },
+    {
+      "entropy": 0.10985926580615342,
+      "epoch": 7.567290283645142,
+      "grad_norm": 0.4799748361110687,
+      "learning_rate": 6.402491332045683e-05,
+      "loss": 0.08672296404838561,
+      "mean_token_accuracy": 0.9698824048042297,
+      "num_tokens": 7274881.0,
+      "step": 3140
+    },
+    {
+      "epoch": 7.567290283645142,
+      "eval_entropy": 0.26657010866015146,
+      "eval_loss": 0.9056380987167358,
+      "eval_mean_token_accuracy": 0.8272326461384806,
+      "eval_num_tokens": 7274881.0,
+      "eval_runtime": 88.2967,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3140
+    },
+    {
+      "entropy": 0.11890473761595785,
+      "epoch": 7.615570307785154,
+      "grad_norm": 0.4081912934780121,
+      "learning_rate": 6.166397771942987e-05,
+      "loss": 0.09396827220916748,
+      "mean_token_accuracy": 0.967512384802103,
+      "num_tokens": 7321163.0,
+      "step": 3160
+    },
+    {
+      "epoch": 7.615570307785154,
+      "eval_entropy": 0.26364776575833226,
+      "eval_loss": 0.9204733967781067,
+      "eval_mean_token_accuracy": 0.827419729714983,
+      "eval_num_tokens": 7321163.0,
+      "eval_runtime": 88.3441,
+      "eval_samples_per_second": 16.074,
+      "eval_steps_per_second": 2.015,
+      "step": 3160
+    },
+    {
+      "entropy": 0.12209134344011545,
+      "epoch": 7.663850331925166,
+      "grad_norm": 0.5159774422645569,
+      "learning_rate": 5.93388342896962e-05,
+      "loss": 0.09292978644371033,
+      "mean_token_accuracy": 0.9658421002328396,
+      "num_tokens": 7366217.0,
+      "step": 3180
+    },
+    {
+      "epoch": 7.663850331925166,
+      "eval_entropy": 0.2714110068725736,
+      "eval_loss": 0.9143092036247253,
+      "eval_mean_token_accuracy": 0.8262251099843657,
+      "eval_num_tokens": 7366217.0,
+      "eval_runtime": 87.9635,
+      "eval_samples_per_second": 16.143,
+      "eval_steps_per_second": 2.024,
+      "step": 3180
+    },
+    {
+      "entropy": 0.11955609126016498,
+      "epoch": 7.712130356065178,
+      "grad_norm": 0.7005976438522339,
+      "learning_rate": 5.705014101934252e-05,
+      "loss": 0.09319776892662049,
+      "mean_token_accuracy": 0.9674513377249241,
+      "num_tokens": 7410887.0,
+      "step": 3200
+    },
+    {
+      "epoch": 7.712130356065178,
+      "eval_entropy": 0.2626291513442993,
+      "eval_loss": 0.91884446144104,
+      "eval_mean_token_accuracy": 0.827563955207889,
+      "eval_num_tokens": 7410887.0,
+      "eval_runtime": 87.8849,
+      "eval_samples_per_second": 16.157,
+      "eval_steps_per_second": 2.025,
+      "step": 3200
+    },
+    {
+      "entropy": 0.11767532899975777,
+      "epoch": 7.76041038020519,
+      "grad_norm": 0.4675362706184387,
+      "learning_rate": 5.479854558149187e-05,
+      "loss": 0.09038087725639343,
+      "mean_token_accuracy": 0.9669680051505566,
+      "num_tokens": 7457578.0,
+      "step": 3220
+    },
+    {
+      "epoch": 7.76041038020519,
+      "eval_entropy": 0.26581037203582486,
+      "eval_loss": 0.9146883487701416,
+      "eval_mean_token_accuracy": 0.8270345812433222,
+      "eval_num_tokens": 7457578.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 3220
+    },
+    {
+      "entropy": 0.11383061073720455,
+      "epoch": 7.808690404345202,
+      "grad_norm": 0.541247546672821,
+      "learning_rate": 5.258468515101918e-05,
+      "loss": 0.08942405581474304,
+      "mean_token_accuracy": 0.968398705124855,
+      "num_tokens": 7505761.0,
+      "step": 3240
+    },
+    {
+      "epoch": 7.808690404345202,
+      "eval_entropy": 0.26687998960861997,
+      "eval_loss": 0.9037085175514221,
+      "eval_mean_token_accuracy": 0.8274977863504646,
+      "eval_num_tokens": 7505761.0,
+      "eval_runtime": 88.5163,
+      "eval_samples_per_second": 16.042,
+      "eval_steps_per_second": 2.011,
+      "step": 3240
+    },
+    {
+      "entropy": 0.12004058873280883,
+      "epoch": 7.856970428485214,
+      "grad_norm": 0.5289552211761475,
+      "learning_rate": 5.040918622423899e-05,
+      "loss": 0.09064805507659912,
+      "mean_token_accuracy": 0.9673330299556255,
+      "num_tokens": 7552557.0,
+      "step": 3260
+    },
+    {
+      "epoch": 7.856970428485214,
+      "eval_entropy": 0.26288379251622085,
+      "eval_loss": 0.9146178364753723,
+      "eval_mean_token_accuracy": 0.8276662448149049,
+      "eval_num_tokens": 7552557.0,
+      "eval_runtime": 88.0723,
+      "eval_samples_per_second": 16.123,
+      "eval_steps_per_second": 2.021,
+      "step": 3260
+    },
+    {
+      "entropy": 0.1180689456872642,
+      "epoch": 7.905250452625227,
+      "grad_norm": 0.4754861891269684,
+      "learning_rate": 4.827266444161377e-05,
+      "loss": 0.09402655363082886,
+      "mean_token_accuracy": 0.9663361787796021,
+      "num_tokens": 7596998.0,
+      "step": 3280
+    },
+    {
+      "epoch": 7.905250452625227,
+      "eval_entropy": 0.261690996922134,
+      "eval_loss": 0.9120186567306519,
+      "eval_mean_token_accuracy": 0.8278995015647974,
+      "eval_num_tokens": 7596998.0,
+      "eval_runtime": 88.6573,
+      "eval_samples_per_second": 16.017,
+      "eval_steps_per_second": 2.008,
+      "step": 3280
+    },
+    {
+      "entropy": 0.1162164019420743,
+      "epoch": 7.953530476765239,
+      "grad_norm": 0.6203398108482361,
+      "learning_rate": 4.61757244135355e-05,
+      "loss": 0.08832768201828003,
+      "mean_token_accuracy": 0.9694220684468746,
+      "num_tokens": 7643288.0,
+      "step": 3300
+    },
+    {
+      "epoch": 7.953530476765239,
+      "eval_entropy": 0.26158678129817664,
+      "eval_loss": 0.9301361441612244,
+      "eval_mean_token_accuracy": 0.8269603885291668,
+      "eval_num_tokens": 7643288.0,
+      "eval_runtime": 88.7064,
+      "eval_samples_per_second": 16.008,
+      "eval_steps_per_second": 2.007,
+      "step": 3300
+    },
+    {
+      "entropy": 0.10877202204489089,
+      "epoch": 8.0,
+      "grad_norm": 3.058673143386841,
+      "learning_rate": 4.411895954922763e-05,
+      "loss": 0.09159615635871887,
+      "mean_token_accuracy": 0.9686878834451947,
+      "num_tokens": 7688608.0,
+      "step": 3320
+    },
+    {
+      "epoch": 8.0,
+      "eval_entropy": 0.2590438057197614,
+      "eval_loss": 0.9262065887451172,
+      "eval_mean_token_accuracy": 0.8277404546067956,
+      "eval_num_tokens": 7688608.0,
+      "eval_runtime": 87.7557,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 3320
+    },
+    {
+      "entropy": 0.09969092034734786,
+      "epoch": 8.048280024140013,
+      "grad_norm": 0.2561911642551422,
+      "learning_rate": 4.2102951888817644e-05,
+      "loss": 0.07228735685348511,
+      "mean_token_accuracy": 0.9743556551635265,
+      "num_tokens": 7732856.0,
+      "step": 3340
+    },
+    {
+      "epoch": 8.048280024140013,
+      "eval_entropy": 0.2462534454933713,
+      "eval_loss": 0.9755983352661133,
+      "eval_mean_token_accuracy": 0.827510209565752,
+      "eval_num_tokens": 7732856.0,
+      "eval_runtime": 88.0297,
+      "eval_samples_per_second": 16.131,
+      "eval_steps_per_second": 2.022,
+      "step": 3340
+    },
+    {
+      "entropy": 0.09477624730207027,
+      "epoch": 8.096560048280024,
+      "grad_norm": 0.3670356571674347,
+      "learning_rate": 4.012827193862633e-05,
+      "loss": 0.0686036467552185,
+      "mean_token_accuracy": 0.9743670664727688,
+      "num_tokens": 7778585.0,
+      "step": 3360
+    },
+    {
+      "epoch": 8.096560048280024,
+      "eval_entropy": 0.2440882951356052,
+      "eval_loss": 1.0118423700332642,
+      "eval_mean_token_accuracy": 0.8271241070849172,
+      "eval_num_tokens": 7778585.0,
+      "eval_runtime": 88.1364,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 3360
+    },
+    {
+      "entropy": 0.09417980071157217,
+      "epoch": 8.144840072420036,
+      "grad_norm": 0.5785111784934998,
+      "learning_rate": 3.819547850972133e-05,
+      "loss": 0.07100698947906495,
+      "mean_token_accuracy": 0.9737804099917412,
+      "num_tokens": 7824287.0,
+      "step": 3380
+    },
+    {
+      "epoch": 8.144840072420036,
+      "eval_entropy": 0.24603172860453637,
+      "eval_loss": 1.0021991729736328,
+      "eval_mean_token_accuracy": 0.827162122793412,
+      "eval_num_tokens": 7824287.0,
+      "eval_runtime": 88.4534,
+      "eval_samples_per_second": 16.054,
+      "eval_steps_per_second": 2.012,
+      "step": 3380
+    },
+    {
+      "entropy": 0.087222247896716,
+      "epoch": 8.193120096560047,
+      "grad_norm": 0.44848504662513733,
+      "learning_rate": 3.6305118559780393e-05,
+      "loss": 0.06325234174728393,
+      "mean_token_accuracy": 0.9769335232675076,
+      "num_tokens": 7874784.0,
+      "step": 3400
+    },
+    {
+      "epoch": 8.193120096560047,
+      "eval_entropy": 0.24544863905129807,
+      "eval_loss": 1.005008339881897,
+      "eval_mean_token_accuracy": 0.8273035240976998,
+      "eval_num_tokens": 7874784.0,
+      "eval_runtime": 88.6737,
+      "eval_samples_per_second": 16.014,
+      "eval_steps_per_second": 2.007,
+      "step": 3400
+    },
+    {
+      "entropy": 0.09037275062873959,
+      "epoch": 8.24140012070006,
+      "grad_norm": 0.23529155552387238,
+      "learning_rate": 3.4457727038308435e-05,
+      "loss": 0.06589569449424744,
+      "mean_token_accuracy": 0.9751470290124417,
+      "num_tokens": 7925240.0,
+      "step": 3420
+    },
+    {
+      "epoch": 8.24140012070006,
+      "eval_entropy": 0.24203423494368456,
+      "eval_loss": 1.0052824020385742,
+      "eval_mean_token_accuracy": 0.8275575503874361,
+      "eval_num_tokens": 7925240.0,
+      "eval_runtime": 88.3486,
+      "eval_samples_per_second": 16.073,
+      "eval_steps_per_second": 2.015,
+      "step": 3420
+    },
+    {
+      "entropy": 0.10373208015225828,
+      "epoch": 8.289680144840073,
+      "grad_norm": 0.24796408414840698,
+      "learning_rate": 3.2653826735253594e-05,
+      "loss": 0.07714771628379821,
+      "mean_token_accuracy": 0.9722891598939896,
+      "num_tokens": 7966685.0,
+      "step": 3440
+    },
+    {
+      "epoch": 8.289680144840073,
+      "eval_entropy": 0.23461157050025597,
+      "eval_loss": 1.012108325958252,
+      "eval_mean_token_accuracy": 0.82825900564033,
+      "eval_num_tokens": 7966685.0,
+      "eval_runtime": 88.1018,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 3440
+    },
+    {
+      "entropy": 0.08967954274266958,
+      "epoch": 8.337960168980084,
+      "grad_norm": 0.5314674973487854,
+      "learning_rate": 3.08939281330635e-05,
+      "loss": 0.06746266484260559,
+      "mean_token_accuracy": 0.9753133699297905,
+      "num_tokens": 8013656.0,
+      "step": 3460
+    },
+    {
+      "epoch": 8.337960168980084,
+      "eval_entropy": 0.23781998589467468,
+      "eval_loss": 1.0119913816452026,
+      "eval_mean_token_accuracy": 0.8273287157664139,
+      "eval_num_tokens": 8013656.0,
+      "eval_runtime": 88.6038,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3460
+    },
+    {
+      "entropy": 0.09280927330255509,
+      "epoch": 8.386240193120097,
+      "grad_norm": 0.30658936500549316,
+      "learning_rate": 2.9178529262224946e-05,
+      "loss": 0.06936421990394592,
+      "mean_token_accuracy": 0.9742105580866337,
+      "num_tokens": 8061108.0,
+      "step": 3480
+    },
+    {
+      "epoch": 8.386240193120097,
+      "eval_entropy": 0.23672426734747512,
+      "eval_loss": 1.0162041187286377,
+      "eval_mean_token_accuracy": 0.8278394281194451,
+      "eval_num_tokens": 8061108.0,
+      "eval_runtime": 88.6015,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 3480
+    },
+    {
+      "entropy": 0.09195122267119586,
+      "epoch": 8.43452021726011,
+      "grad_norm": 0.24512454867362976,
+      "learning_rate": 2.7508115560327256e-05,
+      "loss": 0.06996064782142639,
+      "mean_token_accuracy": 0.974104854464531,
+      "num_tokens": 8106924.0,
+      "step": 3500
+    },
+    {
+      "epoch": 8.43452021726011,
+      "eval_entropy": 0.23758245828781235,
+      "eval_loss": 1.0157747268676758,
+      "eval_mean_token_accuracy": 0.8278501579600773,
+      "eval_num_tokens": 8106924.0,
+      "eval_runtime": 88.5678,
+      "eval_samples_per_second": 16.033,
+      "eval_steps_per_second": 2.01,
+      "step": 3500
+    },
+    {
+      "entropy": 0.09518477590754629,
+      "epoch": 8.48280024140012,
+      "grad_norm": 0.2373228371143341,
+      "learning_rate": 2.5883159734688868e-05,
+      "loss": 0.06832797527313232,
+      "mean_token_accuracy": 0.9743806682527065,
+      "num_tokens": 8153070.0,
+      "step": 3520
+    },
+    {
+      "epoch": 8.48280024140012,
+      "eval_entropy": 0.2379308074210467,
+      "eval_loss": 1.0180530548095703,
+      "eval_mean_token_accuracy": 0.8282477052024241,
+      "eval_num_tokens": 8153070.0,
+      "eval_runtime": 88.4651,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 3520
+    },
+    {
+      "entropy": 0.0906269280705601,
+      "epoch": 8.531080265540133,
+      "grad_norm": 0.37602657079696655,
+      "learning_rate": 2.4304121628587173e-05,
+      "loss": 0.06921588778495788,
+      "mean_token_accuracy": 0.9745571479201317,
+      "num_tokens": 8200066.0,
+      "step": 3540
+    },
+    {
+      "epoch": 8.531080265540133,
+      "eval_entropy": 0.2357490769263064,
+      "eval_loss": 1.017139196395874,
+      "eval_mean_token_accuracy": 0.828713178299786,
+      "eval_num_tokens": 8200066.0,
+      "eval_runtime": 88.6424,
+      "eval_samples_per_second": 16.019,
+      "eval_steps_per_second": 2.008,
+      "step": 3540
+    },
+    {
+      "entropy": 0.09273334196768701,
+      "epoch": 8.579360289680144,
+      "grad_norm": 0.2978838086128235,
+      "learning_rate": 2.277144809112794e-05,
+      "loss": 0.06970887780189514,
+      "mean_token_accuracy": 0.9751374177634716,
+      "num_tokens": 8246157.0,
+      "step": 3560
+    },
+    {
+      "epoch": 8.579360289680144,
+      "eval_entropy": 0.23590933138065123,
+      "eval_loss": 1.019231915473938,
+      "eval_mean_token_accuracy": 0.8279163412163767,
+      "eval_num_tokens": 8246157.0,
+      "eval_runtime": 88.2691,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 3560
+    },
+    {
+      "entropy": 0.09854201921261847,
+      "epoch": 8.627640313820157,
+      "grad_norm": 0.5381243228912354,
+      "learning_rate": 2.1285572850792413e-05,
+      "loss": 0.07123501300811767,
+      "mean_token_accuracy": 0.9723988175392151,
+      "num_tokens": 8291752.0,
+      "step": 3580
+    },
+    {
+      "epoch": 8.627640313820157,
+      "eval_entropy": 0.23592024949494372,
+      "eval_loss": 1.020250678062439,
+      "eval_mean_token_accuracy": 0.8282466226079491,
+      "eval_num_tokens": 8291752.0,
+      "eval_runtime": 87.8939,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 3580
+    },
+    {
+      "entropy": 0.09092000788077712,
+      "epoch": 8.67592033796017,
+      "grad_norm": 0.29089561104774475,
+      "learning_rate": 1.9846916392697493e-05,
+      "loss": 0.06982720494270325,
+      "mean_token_accuracy": 0.9749995484948158,
+      "num_tokens": 8337651.0,
+      "step": 3600
+    },
+    {
+      "epoch": 8.67592033796017,
+      "eval_entropy": 0.23658913382318583,
+      "eval_loss": 1.019862413406372,
+      "eval_mean_token_accuracy": 0.8280127842104836,
+      "eval_num_tokens": 8337651.0,
+      "eval_runtime": 88.2855,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 3600
+    },
+    {
+      "entropy": 0.08708424470387399,
+      "epoch": 8.72420036210018,
+      "grad_norm": 0.29904839396476746,
+      "learning_rate": 1.8455885839603106e-05,
+      "loss": 0.06801370978355407,
+      "mean_token_accuracy": 0.9755801670253277,
+      "num_tokens": 8385330.0,
+      "step": 3620
+    },
+    {
+      "epoch": 8.72420036210018,
+      "eval_entropy": 0.23627035771862845,
+      "eval_loss": 1.0204569101333618,
+      "eval_mean_token_accuracy": 0.8279967894045155,
+      "eval_num_tokens": 8385330.0,
+      "eval_runtime": 88.1753,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 3620
+    },
+    {
+      "entropy": 0.0916255566291511,
+      "epoch": 8.772480386240193,
+      "grad_norm": 0.20571033656597137,
+      "learning_rate": 1.7112874836701502e-05,
+      "loss": 0.0692249894142151,
+      "mean_token_accuracy": 0.9745148062705994,
+      "num_tokens": 8432134.0,
+      "step": 3640
+    },
+    {
+      "epoch": 8.772480386240193,
+      "eval_entropy": 0.2340821772813797,
+      "eval_loss": 1.0212821960449219,
+      "eval_mean_token_accuracy": 0.828373622358515,
+      "eval_num_tokens": 8432134.0,
+      "eval_runtime": 87.2518,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 3640
+    },
+    {
+      "entropy": 0.0926868049427867,
+      "epoch": 8.820760410380204,
+      "grad_norm": 0.2618265450000763,
+      "learning_rate": 1.5818263440220203e-05,
+      "loss": 0.07338647246360779,
+      "mean_token_accuracy": 0.9724708944559097,
+      "num_tokens": 8476758.0,
+      "step": 3660
+    },
+    {
+      "epoch": 8.820760410380204,
+      "eval_entropy": 0.23637074575330433,
+      "eval_loss": 1.0186725854873657,
+      "eval_mean_token_accuracy": 0.8277707605549459,
+      "eval_num_tokens": 8476758.0,
+      "eval_runtime": 88.4695,
+      "eval_samples_per_second": 16.051,
+      "eval_steps_per_second": 2.012,
+      "step": 3660
+    },
+    {
+      "entropy": 0.09175782897509635,
+      "epoch": 8.869040434520217,
+      "grad_norm": 0.2889852821826935,
+      "learning_rate": 1.4572418009870604e-05,
+      "loss": 0.07066041231155396,
+      "mean_token_accuracy": 0.9725673109292984,
+      "num_tokens": 8523294.0,
+      "step": 3680
+    },
+    {
+      "epoch": 8.869040434520217,
+      "eval_entropy": 0.23945164404223473,
+      "eval_loss": 1.0145231485366821,
+      "eval_mean_token_accuracy": 0.8281202523895864,
+      "eval_num_tokens": 8523294.0,
+      "eval_runtime": 88.368,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 3680
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.6046714733746176e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3700/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3700/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3700/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3700/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3700/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3700/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3700/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3700/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3700/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3700/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3700/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3700/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3700/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3700/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..22407ada44ef458b856b92d794cf64c461954ab8
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3700/trainer_state.json
@@ -0,0 +1,3919 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 8.91732045866023,
+  "eval_steps": 20,
+  "global_step": 3700,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    },
+    {
+      "entropy": 0.1326121488586068,
+      "epoch": 7.181050090525045,
+      "grad_norm": 0.4843117296695709,
+      "learning_rate": 8.411870386507724e-05,
+      "loss": 0.09537842869758606,
+      "mean_token_accuracy": 0.9645151920616627,
+      "num_tokens": 6898444.0,
+      "step": 2980
+    },
+    {
+      "epoch": 7.181050090525045,
+      "eval_entropy": 0.27999477783280813,
+      "eval_loss": 0.9031037092208862,
+      "eval_mean_token_accuracy": 0.8261882204018282,
+      "eval_num_tokens": 6898444.0,
+      "eval_runtime": 88.5602,
+      "eval_samples_per_second": 16.034,
+      "eval_steps_per_second": 2.01,
+      "step": 2980
+    },
+    {
+      "entropy": 0.1156751093454659,
+      "epoch": 7.229330114665057,
+      "grad_norm": 0.5297922492027283,
+      "learning_rate": 8.149629351430326e-05,
+      "loss": 0.09203402996063233,
+      "mean_token_accuracy": 0.9689733698964119,
+      "num_tokens": 6947106.0,
+      "step": 3000
+    },
+    {
+      "epoch": 7.229330114665057,
+      "eval_entropy": 0.2740225845508361,
+      "eval_loss": 0.8876440525054932,
+      "eval_mean_token_accuracy": 0.8265116857678703,
+      "eval_num_tokens": 6947106.0,
+      "eval_runtime": 87.7971,
+      "eval_samples_per_second": 16.174,
+      "eval_steps_per_second": 2.027,
+      "step": 3000
+    },
+    {
+      "entropy": 0.12116114404052496,
+      "epoch": 7.27761013880507,
+      "grad_norm": 0.5467398166656494,
+      "learning_rate": 7.890406302404618e-05,
+      "loss": 0.0933056652545929,
+      "mean_token_accuracy": 0.9676455594599247,
+      "num_tokens": 6992784.0,
+      "step": 3020
+    },
+    {
+      "epoch": 7.27761013880507,
+      "eval_entropy": 0.27238342793804876,
+      "eval_loss": 0.9166117310523987,
+      "eval_mean_token_accuracy": 0.825499015242866,
+      "eval_num_tokens": 6992784.0,
+      "eval_runtime": 88.3276,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 3020
+    },
+    {
+      "entropy": 0.11715119928121567,
+      "epoch": 7.325890162945082,
+      "grad_norm": 0.4976702332496643,
+      "learning_rate": 7.634274596487176e-05,
+      "loss": 0.09449659585952759,
+      "mean_token_accuracy": 0.966506478190422,
+      "num_tokens": 7040482.0,
+      "step": 3040
+    },
+    {
+      "epoch": 7.325890162945082,
+      "eval_entropy": 0.2788975556077582,
+      "eval_loss": 0.8833065629005432,
+      "eval_mean_token_accuracy": 0.8269920596915684,
+      "eval_num_tokens": 7040482.0,
+      "eval_runtime": 88.9016,
+      "eval_samples_per_second": 15.973,
+      "eval_steps_per_second": 2.002,
+      "step": 3040
+    },
+    {
+      "entropy": 0.12565911412239075,
+      "epoch": 7.3741701870850935,
+      "grad_norm": 0.5728679895401001,
+      "learning_rate": 7.381306715921025e-05,
+      "loss": 0.09626244902610778,
+      "mean_token_accuracy": 0.9661444149911403,
+      "num_tokens": 7086261.0,
+      "step": 3060
+    },
+    {
+      "epoch": 7.3741701870850935,
+      "eval_entropy": 0.2762492011939542,
+      "eval_loss": 0.8961160778999329,
+      "eval_mean_token_accuracy": 0.826369910092836,
+      "eval_num_tokens": 7086261.0,
+      "eval_runtime": 88.5204,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 3060
+    },
+    {
+      "entropy": 0.12258567921817302,
+      "epoch": 7.422450211225105,
+      "grad_norm": 0.468985915184021,
+      "learning_rate": 7.131574247624031e-05,
+      "loss": 0.09120266437530518,
+      "mean_token_accuracy": 0.9674938447773457,
+      "num_tokens": 7133247.0,
+      "step": 3080
+    },
+    {
+      "epoch": 7.422450211225105,
+      "eval_entropy": 0.2688512086366,
+      "eval_loss": 0.9048654437065125,
+      "eval_mean_token_accuracy": 0.8265582302982888,
+      "eval_num_tokens": 7133247.0,
+      "eval_runtime": 88.164,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 3080
+    },
+    {
+      "entropy": 0.12070536343380808,
+      "epoch": 7.470730235365117,
+      "grad_norm": 0.5310841798782349,
+      "learning_rate": 6.885147862930688e-05,
+      "loss": 0.09386247992515565,
+      "mean_token_accuracy": 0.96664222702384,
+      "num_tokens": 7177810.0,
+      "step": 3100
+    },
+    {
+      "epoch": 7.470730235365117,
+      "eval_entropy": 0.26289632874593305,
+      "eval_loss": 0.902837336063385,
+      "eval_mean_token_accuracy": 0.828052184554968,
+      "eval_num_tokens": 7177810.0,
+      "eval_runtime": 88.1201,
+      "eval_samples_per_second": 16.114,
+      "eval_steps_per_second": 2.02,
+      "step": 3100
+    },
+    {
+      "entropy": 0.11322665181942285,
+      "epoch": 7.51901025950513,
+      "grad_norm": 0.6846067905426025,
+      "learning_rate": 6.642097297592908e-05,
+      "loss": 0.08772538304328918,
+      "mean_token_accuracy": 0.9691362135112286,
+      "num_tokens": 7225302.0,
+      "step": 3120
+    },
+    {
+      "epoch": 7.51901025950513,
+      "eval_entropy": 0.2670605613776807,
+      "eval_loss": 0.9130566120147705,
+      "eval_mean_token_accuracy": 0.826366903741708,
+      "eval_num_tokens": 7225302.0,
+      "eval_runtime": 88.6035,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3120
+    },
+    {
+      "entropy": 0.10985926580615342,
+      "epoch": 7.567290283645142,
+      "grad_norm": 0.4799748361110687,
+      "learning_rate": 6.402491332045683e-05,
+      "loss": 0.08672296404838561,
+      "mean_token_accuracy": 0.9698824048042297,
+      "num_tokens": 7274881.0,
+      "step": 3140
+    },
+    {
+      "epoch": 7.567290283645142,
+      "eval_entropy": 0.26657010866015146,
+      "eval_loss": 0.9056380987167358,
+      "eval_mean_token_accuracy": 0.8272326461384806,
+      "eval_num_tokens": 7274881.0,
+      "eval_runtime": 88.2967,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3140
+    },
+    {
+      "entropy": 0.11890473761595785,
+      "epoch": 7.615570307785154,
+      "grad_norm": 0.4081912934780121,
+      "learning_rate": 6.166397771942987e-05,
+      "loss": 0.09396827220916748,
+      "mean_token_accuracy": 0.967512384802103,
+      "num_tokens": 7321163.0,
+      "step": 3160
+    },
+    {
+      "epoch": 7.615570307785154,
+      "eval_entropy": 0.26364776575833226,
+      "eval_loss": 0.9204733967781067,
+      "eval_mean_token_accuracy": 0.827419729714983,
+      "eval_num_tokens": 7321163.0,
+      "eval_runtime": 88.3441,
+      "eval_samples_per_second": 16.074,
+      "eval_steps_per_second": 2.015,
+      "step": 3160
+    },
+    {
+      "entropy": 0.12209134344011545,
+      "epoch": 7.663850331925166,
+      "grad_norm": 0.5159774422645569,
+      "learning_rate": 5.93388342896962e-05,
+      "loss": 0.09292978644371033,
+      "mean_token_accuracy": 0.9658421002328396,
+      "num_tokens": 7366217.0,
+      "step": 3180
+    },
+    {
+      "epoch": 7.663850331925166,
+      "eval_entropy": 0.2714110068725736,
+      "eval_loss": 0.9143092036247253,
+      "eval_mean_token_accuracy": 0.8262251099843657,
+      "eval_num_tokens": 7366217.0,
+      "eval_runtime": 87.9635,
+      "eval_samples_per_second": 16.143,
+      "eval_steps_per_second": 2.024,
+      "step": 3180
+    },
+    {
+      "entropy": 0.11955609126016498,
+      "epoch": 7.712130356065178,
+      "grad_norm": 0.7005976438522339,
+      "learning_rate": 5.705014101934252e-05,
+      "loss": 0.09319776892662049,
+      "mean_token_accuracy": 0.9674513377249241,
+      "num_tokens": 7410887.0,
+      "step": 3200
+    },
+    {
+      "epoch": 7.712130356065178,
+      "eval_entropy": 0.2626291513442993,
+      "eval_loss": 0.91884446144104,
+      "eval_mean_token_accuracy": 0.827563955207889,
+      "eval_num_tokens": 7410887.0,
+      "eval_runtime": 87.8849,
+      "eval_samples_per_second": 16.157,
+      "eval_steps_per_second": 2.025,
+      "step": 3200
+    },
+    {
+      "entropy": 0.11767532899975777,
+      "epoch": 7.76041038020519,
+      "grad_norm": 0.4675362706184387,
+      "learning_rate": 5.479854558149187e-05,
+      "loss": 0.09038087725639343,
+      "mean_token_accuracy": 0.9669680051505566,
+      "num_tokens": 7457578.0,
+      "step": 3220
+    },
+    {
+      "epoch": 7.76041038020519,
+      "eval_entropy": 0.26581037203582486,
+      "eval_loss": 0.9146883487701416,
+      "eval_mean_token_accuracy": 0.8270345812433222,
+      "eval_num_tokens": 7457578.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 3220
+    },
+    {
+      "entropy": 0.11383061073720455,
+      "epoch": 7.808690404345202,
+      "grad_norm": 0.541247546672821,
+      "learning_rate": 5.258468515101918e-05,
+      "loss": 0.08942405581474304,
+      "mean_token_accuracy": 0.968398705124855,
+      "num_tokens": 7505761.0,
+      "step": 3240
+    },
+    {
+      "epoch": 7.808690404345202,
+      "eval_entropy": 0.26687998960861997,
+      "eval_loss": 0.9037085175514221,
+      "eval_mean_token_accuracy": 0.8274977863504646,
+      "eval_num_tokens": 7505761.0,
+      "eval_runtime": 88.5163,
+      "eval_samples_per_second": 16.042,
+      "eval_steps_per_second": 2.011,
+      "step": 3240
+    },
+    {
+      "entropy": 0.12004058873280883,
+      "epoch": 7.856970428485214,
+      "grad_norm": 0.5289552211761475,
+      "learning_rate": 5.040918622423899e-05,
+      "loss": 0.09064805507659912,
+      "mean_token_accuracy": 0.9673330299556255,
+      "num_tokens": 7552557.0,
+      "step": 3260
+    },
+    {
+      "epoch": 7.856970428485214,
+      "eval_entropy": 0.26288379251622085,
+      "eval_loss": 0.9146178364753723,
+      "eval_mean_token_accuracy": 0.8276662448149049,
+      "eval_num_tokens": 7552557.0,
+      "eval_runtime": 88.0723,
+      "eval_samples_per_second": 16.123,
+      "eval_steps_per_second": 2.021,
+      "step": 3260
+    },
+    {
+      "entropy": 0.1180689456872642,
+      "epoch": 7.905250452625227,
+      "grad_norm": 0.4754861891269684,
+      "learning_rate": 4.827266444161377e-05,
+      "loss": 0.09402655363082886,
+      "mean_token_accuracy": 0.9663361787796021,
+      "num_tokens": 7596998.0,
+      "step": 3280
+    },
+    {
+      "epoch": 7.905250452625227,
+      "eval_entropy": 0.261690996922134,
+      "eval_loss": 0.9120186567306519,
+      "eval_mean_token_accuracy": 0.8278995015647974,
+      "eval_num_tokens": 7596998.0,
+      "eval_runtime": 88.6573,
+      "eval_samples_per_second": 16.017,
+      "eval_steps_per_second": 2.008,
+      "step": 3280
+    },
+    {
+      "entropy": 0.1162164019420743,
+      "epoch": 7.953530476765239,
+      "grad_norm": 0.6203398108482361,
+      "learning_rate": 4.61757244135355e-05,
+      "loss": 0.08832768201828003,
+      "mean_token_accuracy": 0.9694220684468746,
+      "num_tokens": 7643288.0,
+      "step": 3300
+    },
+    {
+      "epoch": 7.953530476765239,
+      "eval_entropy": 0.26158678129817664,
+      "eval_loss": 0.9301361441612244,
+      "eval_mean_token_accuracy": 0.8269603885291668,
+      "eval_num_tokens": 7643288.0,
+      "eval_runtime": 88.7064,
+      "eval_samples_per_second": 16.008,
+      "eval_steps_per_second": 2.007,
+      "step": 3300
+    },
+    {
+      "entropy": 0.10877202204489089,
+      "epoch": 8.0,
+      "grad_norm": 3.058673143386841,
+      "learning_rate": 4.411895954922763e-05,
+      "loss": 0.09159615635871887,
+      "mean_token_accuracy": 0.9686878834451947,
+      "num_tokens": 7688608.0,
+      "step": 3320
+    },
+    {
+      "epoch": 8.0,
+      "eval_entropy": 0.2590438057197614,
+      "eval_loss": 0.9262065887451172,
+      "eval_mean_token_accuracy": 0.8277404546067956,
+      "eval_num_tokens": 7688608.0,
+      "eval_runtime": 87.7557,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 3320
+    },
+    {
+      "entropy": 0.09969092034734786,
+      "epoch": 8.048280024140013,
+      "grad_norm": 0.2561911642551422,
+      "learning_rate": 4.2102951888817644e-05,
+      "loss": 0.07228735685348511,
+      "mean_token_accuracy": 0.9743556551635265,
+      "num_tokens": 7732856.0,
+      "step": 3340
+    },
+    {
+      "epoch": 8.048280024140013,
+      "eval_entropy": 0.2462534454933713,
+      "eval_loss": 0.9755983352661133,
+      "eval_mean_token_accuracy": 0.827510209565752,
+      "eval_num_tokens": 7732856.0,
+      "eval_runtime": 88.0297,
+      "eval_samples_per_second": 16.131,
+      "eval_steps_per_second": 2.022,
+      "step": 3340
+    },
+    {
+      "entropy": 0.09477624730207027,
+      "epoch": 8.096560048280024,
+      "grad_norm": 0.3670356571674347,
+      "learning_rate": 4.012827193862633e-05,
+      "loss": 0.0686036467552185,
+      "mean_token_accuracy": 0.9743670664727688,
+      "num_tokens": 7778585.0,
+      "step": 3360
+    },
+    {
+      "epoch": 8.096560048280024,
+      "eval_entropy": 0.2440882951356052,
+      "eval_loss": 1.0118423700332642,
+      "eval_mean_token_accuracy": 0.8271241070849172,
+      "eval_num_tokens": 7778585.0,
+      "eval_runtime": 88.1364,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 3360
+    },
+    {
+      "entropy": 0.09417980071157217,
+      "epoch": 8.144840072420036,
+      "grad_norm": 0.5785111784934998,
+      "learning_rate": 3.819547850972133e-05,
+      "loss": 0.07100698947906495,
+      "mean_token_accuracy": 0.9737804099917412,
+      "num_tokens": 7824287.0,
+      "step": 3380
+    },
+    {
+      "epoch": 8.144840072420036,
+      "eval_entropy": 0.24603172860453637,
+      "eval_loss": 1.0021991729736328,
+      "eval_mean_token_accuracy": 0.827162122793412,
+      "eval_num_tokens": 7824287.0,
+      "eval_runtime": 88.4534,
+      "eval_samples_per_second": 16.054,
+      "eval_steps_per_second": 2.012,
+      "step": 3380
+    },
+    {
+      "entropy": 0.087222247896716,
+      "epoch": 8.193120096560047,
+      "grad_norm": 0.44848504662513733,
+      "learning_rate": 3.6305118559780393e-05,
+      "loss": 0.06325234174728393,
+      "mean_token_accuracy": 0.9769335232675076,
+      "num_tokens": 7874784.0,
+      "step": 3400
+    },
+    {
+      "epoch": 8.193120096560047,
+      "eval_entropy": 0.24544863905129807,
+      "eval_loss": 1.005008339881897,
+      "eval_mean_token_accuracy": 0.8273035240976998,
+      "eval_num_tokens": 7874784.0,
+      "eval_runtime": 88.6737,
+      "eval_samples_per_second": 16.014,
+      "eval_steps_per_second": 2.007,
+      "step": 3400
+    },
+    {
+      "entropy": 0.09037275062873959,
+      "epoch": 8.24140012070006,
+      "grad_norm": 0.23529155552387238,
+      "learning_rate": 3.4457727038308435e-05,
+      "loss": 0.06589569449424744,
+      "mean_token_accuracy": 0.9751470290124417,
+      "num_tokens": 7925240.0,
+      "step": 3420
+    },
+    {
+      "epoch": 8.24140012070006,
+      "eval_entropy": 0.24203423494368456,
+      "eval_loss": 1.0052824020385742,
+      "eval_mean_token_accuracy": 0.8275575503874361,
+      "eval_num_tokens": 7925240.0,
+      "eval_runtime": 88.3486,
+      "eval_samples_per_second": 16.073,
+      "eval_steps_per_second": 2.015,
+      "step": 3420
+    },
+    {
+      "entropy": 0.10373208015225828,
+      "epoch": 8.289680144840073,
+      "grad_norm": 0.24796408414840698,
+      "learning_rate": 3.2653826735253594e-05,
+      "loss": 0.07714771628379821,
+      "mean_token_accuracy": 0.9722891598939896,
+      "num_tokens": 7966685.0,
+      "step": 3440
+    },
+    {
+      "epoch": 8.289680144840073,
+      "eval_entropy": 0.23461157050025597,
+      "eval_loss": 1.012108325958252,
+      "eval_mean_token_accuracy": 0.82825900564033,
+      "eval_num_tokens": 7966685.0,
+      "eval_runtime": 88.1018,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 3440
+    },
+    {
+      "entropy": 0.08967954274266958,
+      "epoch": 8.337960168980084,
+      "grad_norm": 0.5314674973487854,
+      "learning_rate": 3.08939281330635e-05,
+      "loss": 0.06746266484260559,
+      "mean_token_accuracy": 0.9753133699297905,
+      "num_tokens": 8013656.0,
+      "step": 3460
+    },
+    {
+      "epoch": 8.337960168980084,
+      "eval_entropy": 0.23781998589467468,
+      "eval_loss": 1.0119913816452026,
+      "eval_mean_token_accuracy": 0.8273287157664139,
+      "eval_num_tokens": 8013656.0,
+      "eval_runtime": 88.6038,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3460
+    },
+    {
+      "entropy": 0.09280927330255509,
+      "epoch": 8.386240193120097,
+      "grad_norm": 0.30658936500549316,
+      "learning_rate": 2.9178529262224946e-05,
+      "loss": 0.06936421990394592,
+      "mean_token_accuracy": 0.9742105580866337,
+      "num_tokens": 8061108.0,
+      "step": 3480
+    },
+    {
+      "epoch": 8.386240193120097,
+      "eval_entropy": 0.23672426734747512,
+      "eval_loss": 1.0162041187286377,
+      "eval_mean_token_accuracy": 0.8278394281194451,
+      "eval_num_tokens": 8061108.0,
+      "eval_runtime": 88.6015,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 3480
+    },
+    {
+      "entropy": 0.09195122267119586,
+      "epoch": 8.43452021726011,
+      "grad_norm": 0.24512454867362976,
+      "learning_rate": 2.7508115560327256e-05,
+      "loss": 0.06996064782142639,
+      "mean_token_accuracy": 0.974104854464531,
+      "num_tokens": 8106924.0,
+      "step": 3500
+    },
+    {
+      "epoch": 8.43452021726011,
+      "eval_entropy": 0.23758245828781235,
+      "eval_loss": 1.0157747268676758,
+      "eval_mean_token_accuracy": 0.8278501579600773,
+      "eval_num_tokens": 8106924.0,
+      "eval_runtime": 88.5678,
+      "eval_samples_per_second": 16.033,
+      "eval_steps_per_second": 2.01,
+      "step": 3500
+    },
+    {
+      "entropy": 0.09518477590754629,
+      "epoch": 8.48280024140012,
+      "grad_norm": 0.2373228371143341,
+      "learning_rate": 2.5883159734688868e-05,
+      "loss": 0.06832797527313232,
+      "mean_token_accuracy": 0.9743806682527065,
+      "num_tokens": 8153070.0,
+      "step": 3520
+    },
+    {
+      "epoch": 8.48280024140012,
+      "eval_entropy": 0.2379308074210467,
+      "eval_loss": 1.0180530548095703,
+      "eval_mean_token_accuracy": 0.8282477052024241,
+      "eval_num_tokens": 8153070.0,
+      "eval_runtime": 88.4651,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 3520
+    },
+    {
+      "entropy": 0.0906269280705601,
+      "epoch": 8.531080265540133,
+      "grad_norm": 0.37602657079696655,
+      "learning_rate": 2.4304121628587173e-05,
+      "loss": 0.06921588778495788,
+      "mean_token_accuracy": 0.9745571479201317,
+      "num_tokens": 8200066.0,
+      "step": 3540
+    },
+    {
+      "epoch": 8.531080265540133,
+      "eval_entropy": 0.2357490769263064,
+      "eval_loss": 1.017139196395874,
+      "eval_mean_token_accuracy": 0.828713178299786,
+      "eval_num_tokens": 8200066.0,
+      "eval_runtime": 88.6424,
+      "eval_samples_per_second": 16.019,
+      "eval_steps_per_second": 2.008,
+      "step": 3540
+    },
+    {
+      "entropy": 0.09273334196768701,
+      "epoch": 8.579360289680144,
+      "grad_norm": 0.2978838086128235,
+      "learning_rate": 2.277144809112794e-05,
+      "loss": 0.06970887780189514,
+      "mean_token_accuracy": 0.9751374177634716,
+      "num_tokens": 8246157.0,
+      "step": 3560
+    },
+    {
+      "epoch": 8.579360289680144,
+      "eval_entropy": 0.23590933138065123,
+      "eval_loss": 1.019231915473938,
+      "eval_mean_token_accuracy": 0.8279163412163767,
+      "eval_num_tokens": 8246157.0,
+      "eval_runtime": 88.2691,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 3560
+    },
+    {
+      "entropy": 0.09854201921261847,
+      "epoch": 8.627640313820157,
+      "grad_norm": 0.5381243228912354,
+      "learning_rate": 2.1285572850792413e-05,
+      "loss": 0.07123501300811767,
+      "mean_token_accuracy": 0.9723988175392151,
+      "num_tokens": 8291752.0,
+      "step": 3580
+    },
+    {
+      "epoch": 8.627640313820157,
+      "eval_entropy": 0.23592024949494372,
+      "eval_loss": 1.020250678062439,
+      "eval_mean_token_accuracy": 0.8282466226079491,
+      "eval_num_tokens": 8291752.0,
+      "eval_runtime": 87.8939,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 3580
+    },
+    {
+      "entropy": 0.09092000788077712,
+      "epoch": 8.67592033796017,
+      "grad_norm": 0.29089561104774475,
+      "learning_rate": 1.9846916392697493e-05,
+      "loss": 0.06982720494270325,
+      "mean_token_accuracy": 0.9749995484948158,
+      "num_tokens": 8337651.0,
+      "step": 3600
+    },
+    {
+      "epoch": 8.67592033796017,
+      "eval_entropy": 0.23658913382318583,
+      "eval_loss": 1.019862413406372,
+      "eval_mean_token_accuracy": 0.8280127842104836,
+      "eval_num_tokens": 8337651.0,
+      "eval_runtime": 88.2855,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 3600
+    },
+    {
+      "entropy": 0.08708424470387399,
+      "epoch": 8.72420036210018,
+      "grad_norm": 0.29904839396476746,
+      "learning_rate": 1.8455885839603106e-05,
+      "loss": 0.06801370978355407,
+      "mean_token_accuracy": 0.9755801670253277,
+      "num_tokens": 8385330.0,
+      "step": 3620
+    },
+    {
+      "epoch": 8.72420036210018,
+      "eval_entropy": 0.23627035771862845,
+      "eval_loss": 1.0204569101333618,
+      "eval_mean_token_accuracy": 0.8279967894045155,
+      "eval_num_tokens": 8385330.0,
+      "eval_runtime": 88.1753,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 3620
+    },
+    {
+      "entropy": 0.0916255566291511,
+      "epoch": 8.772480386240193,
+      "grad_norm": 0.20571033656597137,
+      "learning_rate": 1.7112874836701502e-05,
+      "loss": 0.0692249894142151,
+      "mean_token_accuracy": 0.9745148062705994,
+      "num_tokens": 8432134.0,
+      "step": 3640
+    },
+    {
+      "epoch": 8.772480386240193,
+      "eval_entropy": 0.2340821772813797,
+      "eval_loss": 1.0212821960449219,
+      "eval_mean_token_accuracy": 0.828373622358515,
+      "eval_num_tokens": 8432134.0,
+      "eval_runtime": 87.2518,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 3640
+    },
+    {
+      "entropy": 0.0926868049427867,
+      "epoch": 8.820760410380204,
+      "grad_norm": 0.2618265450000763,
+      "learning_rate": 1.5818263440220203e-05,
+      "loss": 0.07338647246360779,
+      "mean_token_accuracy": 0.9724708944559097,
+      "num_tokens": 8476758.0,
+      "step": 3660
+    },
+    {
+      "epoch": 8.820760410380204,
+      "eval_entropy": 0.23637074575330433,
+      "eval_loss": 1.0186725854873657,
+      "eval_mean_token_accuracy": 0.8277707605549459,
+      "eval_num_tokens": 8476758.0,
+      "eval_runtime": 88.4695,
+      "eval_samples_per_second": 16.051,
+      "eval_steps_per_second": 2.012,
+      "step": 3660
+    },
+    {
+      "entropy": 0.09175782897509635,
+      "epoch": 8.869040434520217,
+      "grad_norm": 0.2889852821826935,
+      "learning_rate": 1.4572418009870604e-05,
+      "loss": 0.07066041231155396,
+      "mean_token_accuracy": 0.9725673109292984,
+      "num_tokens": 8523294.0,
+      "step": 3680
+    },
+    {
+      "epoch": 8.869040434520217,
+      "eval_entropy": 0.23945164404223473,
+      "eval_loss": 1.0145231485366821,
+      "eval_mean_token_accuracy": 0.8281202523895864,
+      "eval_num_tokens": 8523294.0,
+      "eval_runtime": 88.368,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 3680
+    },
+    {
+      "entropy": 0.09265611860901117,
+      "epoch": 8.91732045866023,
+      "grad_norm": 0.41092950105667114,
+      "learning_rate": 1.3375691105172625e-05,
+      "loss": 0.06975027322769164,
+      "mean_token_accuracy": 0.974636297672987,
+      "num_tokens": 8570358.0,
+      "step": 3700
+    },
+    {
+      "epoch": 8.91732045866023,
+      "eval_entropy": 0.239588919901446,
+      "eval_loss": 1.014059066772461,
+      "eval_mean_token_accuracy": 0.828028041660116,
+      "eval_num_tokens": 8570358.0,
+      "eval_runtime": 88.1039,
+      "eval_samples_per_second": 16.117,
+      "eval_steps_per_second": 2.02,
+      "step": 3700
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.6235566437227315e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3720/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3720/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3720/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3720/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3720/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3720/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3720/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3720/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3720/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3720/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3720/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3720/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3720/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3720/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..09cb9abcc13e89b01c2aa4b9a6e99cf53549a24a
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3720/trainer_state.json
@@ -0,0 +1,3940 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 8.965600482800241,
+  "eval_steps": 20,
+  "global_step": 3720,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    },
+    {
+      "entropy": 0.1326121488586068,
+      "epoch": 7.181050090525045,
+      "grad_norm": 0.4843117296695709,
+      "learning_rate": 8.411870386507724e-05,
+      "loss": 0.09537842869758606,
+      "mean_token_accuracy": 0.9645151920616627,
+      "num_tokens": 6898444.0,
+      "step": 2980
+    },
+    {
+      "epoch": 7.181050090525045,
+      "eval_entropy": 0.27999477783280813,
+      "eval_loss": 0.9031037092208862,
+      "eval_mean_token_accuracy": 0.8261882204018282,
+      "eval_num_tokens": 6898444.0,
+      "eval_runtime": 88.5602,
+      "eval_samples_per_second": 16.034,
+      "eval_steps_per_second": 2.01,
+      "step": 2980
+    },
+    {
+      "entropy": 0.1156751093454659,
+      "epoch": 7.229330114665057,
+      "grad_norm": 0.5297922492027283,
+      "learning_rate": 8.149629351430326e-05,
+      "loss": 0.09203402996063233,
+      "mean_token_accuracy": 0.9689733698964119,
+      "num_tokens": 6947106.0,
+      "step": 3000
+    },
+    {
+      "epoch": 7.229330114665057,
+      "eval_entropy": 0.2740225845508361,
+      "eval_loss": 0.8876440525054932,
+      "eval_mean_token_accuracy": 0.8265116857678703,
+      "eval_num_tokens": 6947106.0,
+      "eval_runtime": 87.7971,
+      "eval_samples_per_second": 16.174,
+      "eval_steps_per_second": 2.027,
+      "step": 3000
+    },
+    {
+      "entropy": 0.12116114404052496,
+      "epoch": 7.27761013880507,
+      "grad_norm": 0.5467398166656494,
+      "learning_rate": 7.890406302404618e-05,
+      "loss": 0.0933056652545929,
+      "mean_token_accuracy": 0.9676455594599247,
+      "num_tokens": 6992784.0,
+      "step": 3020
+    },
+    {
+      "epoch": 7.27761013880507,
+      "eval_entropy": 0.27238342793804876,
+      "eval_loss": 0.9166117310523987,
+      "eval_mean_token_accuracy": 0.825499015242866,
+      "eval_num_tokens": 6992784.0,
+      "eval_runtime": 88.3276,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 3020
+    },
+    {
+      "entropy": 0.11715119928121567,
+      "epoch": 7.325890162945082,
+      "grad_norm": 0.4976702332496643,
+      "learning_rate": 7.634274596487176e-05,
+      "loss": 0.09449659585952759,
+      "mean_token_accuracy": 0.966506478190422,
+      "num_tokens": 7040482.0,
+      "step": 3040
+    },
+    {
+      "epoch": 7.325890162945082,
+      "eval_entropy": 0.2788975556077582,
+      "eval_loss": 0.8833065629005432,
+      "eval_mean_token_accuracy": 0.8269920596915684,
+      "eval_num_tokens": 7040482.0,
+      "eval_runtime": 88.9016,
+      "eval_samples_per_second": 15.973,
+      "eval_steps_per_second": 2.002,
+      "step": 3040
+    },
+    {
+      "entropy": 0.12565911412239075,
+      "epoch": 7.3741701870850935,
+      "grad_norm": 0.5728679895401001,
+      "learning_rate": 7.381306715921025e-05,
+      "loss": 0.09626244902610778,
+      "mean_token_accuracy": 0.9661444149911403,
+      "num_tokens": 7086261.0,
+      "step": 3060
+    },
+    {
+      "epoch": 7.3741701870850935,
+      "eval_entropy": 0.2762492011939542,
+      "eval_loss": 0.8961160778999329,
+      "eval_mean_token_accuracy": 0.826369910092836,
+      "eval_num_tokens": 7086261.0,
+      "eval_runtime": 88.5204,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 3060
+    },
+    {
+      "entropy": 0.12258567921817302,
+      "epoch": 7.422450211225105,
+      "grad_norm": 0.468985915184021,
+      "learning_rate": 7.131574247624031e-05,
+      "loss": 0.09120266437530518,
+      "mean_token_accuracy": 0.9674938447773457,
+      "num_tokens": 7133247.0,
+      "step": 3080
+    },
+    {
+      "epoch": 7.422450211225105,
+      "eval_entropy": 0.2688512086366,
+      "eval_loss": 0.9048654437065125,
+      "eval_mean_token_accuracy": 0.8265582302982888,
+      "eval_num_tokens": 7133247.0,
+      "eval_runtime": 88.164,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 3080
+    },
+    {
+      "entropy": 0.12070536343380808,
+      "epoch": 7.470730235365117,
+      "grad_norm": 0.5310841798782349,
+      "learning_rate": 6.885147862930688e-05,
+      "loss": 0.09386247992515565,
+      "mean_token_accuracy": 0.96664222702384,
+      "num_tokens": 7177810.0,
+      "step": 3100
+    },
+    {
+      "epoch": 7.470730235365117,
+      "eval_entropy": 0.26289632874593305,
+      "eval_loss": 0.902837336063385,
+      "eval_mean_token_accuracy": 0.828052184554968,
+      "eval_num_tokens": 7177810.0,
+      "eval_runtime": 88.1201,
+      "eval_samples_per_second": 16.114,
+      "eval_steps_per_second": 2.02,
+      "step": 3100
+    },
+    {
+      "entropy": 0.11322665181942285,
+      "epoch": 7.51901025950513,
+      "grad_norm": 0.6846067905426025,
+      "learning_rate": 6.642097297592908e-05,
+      "loss": 0.08772538304328918,
+      "mean_token_accuracy": 0.9691362135112286,
+      "num_tokens": 7225302.0,
+      "step": 3120
+    },
+    {
+      "epoch": 7.51901025950513,
+      "eval_entropy": 0.2670605613776807,
+      "eval_loss": 0.9130566120147705,
+      "eval_mean_token_accuracy": 0.826366903741708,
+      "eval_num_tokens": 7225302.0,
+      "eval_runtime": 88.6035,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3120
+    },
+    {
+      "entropy": 0.10985926580615342,
+      "epoch": 7.567290283645142,
+      "grad_norm": 0.4799748361110687,
+      "learning_rate": 6.402491332045683e-05,
+      "loss": 0.08672296404838561,
+      "mean_token_accuracy": 0.9698824048042297,
+      "num_tokens": 7274881.0,
+      "step": 3140
+    },
+    {
+      "epoch": 7.567290283645142,
+      "eval_entropy": 0.26657010866015146,
+      "eval_loss": 0.9056380987167358,
+      "eval_mean_token_accuracy": 0.8272326461384806,
+      "eval_num_tokens": 7274881.0,
+      "eval_runtime": 88.2967,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3140
+    },
+    {
+      "entropy": 0.11890473761595785,
+      "epoch": 7.615570307785154,
+      "grad_norm": 0.4081912934780121,
+      "learning_rate": 6.166397771942987e-05,
+      "loss": 0.09396827220916748,
+      "mean_token_accuracy": 0.967512384802103,
+      "num_tokens": 7321163.0,
+      "step": 3160
+    },
+    {
+      "epoch": 7.615570307785154,
+      "eval_entropy": 0.26364776575833226,
+      "eval_loss": 0.9204733967781067,
+      "eval_mean_token_accuracy": 0.827419729714983,
+      "eval_num_tokens": 7321163.0,
+      "eval_runtime": 88.3441,
+      "eval_samples_per_second": 16.074,
+      "eval_steps_per_second": 2.015,
+      "step": 3160
+    },
+    {
+      "entropy": 0.12209134344011545,
+      "epoch": 7.663850331925166,
+      "grad_norm": 0.5159774422645569,
+      "learning_rate": 5.93388342896962e-05,
+      "loss": 0.09292978644371033,
+      "mean_token_accuracy": 0.9658421002328396,
+      "num_tokens": 7366217.0,
+      "step": 3180
+    },
+    {
+      "epoch": 7.663850331925166,
+      "eval_entropy": 0.2714110068725736,
+      "eval_loss": 0.9143092036247253,
+      "eval_mean_token_accuracy": 0.8262251099843657,
+      "eval_num_tokens": 7366217.0,
+      "eval_runtime": 87.9635,
+      "eval_samples_per_second": 16.143,
+      "eval_steps_per_second": 2.024,
+      "step": 3180
+    },
+    {
+      "entropy": 0.11955609126016498,
+      "epoch": 7.712130356065178,
+      "grad_norm": 0.7005976438522339,
+      "learning_rate": 5.705014101934252e-05,
+      "loss": 0.09319776892662049,
+      "mean_token_accuracy": 0.9674513377249241,
+      "num_tokens": 7410887.0,
+      "step": 3200
+    },
+    {
+      "epoch": 7.712130356065178,
+      "eval_entropy": 0.2626291513442993,
+      "eval_loss": 0.91884446144104,
+      "eval_mean_token_accuracy": 0.827563955207889,
+      "eval_num_tokens": 7410887.0,
+      "eval_runtime": 87.8849,
+      "eval_samples_per_second": 16.157,
+      "eval_steps_per_second": 2.025,
+      "step": 3200
+    },
+    {
+      "entropy": 0.11767532899975777,
+      "epoch": 7.76041038020519,
+      "grad_norm": 0.4675362706184387,
+      "learning_rate": 5.479854558149187e-05,
+      "loss": 0.09038087725639343,
+      "mean_token_accuracy": 0.9669680051505566,
+      "num_tokens": 7457578.0,
+      "step": 3220
+    },
+    {
+      "epoch": 7.76041038020519,
+      "eval_entropy": 0.26581037203582486,
+      "eval_loss": 0.9146883487701416,
+      "eval_mean_token_accuracy": 0.8270345812433222,
+      "eval_num_tokens": 7457578.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 3220
+    },
+    {
+      "entropy": 0.11383061073720455,
+      "epoch": 7.808690404345202,
+      "grad_norm": 0.541247546672821,
+      "learning_rate": 5.258468515101918e-05,
+      "loss": 0.08942405581474304,
+      "mean_token_accuracy": 0.968398705124855,
+      "num_tokens": 7505761.0,
+      "step": 3240
+    },
+    {
+      "epoch": 7.808690404345202,
+      "eval_entropy": 0.26687998960861997,
+      "eval_loss": 0.9037085175514221,
+      "eval_mean_token_accuracy": 0.8274977863504646,
+      "eval_num_tokens": 7505761.0,
+      "eval_runtime": 88.5163,
+      "eval_samples_per_second": 16.042,
+      "eval_steps_per_second": 2.011,
+      "step": 3240
+    },
+    {
+      "entropy": 0.12004058873280883,
+      "epoch": 7.856970428485214,
+      "grad_norm": 0.5289552211761475,
+      "learning_rate": 5.040918622423899e-05,
+      "loss": 0.09064805507659912,
+      "mean_token_accuracy": 0.9673330299556255,
+      "num_tokens": 7552557.0,
+      "step": 3260
+    },
+    {
+      "epoch": 7.856970428485214,
+      "eval_entropy": 0.26288379251622085,
+      "eval_loss": 0.9146178364753723,
+      "eval_mean_token_accuracy": 0.8276662448149049,
+      "eval_num_tokens": 7552557.0,
+      "eval_runtime": 88.0723,
+      "eval_samples_per_second": 16.123,
+      "eval_steps_per_second": 2.021,
+      "step": 3260
+    },
+    {
+      "entropy": 0.1180689456872642,
+      "epoch": 7.905250452625227,
+      "grad_norm": 0.4754861891269684,
+      "learning_rate": 4.827266444161377e-05,
+      "loss": 0.09402655363082886,
+      "mean_token_accuracy": 0.9663361787796021,
+      "num_tokens": 7596998.0,
+      "step": 3280
+    },
+    {
+      "epoch": 7.905250452625227,
+      "eval_entropy": 0.261690996922134,
+      "eval_loss": 0.9120186567306519,
+      "eval_mean_token_accuracy": 0.8278995015647974,
+      "eval_num_tokens": 7596998.0,
+      "eval_runtime": 88.6573,
+      "eval_samples_per_second": 16.017,
+      "eval_steps_per_second": 2.008,
+      "step": 3280
+    },
+    {
+      "entropy": 0.1162164019420743,
+      "epoch": 7.953530476765239,
+      "grad_norm": 0.6203398108482361,
+      "learning_rate": 4.61757244135355e-05,
+      "loss": 0.08832768201828003,
+      "mean_token_accuracy": 0.9694220684468746,
+      "num_tokens": 7643288.0,
+      "step": 3300
+    },
+    {
+      "epoch": 7.953530476765239,
+      "eval_entropy": 0.26158678129817664,
+      "eval_loss": 0.9301361441612244,
+      "eval_mean_token_accuracy": 0.8269603885291668,
+      "eval_num_tokens": 7643288.0,
+      "eval_runtime": 88.7064,
+      "eval_samples_per_second": 16.008,
+      "eval_steps_per_second": 2.007,
+      "step": 3300
+    },
+    {
+      "entropy": 0.10877202204489089,
+      "epoch": 8.0,
+      "grad_norm": 3.058673143386841,
+      "learning_rate": 4.411895954922763e-05,
+      "loss": 0.09159615635871887,
+      "mean_token_accuracy": 0.9686878834451947,
+      "num_tokens": 7688608.0,
+      "step": 3320
+    },
+    {
+      "epoch": 8.0,
+      "eval_entropy": 0.2590438057197614,
+      "eval_loss": 0.9262065887451172,
+      "eval_mean_token_accuracy": 0.8277404546067956,
+      "eval_num_tokens": 7688608.0,
+      "eval_runtime": 87.7557,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 3320
+    },
+    {
+      "entropy": 0.09969092034734786,
+      "epoch": 8.048280024140013,
+      "grad_norm": 0.2561911642551422,
+      "learning_rate": 4.2102951888817644e-05,
+      "loss": 0.07228735685348511,
+      "mean_token_accuracy": 0.9743556551635265,
+      "num_tokens": 7732856.0,
+      "step": 3340
+    },
+    {
+      "epoch": 8.048280024140013,
+      "eval_entropy": 0.2462534454933713,
+      "eval_loss": 0.9755983352661133,
+      "eval_mean_token_accuracy": 0.827510209565752,
+      "eval_num_tokens": 7732856.0,
+      "eval_runtime": 88.0297,
+      "eval_samples_per_second": 16.131,
+      "eval_steps_per_second": 2.022,
+      "step": 3340
+    },
+    {
+      "entropy": 0.09477624730207027,
+      "epoch": 8.096560048280024,
+      "grad_norm": 0.3670356571674347,
+      "learning_rate": 4.012827193862633e-05,
+      "loss": 0.0686036467552185,
+      "mean_token_accuracy": 0.9743670664727688,
+      "num_tokens": 7778585.0,
+      "step": 3360
+    },
+    {
+      "epoch": 8.096560048280024,
+      "eval_entropy": 0.2440882951356052,
+      "eval_loss": 1.0118423700332642,
+      "eval_mean_token_accuracy": 0.8271241070849172,
+      "eval_num_tokens": 7778585.0,
+      "eval_runtime": 88.1364,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 3360
+    },
+    {
+      "entropy": 0.09417980071157217,
+      "epoch": 8.144840072420036,
+      "grad_norm": 0.5785111784934998,
+      "learning_rate": 3.819547850972133e-05,
+      "loss": 0.07100698947906495,
+      "mean_token_accuracy": 0.9737804099917412,
+      "num_tokens": 7824287.0,
+      "step": 3380
+    },
+    {
+      "epoch": 8.144840072420036,
+      "eval_entropy": 0.24603172860453637,
+      "eval_loss": 1.0021991729736328,
+      "eval_mean_token_accuracy": 0.827162122793412,
+      "eval_num_tokens": 7824287.0,
+      "eval_runtime": 88.4534,
+      "eval_samples_per_second": 16.054,
+      "eval_steps_per_second": 2.012,
+      "step": 3380
+    },
+    {
+      "entropy": 0.087222247896716,
+      "epoch": 8.193120096560047,
+      "grad_norm": 0.44848504662513733,
+      "learning_rate": 3.6305118559780393e-05,
+      "loss": 0.06325234174728393,
+      "mean_token_accuracy": 0.9769335232675076,
+      "num_tokens": 7874784.0,
+      "step": 3400
+    },
+    {
+      "epoch": 8.193120096560047,
+      "eval_entropy": 0.24544863905129807,
+      "eval_loss": 1.005008339881897,
+      "eval_mean_token_accuracy": 0.8273035240976998,
+      "eval_num_tokens": 7874784.0,
+      "eval_runtime": 88.6737,
+      "eval_samples_per_second": 16.014,
+      "eval_steps_per_second": 2.007,
+      "step": 3400
+    },
+    {
+      "entropy": 0.09037275062873959,
+      "epoch": 8.24140012070006,
+      "grad_norm": 0.23529155552387238,
+      "learning_rate": 3.4457727038308435e-05,
+      "loss": 0.06589569449424744,
+      "mean_token_accuracy": 0.9751470290124417,
+      "num_tokens": 7925240.0,
+      "step": 3420
+    },
+    {
+      "epoch": 8.24140012070006,
+      "eval_entropy": 0.24203423494368456,
+      "eval_loss": 1.0052824020385742,
+      "eval_mean_token_accuracy": 0.8275575503874361,
+      "eval_num_tokens": 7925240.0,
+      "eval_runtime": 88.3486,
+      "eval_samples_per_second": 16.073,
+      "eval_steps_per_second": 2.015,
+      "step": 3420
+    },
+    {
+      "entropy": 0.10373208015225828,
+      "epoch": 8.289680144840073,
+      "grad_norm": 0.24796408414840698,
+      "learning_rate": 3.2653826735253594e-05,
+      "loss": 0.07714771628379821,
+      "mean_token_accuracy": 0.9722891598939896,
+      "num_tokens": 7966685.0,
+      "step": 3440
+    },
+    {
+      "epoch": 8.289680144840073,
+      "eval_entropy": 0.23461157050025597,
+      "eval_loss": 1.012108325958252,
+      "eval_mean_token_accuracy": 0.82825900564033,
+      "eval_num_tokens": 7966685.0,
+      "eval_runtime": 88.1018,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 3440
+    },
+    {
+      "entropy": 0.08967954274266958,
+      "epoch": 8.337960168980084,
+      "grad_norm": 0.5314674973487854,
+      "learning_rate": 3.08939281330635e-05,
+      "loss": 0.06746266484260559,
+      "mean_token_accuracy": 0.9753133699297905,
+      "num_tokens": 8013656.0,
+      "step": 3460
+    },
+    {
+      "epoch": 8.337960168980084,
+      "eval_entropy": 0.23781998589467468,
+      "eval_loss": 1.0119913816452026,
+      "eval_mean_token_accuracy": 0.8273287157664139,
+      "eval_num_tokens": 8013656.0,
+      "eval_runtime": 88.6038,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3460
+    },
+    {
+      "entropy": 0.09280927330255509,
+      "epoch": 8.386240193120097,
+      "grad_norm": 0.30658936500549316,
+      "learning_rate": 2.9178529262224946e-05,
+      "loss": 0.06936421990394592,
+      "mean_token_accuracy": 0.9742105580866337,
+      "num_tokens": 8061108.0,
+      "step": 3480
+    },
+    {
+      "epoch": 8.386240193120097,
+      "eval_entropy": 0.23672426734747512,
+      "eval_loss": 1.0162041187286377,
+      "eval_mean_token_accuracy": 0.8278394281194451,
+      "eval_num_tokens": 8061108.0,
+      "eval_runtime": 88.6015,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 3480
+    },
+    {
+      "entropy": 0.09195122267119586,
+      "epoch": 8.43452021726011,
+      "grad_norm": 0.24512454867362976,
+      "learning_rate": 2.7508115560327256e-05,
+      "loss": 0.06996064782142639,
+      "mean_token_accuracy": 0.974104854464531,
+      "num_tokens": 8106924.0,
+      "step": 3500
+    },
+    {
+      "epoch": 8.43452021726011,
+      "eval_entropy": 0.23758245828781235,
+      "eval_loss": 1.0157747268676758,
+      "eval_mean_token_accuracy": 0.8278501579600773,
+      "eval_num_tokens": 8106924.0,
+      "eval_runtime": 88.5678,
+      "eval_samples_per_second": 16.033,
+      "eval_steps_per_second": 2.01,
+      "step": 3500
+    },
+    {
+      "entropy": 0.09518477590754629,
+      "epoch": 8.48280024140012,
+      "grad_norm": 0.2373228371143341,
+      "learning_rate": 2.5883159734688868e-05,
+      "loss": 0.06832797527313232,
+      "mean_token_accuracy": 0.9743806682527065,
+      "num_tokens": 8153070.0,
+      "step": 3520
+    },
+    {
+      "epoch": 8.48280024140012,
+      "eval_entropy": 0.2379308074210467,
+      "eval_loss": 1.0180530548095703,
+      "eval_mean_token_accuracy": 0.8282477052024241,
+      "eval_num_tokens": 8153070.0,
+      "eval_runtime": 88.4651,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 3520
+    },
+    {
+      "entropy": 0.0906269280705601,
+      "epoch": 8.531080265540133,
+      "grad_norm": 0.37602657079696655,
+      "learning_rate": 2.4304121628587173e-05,
+      "loss": 0.06921588778495788,
+      "mean_token_accuracy": 0.9745571479201317,
+      "num_tokens": 8200066.0,
+      "step": 3540
+    },
+    {
+      "epoch": 8.531080265540133,
+      "eval_entropy": 0.2357490769263064,
+      "eval_loss": 1.017139196395874,
+      "eval_mean_token_accuracy": 0.828713178299786,
+      "eval_num_tokens": 8200066.0,
+      "eval_runtime": 88.6424,
+      "eval_samples_per_second": 16.019,
+      "eval_steps_per_second": 2.008,
+      "step": 3540
+    },
+    {
+      "entropy": 0.09273334196768701,
+      "epoch": 8.579360289680144,
+      "grad_norm": 0.2978838086128235,
+      "learning_rate": 2.277144809112794e-05,
+      "loss": 0.06970887780189514,
+      "mean_token_accuracy": 0.9751374177634716,
+      "num_tokens": 8246157.0,
+      "step": 3560
+    },
+    {
+      "epoch": 8.579360289680144,
+      "eval_entropy": 0.23590933138065123,
+      "eval_loss": 1.019231915473938,
+      "eval_mean_token_accuracy": 0.8279163412163767,
+      "eval_num_tokens": 8246157.0,
+      "eval_runtime": 88.2691,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 3560
+    },
+    {
+      "entropy": 0.09854201921261847,
+      "epoch": 8.627640313820157,
+      "grad_norm": 0.5381243228912354,
+      "learning_rate": 2.1285572850792413e-05,
+      "loss": 0.07123501300811767,
+      "mean_token_accuracy": 0.9723988175392151,
+      "num_tokens": 8291752.0,
+      "step": 3580
+    },
+    {
+      "epoch": 8.627640313820157,
+      "eval_entropy": 0.23592024949494372,
+      "eval_loss": 1.020250678062439,
+      "eval_mean_token_accuracy": 0.8282466226079491,
+      "eval_num_tokens": 8291752.0,
+      "eval_runtime": 87.8939,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 3580
+    },
+    {
+      "entropy": 0.09092000788077712,
+      "epoch": 8.67592033796017,
+      "grad_norm": 0.29089561104774475,
+      "learning_rate": 1.9846916392697493e-05,
+      "loss": 0.06982720494270325,
+      "mean_token_accuracy": 0.9749995484948158,
+      "num_tokens": 8337651.0,
+      "step": 3600
+    },
+    {
+      "epoch": 8.67592033796017,
+      "eval_entropy": 0.23658913382318583,
+      "eval_loss": 1.019862413406372,
+      "eval_mean_token_accuracy": 0.8280127842104836,
+      "eval_num_tokens": 8337651.0,
+      "eval_runtime": 88.2855,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 3600
+    },
+    {
+      "entropy": 0.08708424470387399,
+      "epoch": 8.72420036210018,
+      "grad_norm": 0.29904839396476746,
+      "learning_rate": 1.8455885839603106e-05,
+      "loss": 0.06801370978355407,
+      "mean_token_accuracy": 0.9755801670253277,
+      "num_tokens": 8385330.0,
+      "step": 3620
+    },
+    {
+      "epoch": 8.72420036210018,
+      "eval_entropy": 0.23627035771862845,
+      "eval_loss": 1.0204569101333618,
+      "eval_mean_token_accuracy": 0.8279967894045155,
+      "eval_num_tokens": 8385330.0,
+      "eval_runtime": 88.1753,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 3620
+    },
+    {
+      "entropy": 0.0916255566291511,
+      "epoch": 8.772480386240193,
+      "grad_norm": 0.20571033656597137,
+      "learning_rate": 1.7112874836701502e-05,
+      "loss": 0.0692249894142151,
+      "mean_token_accuracy": 0.9745148062705994,
+      "num_tokens": 8432134.0,
+      "step": 3640
+    },
+    {
+      "epoch": 8.772480386240193,
+      "eval_entropy": 0.2340821772813797,
+      "eval_loss": 1.0212821960449219,
+      "eval_mean_token_accuracy": 0.828373622358515,
+      "eval_num_tokens": 8432134.0,
+      "eval_runtime": 87.2518,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 3640
+    },
+    {
+      "entropy": 0.0926868049427867,
+      "epoch": 8.820760410380204,
+      "grad_norm": 0.2618265450000763,
+      "learning_rate": 1.5818263440220203e-05,
+      "loss": 0.07338647246360779,
+      "mean_token_accuracy": 0.9724708944559097,
+      "num_tokens": 8476758.0,
+      "step": 3660
+    },
+    {
+      "epoch": 8.820760410380204,
+      "eval_entropy": 0.23637074575330433,
+      "eval_loss": 1.0186725854873657,
+      "eval_mean_token_accuracy": 0.8277707605549459,
+      "eval_num_tokens": 8476758.0,
+      "eval_runtime": 88.4695,
+      "eval_samples_per_second": 16.051,
+      "eval_steps_per_second": 2.012,
+      "step": 3660
+    },
+    {
+      "entropy": 0.09175782897509635,
+      "epoch": 8.869040434520217,
+      "grad_norm": 0.2889852821826935,
+      "learning_rate": 1.4572418009870604e-05,
+      "loss": 0.07066041231155396,
+      "mean_token_accuracy": 0.9725673109292984,
+      "num_tokens": 8523294.0,
+      "step": 3680
+    },
+    {
+      "epoch": 8.869040434520217,
+      "eval_entropy": 0.23945164404223473,
+      "eval_loss": 1.0145231485366821,
+      "eval_mean_token_accuracy": 0.8281202523895864,
+      "eval_num_tokens": 8523294.0,
+      "eval_runtime": 88.368,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 3680
+    },
+    {
+      "entropy": 0.09265611860901117,
+      "epoch": 8.91732045866023,
+      "grad_norm": 0.41092950105667114,
+      "learning_rate": 1.3375691105172625e-05,
+      "loss": 0.06975027322769164,
+      "mean_token_accuracy": 0.974636297672987,
+      "num_tokens": 8570358.0,
+      "step": 3700
+    },
+    {
+      "epoch": 8.91732045866023,
+      "eval_entropy": 0.239588919901446,
+      "eval_loss": 1.014059066772461,
+      "eval_mean_token_accuracy": 0.828028041660116,
+      "eval_num_tokens": 8570358.0,
+      "eval_runtime": 88.1039,
+      "eval_samples_per_second": 16.117,
+      "eval_steps_per_second": 2.02,
+      "step": 3700
+    },
+    {
+      "entropy": 0.0896465003490448,
+      "epoch": 8.965600482800241,
+      "grad_norm": 0.24000254273414612,
+      "learning_rate": 1.2228421385684305e-05,
+      "loss": 0.06743953824043274,
+      "mean_token_accuracy": 0.974006575345993,
+      "num_tokens": 8618246.0,
+      "step": 3720
+    },
+    {
+      "epoch": 8.965600482800241,
+      "eval_entropy": 0.2390461369846644,
+      "eval_loss": 1.011997938156128,
+      "eval_mean_token_accuracy": 0.8284533211354459,
+      "eval_num_tokens": 8618246.0,
+      "eval_runtime": 88.6202,
+      "eval_samples_per_second": 16.023,
+      "eval_steps_per_second": 2.009,
+      "step": 3720
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.6437102406732595e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3740/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3740/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3740/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3740/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3740/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3740/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3740/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3740/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3740/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3740/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3740/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3740/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3740/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3740/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..bacad24a5b7c9a8de0255a2c23ae3882702f7edd
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3740/trainer_state.json
@@ -0,0 +1,3961 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 9.012070006035003,
+  "eval_steps": 20,
+  "global_step": 3740,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    },
+    {
+      "entropy": 0.1326121488586068,
+      "epoch": 7.181050090525045,
+      "grad_norm": 0.4843117296695709,
+      "learning_rate": 8.411870386507724e-05,
+      "loss": 0.09537842869758606,
+      "mean_token_accuracy": 0.9645151920616627,
+      "num_tokens": 6898444.0,
+      "step": 2980
+    },
+    {
+      "epoch": 7.181050090525045,
+      "eval_entropy": 0.27999477783280813,
+      "eval_loss": 0.9031037092208862,
+      "eval_mean_token_accuracy": 0.8261882204018282,
+      "eval_num_tokens": 6898444.0,
+      "eval_runtime": 88.5602,
+      "eval_samples_per_second": 16.034,
+      "eval_steps_per_second": 2.01,
+      "step": 2980
+    },
+    {
+      "entropy": 0.1156751093454659,
+      "epoch": 7.229330114665057,
+      "grad_norm": 0.5297922492027283,
+      "learning_rate": 8.149629351430326e-05,
+      "loss": 0.09203402996063233,
+      "mean_token_accuracy": 0.9689733698964119,
+      "num_tokens": 6947106.0,
+      "step": 3000
+    },
+    {
+      "epoch": 7.229330114665057,
+      "eval_entropy": 0.2740225845508361,
+      "eval_loss": 0.8876440525054932,
+      "eval_mean_token_accuracy": 0.8265116857678703,
+      "eval_num_tokens": 6947106.0,
+      "eval_runtime": 87.7971,
+      "eval_samples_per_second": 16.174,
+      "eval_steps_per_second": 2.027,
+      "step": 3000
+    },
+    {
+      "entropy": 0.12116114404052496,
+      "epoch": 7.27761013880507,
+      "grad_norm": 0.5467398166656494,
+      "learning_rate": 7.890406302404618e-05,
+      "loss": 0.0933056652545929,
+      "mean_token_accuracy": 0.9676455594599247,
+      "num_tokens": 6992784.0,
+      "step": 3020
+    },
+    {
+      "epoch": 7.27761013880507,
+      "eval_entropy": 0.27238342793804876,
+      "eval_loss": 0.9166117310523987,
+      "eval_mean_token_accuracy": 0.825499015242866,
+      "eval_num_tokens": 6992784.0,
+      "eval_runtime": 88.3276,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 3020
+    },
+    {
+      "entropy": 0.11715119928121567,
+      "epoch": 7.325890162945082,
+      "grad_norm": 0.4976702332496643,
+      "learning_rate": 7.634274596487176e-05,
+      "loss": 0.09449659585952759,
+      "mean_token_accuracy": 0.966506478190422,
+      "num_tokens": 7040482.0,
+      "step": 3040
+    },
+    {
+      "epoch": 7.325890162945082,
+      "eval_entropy": 0.2788975556077582,
+      "eval_loss": 0.8833065629005432,
+      "eval_mean_token_accuracy": 0.8269920596915684,
+      "eval_num_tokens": 7040482.0,
+      "eval_runtime": 88.9016,
+      "eval_samples_per_second": 15.973,
+      "eval_steps_per_second": 2.002,
+      "step": 3040
+    },
+    {
+      "entropy": 0.12565911412239075,
+      "epoch": 7.3741701870850935,
+      "grad_norm": 0.5728679895401001,
+      "learning_rate": 7.381306715921025e-05,
+      "loss": 0.09626244902610778,
+      "mean_token_accuracy": 0.9661444149911403,
+      "num_tokens": 7086261.0,
+      "step": 3060
+    },
+    {
+      "epoch": 7.3741701870850935,
+      "eval_entropy": 0.2762492011939542,
+      "eval_loss": 0.8961160778999329,
+      "eval_mean_token_accuracy": 0.826369910092836,
+      "eval_num_tokens": 7086261.0,
+      "eval_runtime": 88.5204,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 3060
+    },
+    {
+      "entropy": 0.12258567921817302,
+      "epoch": 7.422450211225105,
+      "grad_norm": 0.468985915184021,
+      "learning_rate": 7.131574247624031e-05,
+      "loss": 0.09120266437530518,
+      "mean_token_accuracy": 0.9674938447773457,
+      "num_tokens": 7133247.0,
+      "step": 3080
+    },
+    {
+      "epoch": 7.422450211225105,
+      "eval_entropy": 0.2688512086366,
+      "eval_loss": 0.9048654437065125,
+      "eval_mean_token_accuracy": 0.8265582302982888,
+      "eval_num_tokens": 7133247.0,
+      "eval_runtime": 88.164,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 3080
+    },
+    {
+      "entropy": 0.12070536343380808,
+      "epoch": 7.470730235365117,
+      "grad_norm": 0.5310841798782349,
+      "learning_rate": 6.885147862930688e-05,
+      "loss": 0.09386247992515565,
+      "mean_token_accuracy": 0.96664222702384,
+      "num_tokens": 7177810.0,
+      "step": 3100
+    },
+    {
+      "epoch": 7.470730235365117,
+      "eval_entropy": 0.26289632874593305,
+      "eval_loss": 0.902837336063385,
+      "eval_mean_token_accuracy": 0.828052184554968,
+      "eval_num_tokens": 7177810.0,
+      "eval_runtime": 88.1201,
+      "eval_samples_per_second": 16.114,
+      "eval_steps_per_second": 2.02,
+      "step": 3100
+    },
+    {
+      "entropy": 0.11322665181942285,
+      "epoch": 7.51901025950513,
+      "grad_norm": 0.6846067905426025,
+      "learning_rate": 6.642097297592908e-05,
+      "loss": 0.08772538304328918,
+      "mean_token_accuracy": 0.9691362135112286,
+      "num_tokens": 7225302.0,
+      "step": 3120
+    },
+    {
+      "epoch": 7.51901025950513,
+      "eval_entropy": 0.2670605613776807,
+      "eval_loss": 0.9130566120147705,
+      "eval_mean_token_accuracy": 0.826366903741708,
+      "eval_num_tokens": 7225302.0,
+      "eval_runtime": 88.6035,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3120
+    },
+    {
+      "entropy": 0.10985926580615342,
+      "epoch": 7.567290283645142,
+      "grad_norm": 0.4799748361110687,
+      "learning_rate": 6.402491332045683e-05,
+      "loss": 0.08672296404838561,
+      "mean_token_accuracy": 0.9698824048042297,
+      "num_tokens": 7274881.0,
+      "step": 3140
+    },
+    {
+      "epoch": 7.567290283645142,
+      "eval_entropy": 0.26657010866015146,
+      "eval_loss": 0.9056380987167358,
+      "eval_mean_token_accuracy": 0.8272326461384806,
+      "eval_num_tokens": 7274881.0,
+      "eval_runtime": 88.2967,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3140
+    },
+    {
+      "entropy": 0.11890473761595785,
+      "epoch": 7.615570307785154,
+      "grad_norm": 0.4081912934780121,
+      "learning_rate": 6.166397771942987e-05,
+      "loss": 0.09396827220916748,
+      "mean_token_accuracy": 0.967512384802103,
+      "num_tokens": 7321163.0,
+      "step": 3160
+    },
+    {
+      "epoch": 7.615570307785154,
+      "eval_entropy": 0.26364776575833226,
+      "eval_loss": 0.9204733967781067,
+      "eval_mean_token_accuracy": 0.827419729714983,
+      "eval_num_tokens": 7321163.0,
+      "eval_runtime": 88.3441,
+      "eval_samples_per_second": 16.074,
+      "eval_steps_per_second": 2.015,
+      "step": 3160
+    },
+    {
+      "entropy": 0.12209134344011545,
+      "epoch": 7.663850331925166,
+      "grad_norm": 0.5159774422645569,
+      "learning_rate": 5.93388342896962e-05,
+      "loss": 0.09292978644371033,
+      "mean_token_accuracy": 0.9658421002328396,
+      "num_tokens": 7366217.0,
+      "step": 3180
+    },
+    {
+      "epoch": 7.663850331925166,
+      "eval_entropy": 0.2714110068725736,
+      "eval_loss": 0.9143092036247253,
+      "eval_mean_token_accuracy": 0.8262251099843657,
+      "eval_num_tokens": 7366217.0,
+      "eval_runtime": 87.9635,
+      "eval_samples_per_second": 16.143,
+      "eval_steps_per_second": 2.024,
+      "step": 3180
+    },
+    {
+      "entropy": 0.11955609126016498,
+      "epoch": 7.712130356065178,
+      "grad_norm": 0.7005976438522339,
+      "learning_rate": 5.705014101934252e-05,
+      "loss": 0.09319776892662049,
+      "mean_token_accuracy": 0.9674513377249241,
+      "num_tokens": 7410887.0,
+      "step": 3200
+    },
+    {
+      "epoch": 7.712130356065178,
+      "eval_entropy": 0.2626291513442993,
+      "eval_loss": 0.91884446144104,
+      "eval_mean_token_accuracy": 0.827563955207889,
+      "eval_num_tokens": 7410887.0,
+      "eval_runtime": 87.8849,
+      "eval_samples_per_second": 16.157,
+      "eval_steps_per_second": 2.025,
+      "step": 3200
+    },
+    {
+      "entropy": 0.11767532899975777,
+      "epoch": 7.76041038020519,
+      "grad_norm": 0.4675362706184387,
+      "learning_rate": 5.479854558149187e-05,
+      "loss": 0.09038087725639343,
+      "mean_token_accuracy": 0.9669680051505566,
+      "num_tokens": 7457578.0,
+      "step": 3220
+    },
+    {
+      "epoch": 7.76041038020519,
+      "eval_entropy": 0.26581037203582486,
+      "eval_loss": 0.9146883487701416,
+      "eval_mean_token_accuracy": 0.8270345812433222,
+      "eval_num_tokens": 7457578.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 3220
+    },
+    {
+      "entropy": 0.11383061073720455,
+      "epoch": 7.808690404345202,
+      "grad_norm": 0.541247546672821,
+      "learning_rate": 5.258468515101918e-05,
+      "loss": 0.08942405581474304,
+      "mean_token_accuracy": 0.968398705124855,
+      "num_tokens": 7505761.0,
+      "step": 3240
+    },
+    {
+      "epoch": 7.808690404345202,
+      "eval_entropy": 0.26687998960861997,
+      "eval_loss": 0.9037085175514221,
+      "eval_mean_token_accuracy": 0.8274977863504646,
+      "eval_num_tokens": 7505761.0,
+      "eval_runtime": 88.5163,
+      "eval_samples_per_second": 16.042,
+      "eval_steps_per_second": 2.011,
+      "step": 3240
+    },
+    {
+      "entropy": 0.12004058873280883,
+      "epoch": 7.856970428485214,
+      "grad_norm": 0.5289552211761475,
+      "learning_rate": 5.040918622423899e-05,
+      "loss": 0.09064805507659912,
+      "mean_token_accuracy": 0.9673330299556255,
+      "num_tokens": 7552557.0,
+      "step": 3260
+    },
+    {
+      "epoch": 7.856970428485214,
+      "eval_entropy": 0.26288379251622085,
+      "eval_loss": 0.9146178364753723,
+      "eval_mean_token_accuracy": 0.8276662448149049,
+      "eval_num_tokens": 7552557.0,
+      "eval_runtime": 88.0723,
+      "eval_samples_per_second": 16.123,
+      "eval_steps_per_second": 2.021,
+      "step": 3260
+    },
+    {
+      "entropy": 0.1180689456872642,
+      "epoch": 7.905250452625227,
+      "grad_norm": 0.4754861891269684,
+      "learning_rate": 4.827266444161377e-05,
+      "loss": 0.09402655363082886,
+      "mean_token_accuracy": 0.9663361787796021,
+      "num_tokens": 7596998.0,
+      "step": 3280
+    },
+    {
+      "epoch": 7.905250452625227,
+      "eval_entropy": 0.261690996922134,
+      "eval_loss": 0.9120186567306519,
+      "eval_mean_token_accuracy": 0.8278995015647974,
+      "eval_num_tokens": 7596998.0,
+      "eval_runtime": 88.6573,
+      "eval_samples_per_second": 16.017,
+      "eval_steps_per_second": 2.008,
+      "step": 3280
+    },
+    {
+      "entropy": 0.1162164019420743,
+      "epoch": 7.953530476765239,
+      "grad_norm": 0.6203398108482361,
+      "learning_rate": 4.61757244135355e-05,
+      "loss": 0.08832768201828003,
+      "mean_token_accuracy": 0.9694220684468746,
+      "num_tokens": 7643288.0,
+      "step": 3300
+    },
+    {
+      "epoch": 7.953530476765239,
+      "eval_entropy": 0.26158678129817664,
+      "eval_loss": 0.9301361441612244,
+      "eval_mean_token_accuracy": 0.8269603885291668,
+      "eval_num_tokens": 7643288.0,
+      "eval_runtime": 88.7064,
+      "eval_samples_per_second": 16.008,
+      "eval_steps_per_second": 2.007,
+      "step": 3300
+    },
+    {
+      "entropy": 0.10877202204489089,
+      "epoch": 8.0,
+      "grad_norm": 3.058673143386841,
+      "learning_rate": 4.411895954922763e-05,
+      "loss": 0.09159615635871887,
+      "mean_token_accuracy": 0.9686878834451947,
+      "num_tokens": 7688608.0,
+      "step": 3320
+    },
+    {
+      "epoch": 8.0,
+      "eval_entropy": 0.2590438057197614,
+      "eval_loss": 0.9262065887451172,
+      "eval_mean_token_accuracy": 0.8277404546067956,
+      "eval_num_tokens": 7688608.0,
+      "eval_runtime": 87.7557,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 3320
+    },
+    {
+      "entropy": 0.09969092034734786,
+      "epoch": 8.048280024140013,
+      "grad_norm": 0.2561911642551422,
+      "learning_rate": 4.2102951888817644e-05,
+      "loss": 0.07228735685348511,
+      "mean_token_accuracy": 0.9743556551635265,
+      "num_tokens": 7732856.0,
+      "step": 3340
+    },
+    {
+      "epoch": 8.048280024140013,
+      "eval_entropy": 0.2462534454933713,
+      "eval_loss": 0.9755983352661133,
+      "eval_mean_token_accuracy": 0.827510209565752,
+      "eval_num_tokens": 7732856.0,
+      "eval_runtime": 88.0297,
+      "eval_samples_per_second": 16.131,
+      "eval_steps_per_second": 2.022,
+      "step": 3340
+    },
+    {
+      "entropy": 0.09477624730207027,
+      "epoch": 8.096560048280024,
+      "grad_norm": 0.3670356571674347,
+      "learning_rate": 4.012827193862633e-05,
+      "loss": 0.0686036467552185,
+      "mean_token_accuracy": 0.9743670664727688,
+      "num_tokens": 7778585.0,
+      "step": 3360
+    },
+    {
+      "epoch": 8.096560048280024,
+      "eval_entropy": 0.2440882951356052,
+      "eval_loss": 1.0118423700332642,
+      "eval_mean_token_accuracy": 0.8271241070849172,
+      "eval_num_tokens": 7778585.0,
+      "eval_runtime": 88.1364,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 3360
+    },
+    {
+      "entropy": 0.09417980071157217,
+      "epoch": 8.144840072420036,
+      "grad_norm": 0.5785111784934998,
+      "learning_rate": 3.819547850972133e-05,
+      "loss": 0.07100698947906495,
+      "mean_token_accuracy": 0.9737804099917412,
+      "num_tokens": 7824287.0,
+      "step": 3380
+    },
+    {
+      "epoch": 8.144840072420036,
+      "eval_entropy": 0.24603172860453637,
+      "eval_loss": 1.0021991729736328,
+      "eval_mean_token_accuracy": 0.827162122793412,
+      "eval_num_tokens": 7824287.0,
+      "eval_runtime": 88.4534,
+      "eval_samples_per_second": 16.054,
+      "eval_steps_per_second": 2.012,
+      "step": 3380
+    },
+    {
+      "entropy": 0.087222247896716,
+      "epoch": 8.193120096560047,
+      "grad_norm": 0.44848504662513733,
+      "learning_rate": 3.6305118559780393e-05,
+      "loss": 0.06325234174728393,
+      "mean_token_accuracy": 0.9769335232675076,
+      "num_tokens": 7874784.0,
+      "step": 3400
+    },
+    {
+      "epoch": 8.193120096560047,
+      "eval_entropy": 0.24544863905129807,
+      "eval_loss": 1.005008339881897,
+      "eval_mean_token_accuracy": 0.8273035240976998,
+      "eval_num_tokens": 7874784.0,
+      "eval_runtime": 88.6737,
+      "eval_samples_per_second": 16.014,
+      "eval_steps_per_second": 2.007,
+      "step": 3400
+    },
+    {
+      "entropy": 0.09037275062873959,
+      "epoch": 8.24140012070006,
+      "grad_norm": 0.23529155552387238,
+      "learning_rate": 3.4457727038308435e-05,
+      "loss": 0.06589569449424744,
+      "mean_token_accuracy": 0.9751470290124417,
+      "num_tokens": 7925240.0,
+      "step": 3420
+    },
+    {
+      "epoch": 8.24140012070006,
+      "eval_entropy": 0.24203423494368456,
+      "eval_loss": 1.0052824020385742,
+      "eval_mean_token_accuracy": 0.8275575503874361,
+      "eval_num_tokens": 7925240.0,
+      "eval_runtime": 88.3486,
+      "eval_samples_per_second": 16.073,
+      "eval_steps_per_second": 2.015,
+      "step": 3420
+    },
+    {
+      "entropy": 0.10373208015225828,
+      "epoch": 8.289680144840073,
+      "grad_norm": 0.24796408414840698,
+      "learning_rate": 3.2653826735253594e-05,
+      "loss": 0.07714771628379821,
+      "mean_token_accuracy": 0.9722891598939896,
+      "num_tokens": 7966685.0,
+      "step": 3440
+    },
+    {
+      "epoch": 8.289680144840073,
+      "eval_entropy": 0.23461157050025597,
+      "eval_loss": 1.012108325958252,
+      "eval_mean_token_accuracy": 0.82825900564033,
+      "eval_num_tokens": 7966685.0,
+      "eval_runtime": 88.1018,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 3440
+    },
+    {
+      "entropy": 0.08967954274266958,
+      "epoch": 8.337960168980084,
+      "grad_norm": 0.5314674973487854,
+      "learning_rate": 3.08939281330635e-05,
+      "loss": 0.06746266484260559,
+      "mean_token_accuracy": 0.9753133699297905,
+      "num_tokens": 8013656.0,
+      "step": 3460
+    },
+    {
+      "epoch": 8.337960168980084,
+      "eval_entropy": 0.23781998589467468,
+      "eval_loss": 1.0119913816452026,
+      "eval_mean_token_accuracy": 0.8273287157664139,
+      "eval_num_tokens": 8013656.0,
+      "eval_runtime": 88.6038,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3460
+    },
+    {
+      "entropy": 0.09280927330255509,
+      "epoch": 8.386240193120097,
+      "grad_norm": 0.30658936500549316,
+      "learning_rate": 2.9178529262224946e-05,
+      "loss": 0.06936421990394592,
+      "mean_token_accuracy": 0.9742105580866337,
+      "num_tokens": 8061108.0,
+      "step": 3480
+    },
+    {
+      "epoch": 8.386240193120097,
+      "eval_entropy": 0.23672426734747512,
+      "eval_loss": 1.0162041187286377,
+      "eval_mean_token_accuracy": 0.8278394281194451,
+      "eval_num_tokens": 8061108.0,
+      "eval_runtime": 88.6015,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 3480
+    },
+    {
+      "entropy": 0.09195122267119586,
+      "epoch": 8.43452021726011,
+      "grad_norm": 0.24512454867362976,
+      "learning_rate": 2.7508115560327256e-05,
+      "loss": 0.06996064782142639,
+      "mean_token_accuracy": 0.974104854464531,
+      "num_tokens": 8106924.0,
+      "step": 3500
+    },
+    {
+      "epoch": 8.43452021726011,
+      "eval_entropy": 0.23758245828781235,
+      "eval_loss": 1.0157747268676758,
+      "eval_mean_token_accuracy": 0.8278501579600773,
+      "eval_num_tokens": 8106924.0,
+      "eval_runtime": 88.5678,
+      "eval_samples_per_second": 16.033,
+      "eval_steps_per_second": 2.01,
+      "step": 3500
+    },
+    {
+      "entropy": 0.09518477590754629,
+      "epoch": 8.48280024140012,
+      "grad_norm": 0.2373228371143341,
+      "learning_rate": 2.5883159734688868e-05,
+      "loss": 0.06832797527313232,
+      "mean_token_accuracy": 0.9743806682527065,
+      "num_tokens": 8153070.0,
+      "step": 3520
+    },
+    {
+      "epoch": 8.48280024140012,
+      "eval_entropy": 0.2379308074210467,
+      "eval_loss": 1.0180530548095703,
+      "eval_mean_token_accuracy": 0.8282477052024241,
+      "eval_num_tokens": 8153070.0,
+      "eval_runtime": 88.4651,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 3520
+    },
+    {
+      "entropy": 0.0906269280705601,
+      "epoch": 8.531080265540133,
+      "grad_norm": 0.37602657079696655,
+      "learning_rate": 2.4304121628587173e-05,
+      "loss": 0.06921588778495788,
+      "mean_token_accuracy": 0.9745571479201317,
+      "num_tokens": 8200066.0,
+      "step": 3540
+    },
+    {
+      "epoch": 8.531080265540133,
+      "eval_entropy": 0.2357490769263064,
+      "eval_loss": 1.017139196395874,
+      "eval_mean_token_accuracy": 0.828713178299786,
+      "eval_num_tokens": 8200066.0,
+      "eval_runtime": 88.6424,
+      "eval_samples_per_second": 16.019,
+      "eval_steps_per_second": 2.008,
+      "step": 3540
+    },
+    {
+      "entropy": 0.09273334196768701,
+      "epoch": 8.579360289680144,
+      "grad_norm": 0.2978838086128235,
+      "learning_rate": 2.277144809112794e-05,
+      "loss": 0.06970887780189514,
+      "mean_token_accuracy": 0.9751374177634716,
+      "num_tokens": 8246157.0,
+      "step": 3560
+    },
+    {
+      "epoch": 8.579360289680144,
+      "eval_entropy": 0.23590933138065123,
+      "eval_loss": 1.019231915473938,
+      "eval_mean_token_accuracy": 0.8279163412163767,
+      "eval_num_tokens": 8246157.0,
+      "eval_runtime": 88.2691,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 3560
+    },
+    {
+      "entropy": 0.09854201921261847,
+      "epoch": 8.627640313820157,
+      "grad_norm": 0.5381243228912354,
+      "learning_rate": 2.1285572850792413e-05,
+      "loss": 0.07123501300811767,
+      "mean_token_accuracy": 0.9723988175392151,
+      "num_tokens": 8291752.0,
+      "step": 3580
+    },
+    {
+      "epoch": 8.627640313820157,
+      "eval_entropy": 0.23592024949494372,
+      "eval_loss": 1.020250678062439,
+      "eval_mean_token_accuracy": 0.8282466226079491,
+      "eval_num_tokens": 8291752.0,
+      "eval_runtime": 87.8939,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 3580
+    },
+    {
+      "entropy": 0.09092000788077712,
+      "epoch": 8.67592033796017,
+      "grad_norm": 0.29089561104774475,
+      "learning_rate": 1.9846916392697493e-05,
+      "loss": 0.06982720494270325,
+      "mean_token_accuracy": 0.9749995484948158,
+      "num_tokens": 8337651.0,
+      "step": 3600
+    },
+    {
+      "epoch": 8.67592033796017,
+      "eval_entropy": 0.23658913382318583,
+      "eval_loss": 1.019862413406372,
+      "eval_mean_token_accuracy": 0.8280127842104836,
+      "eval_num_tokens": 8337651.0,
+      "eval_runtime": 88.2855,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 3600
+    },
+    {
+      "entropy": 0.08708424470387399,
+      "epoch": 8.72420036210018,
+      "grad_norm": 0.29904839396476746,
+      "learning_rate": 1.8455885839603106e-05,
+      "loss": 0.06801370978355407,
+      "mean_token_accuracy": 0.9755801670253277,
+      "num_tokens": 8385330.0,
+      "step": 3620
+    },
+    {
+      "epoch": 8.72420036210018,
+      "eval_entropy": 0.23627035771862845,
+      "eval_loss": 1.0204569101333618,
+      "eval_mean_token_accuracy": 0.8279967894045155,
+      "eval_num_tokens": 8385330.0,
+      "eval_runtime": 88.1753,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 3620
+    },
+    {
+      "entropy": 0.0916255566291511,
+      "epoch": 8.772480386240193,
+      "grad_norm": 0.20571033656597137,
+      "learning_rate": 1.7112874836701502e-05,
+      "loss": 0.0692249894142151,
+      "mean_token_accuracy": 0.9745148062705994,
+      "num_tokens": 8432134.0,
+      "step": 3640
+    },
+    {
+      "epoch": 8.772480386240193,
+      "eval_entropy": 0.2340821772813797,
+      "eval_loss": 1.0212821960449219,
+      "eval_mean_token_accuracy": 0.828373622358515,
+      "eval_num_tokens": 8432134.0,
+      "eval_runtime": 87.2518,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 3640
+    },
+    {
+      "entropy": 0.0926868049427867,
+      "epoch": 8.820760410380204,
+      "grad_norm": 0.2618265450000763,
+      "learning_rate": 1.5818263440220203e-05,
+      "loss": 0.07338647246360779,
+      "mean_token_accuracy": 0.9724708944559097,
+      "num_tokens": 8476758.0,
+      "step": 3660
+    },
+    {
+      "epoch": 8.820760410380204,
+      "eval_entropy": 0.23637074575330433,
+      "eval_loss": 1.0186725854873657,
+      "eval_mean_token_accuracy": 0.8277707605549459,
+      "eval_num_tokens": 8476758.0,
+      "eval_runtime": 88.4695,
+      "eval_samples_per_second": 16.051,
+      "eval_steps_per_second": 2.012,
+      "step": 3660
+    },
+    {
+      "entropy": 0.09175782897509635,
+      "epoch": 8.869040434520217,
+      "grad_norm": 0.2889852821826935,
+      "learning_rate": 1.4572418009870604e-05,
+      "loss": 0.07066041231155396,
+      "mean_token_accuracy": 0.9725673109292984,
+      "num_tokens": 8523294.0,
+      "step": 3680
+    },
+    {
+      "epoch": 8.869040434520217,
+      "eval_entropy": 0.23945164404223473,
+      "eval_loss": 1.0145231485366821,
+      "eval_mean_token_accuracy": 0.8281202523895864,
+      "eval_num_tokens": 8523294.0,
+      "eval_runtime": 88.368,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 3680
+    },
+    {
+      "entropy": 0.09265611860901117,
+      "epoch": 8.91732045866023,
+      "grad_norm": 0.41092950105667114,
+      "learning_rate": 1.3375691105172625e-05,
+      "loss": 0.06975027322769164,
+      "mean_token_accuracy": 0.974636297672987,
+      "num_tokens": 8570358.0,
+      "step": 3700
+    },
+    {
+      "epoch": 8.91732045866023,
+      "eval_entropy": 0.239588919901446,
+      "eval_loss": 1.014059066772461,
+      "eval_mean_token_accuracy": 0.828028041660116,
+      "eval_num_tokens": 8570358.0,
+      "eval_runtime": 88.1039,
+      "eval_samples_per_second": 16.117,
+      "eval_steps_per_second": 2.02,
+      "step": 3700
+    },
+    {
+      "entropy": 0.0896465003490448,
+      "epoch": 8.965600482800241,
+      "grad_norm": 0.24000254273414612,
+      "learning_rate": 1.2228421385684305e-05,
+      "loss": 0.06743953824043274,
+      "mean_token_accuracy": 0.974006575345993,
+      "num_tokens": 8618246.0,
+      "step": 3720
+    },
+    {
+      "epoch": 8.965600482800241,
+      "eval_entropy": 0.2390461369846644,
+      "eval_loss": 1.011997938156128,
+      "eval_mean_token_accuracy": 0.8284533211354459,
+      "eval_num_tokens": 8618246.0,
+      "eval_runtime": 88.6202,
+      "eval_samples_per_second": 16.023,
+      "eval_steps_per_second": 2.009,
+      "step": 3720
+    },
+    {
+      "entropy": 0.09575109251520851,
+      "epoch": 9.012070006035003,
+      "grad_norm": 0.18088576197624207,
+      "learning_rate": 1.1130933515165429e-05,
+      "loss": 0.07295305132865906,
+      "mean_token_accuracy": 0.9739722857227573,
+      "num_tokens": 8661177.0,
+      "step": 3740
+    },
+    {
+      "epoch": 9.012070006035003,
+      "eval_entropy": 0.23728882387447892,
+      "eval_loss": 1.0134038925170898,
+      "eval_mean_token_accuracy": 0.8286508951294288,
+      "eval_num_tokens": 8661177.0,
+      "eval_runtime": 88.6434,
+      "eval_samples_per_second": 16.019,
+      "eval_steps_per_second": 2.008,
+      "step": 3740
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.662200233853747e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3760/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3760/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3760/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3760/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3760/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3760/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3760/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3760/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3760/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3760/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3760/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3760/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3760/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3760/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..f76776ca4eeda3bf7bc462e883fefa9e10a7509f
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3760/trainer_state.json
@@ -0,0 +1,3982 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 9.060350030175016,
+  "eval_steps": 20,
+  "global_step": 3760,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    },
+    {
+      "entropy": 0.1326121488586068,
+      "epoch": 7.181050090525045,
+      "grad_norm": 0.4843117296695709,
+      "learning_rate": 8.411870386507724e-05,
+      "loss": 0.09537842869758606,
+      "mean_token_accuracy": 0.9645151920616627,
+      "num_tokens": 6898444.0,
+      "step": 2980
+    },
+    {
+      "epoch": 7.181050090525045,
+      "eval_entropy": 0.27999477783280813,
+      "eval_loss": 0.9031037092208862,
+      "eval_mean_token_accuracy": 0.8261882204018282,
+      "eval_num_tokens": 6898444.0,
+      "eval_runtime": 88.5602,
+      "eval_samples_per_second": 16.034,
+      "eval_steps_per_second": 2.01,
+      "step": 2980
+    },
+    {
+      "entropy": 0.1156751093454659,
+      "epoch": 7.229330114665057,
+      "grad_norm": 0.5297922492027283,
+      "learning_rate": 8.149629351430326e-05,
+      "loss": 0.09203402996063233,
+      "mean_token_accuracy": 0.9689733698964119,
+      "num_tokens": 6947106.0,
+      "step": 3000
+    },
+    {
+      "epoch": 7.229330114665057,
+      "eval_entropy": 0.2740225845508361,
+      "eval_loss": 0.8876440525054932,
+      "eval_mean_token_accuracy": 0.8265116857678703,
+      "eval_num_tokens": 6947106.0,
+      "eval_runtime": 87.7971,
+      "eval_samples_per_second": 16.174,
+      "eval_steps_per_second": 2.027,
+      "step": 3000
+    },
+    {
+      "entropy": 0.12116114404052496,
+      "epoch": 7.27761013880507,
+      "grad_norm": 0.5467398166656494,
+      "learning_rate": 7.890406302404618e-05,
+      "loss": 0.0933056652545929,
+      "mean_token_accuracy": 0.9676455594599247,
+      "num_tokens": 6992784.0,
+      "step": 3020
+    },
+    {
+      "epoch": 7.27761013880507,
+      "eval_entropy": 0.27238342793804876,
+      "eval_loss": 0.9166117310523987,
+      "eval_mean_token_accuracy": 0.825499015242866,
+      "eval_num_tokens": 6992784.0,
+      "eval_runtime": 88.3276,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 3020
+    },
+    {
+      "entropy": 0.11715119928121567,
+      "epoch": 7.325890162945082,
+      "grad_norm": 0.4976702332496643,
+      "learning_rate": 7.634274596487176e-05,
+      "loss": 0.09449659585952759,
+      "mean_token_accuracy": 0.966506478190422,
+      "num_tokens": 7040482.0,
+      "step": 3040
+    },
+    {
+      "epoch": 7.325890162945082,
+      "eval_entropy": 0.2788975556077582,
+      "eval_loss": 0.8833065629005432,
+      "eval_mean_token_accuracy": 0.8269920596915684,
+      "eval_num_tokens": 7040482.0,
+      "eval_runtime": 88.9016,
+      "eval_samples_per_second": 15.973,
+      "eval_steps_per_second": 2.002,
+      "step": 3040
+    },
+    {
+      "entropy": 0.12565911412239075,
+      "epoch": 7.3741701870850935,
+      "grad_norm": 0.5728679895401001,
+      "learning_rate": 7.381306715921025e-05,
+      "loss": 0.09626244902610778,
+      "mean_token_accuracy": 0.9661444149911403,
+      "num_tokens": 7086261.0,
+      "step": 3060
+    },
+    {
+      "epoch": 7.3741701870850935,
+      "eval_entropy": 0.2762492011939542,
+      "eval_loss": 0.8961160778999329,
+      "eval_mean_token_accuracy": 0.826369910092836,
+      "eval_num_tokens": 7086261.0,
+      "eval_runtime": 88.5204,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 3060
+    },
+    {
+      "entropy": 0.12258567921817302,
+      "epoch": 7.422450211225105,
+      "grad_norm": 0.468985915184021,
+      "learning_rate": 7.131574247624031e-05,
+      "loss": 0.09120266437530518,
+      "mean_token_accuracy": 0.9674938447773457,
+      "num_tokens": 7133247.0,
+      "step": 3080
+    },
+    {
+      "epoch": 7.422450211225105,
+      "eval_entropy": 0.2688512086366,
+      "eval_loss": 0.9048654437065125,
+      "eval_mean_token_accuracy": 0.8265582302982888,
+      "eval_num_tokens": 7133247.0,
+      "eval_runtime": 88.164,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 3080
+    },
+    {
+      "entropy": 0.12070536343380808,
+      "epoch": 7.470730235365117,
+      "grad_norm": 0.5310841798782349,
+      "learning_rate": 6.885147862930688e-05,
+      "loss": 0.09386247992515565,
+      "mean_token_accuracy": 0.96664222702384,
+      "num_tokens": 7177810.0,
+      "step": 3100
+    },
+    {
+      "epoch": 7.470730235365117,
+      "eval_entropy": 0.26289632874593305,
+      "eval_loss": 0.902837336063385,
+      "eval_mean_token_accuracy": 0.828052184554968,
+      "eval_num_tokens": 7177810.0,
+      "eval_runtime": 88.1201,
+      "eval_samples_per_second": 16.114,
+      "eval_steps_per_second": 2.02,
+      "step": 3100
+    },
+    {
+      "entropy": 0.11322665181942285,
+      "epoch": 7.51901025950513,
+      "grad_norm": 0.6846067905426025,
+      "learning_rate": 6.642097297592908e-05,
+      "loss": 0.08772538304328918,
+      "mean_token_accuracy": 0.9691362135112286,
+      "num_tokens": 7225302.0,
+      "step": 3120
+    },
+    {
+      "epoch": 7.51901025950513,
+      "eval_entropy": 0.2670605613776807,
+      "eval_loss": 0.9130566120147705,
+      "eval_mean_token_accuracy": 0.826366903741708,
+      "eval_num_tokens": 7225302.0,
+      "eval_runtime": 88.6035,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3120
+    },
+    {
+      "entropy": 0.10985926580615342,
+      "epoch": 7.567290283645142,
+      "grad_norm": 0.4799748361110687,
+      "learning_rate": 6.402491332045683e-05,
+      "loss": 0.08672296404838561,
+      "mean_token_accuracy": 0.9698824048042297,
+      "num_tokens": 7274881.0,
+      "step": 3140
+    },
+    {
+      "epoch": 7.567290283645142,
+      "eval_entropy": 0.26657010866015146,
+      "eval_loss": 0.9056380987167358,
+      "eval_mean_token_accuracy": 0.8272326461384806,
+      "eval_num_tokens": 7274881.0,
+      "eval_runtime": 88.2967,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3140
+    },
+    {
+      "entropy": 0.11890473761595785,
+      "epoch": 7.615570307785154,
+      "grad_norm": 0.4081912934780121,
+      "learning_rate": 6.166397771942987e-05,
+      "loss": 0.09396827220916748,
+      "mean_token_accuracy": 0.967512384802103,
+      "num_tokens": 7321163.0,
+      "step": 3160
+    },
+    {
+      "epoch": 7.615570307785154,
+      "eval_entropy": 0.26364776575833226,
+      "eval_loss": 0.9204733967781067,
+      "eval_mean_token_accuracy": 0.827419729714983,
+      "eval_num_tokens": 7321163.0,
+      "eval_runtime": 88.3441,
+      "eval_samples_per_second": 16.074,
+      "eval_steps_per_second": 2.015,
+      "step": 3160
+    },
+    {
+      "entropy": 0.12209134344011545,
+      "epoch": 7.663850331925166,
+      "grad_norm": 0.5159774422645569,
+      "learning_rate": 5.93388342896962e-05,
+      "loss": 0.09292978644371033,
+      "mean_token_accuracy": 0.9658421002328396,
+      "num_tokens": 7366217.0,
+      "step": 3180
+    },
+    {
+      "epoch": 7.663850331925166,
+      "eval_entropy": 0.2714110068725736,
+      "eval_loss": 0.9143092036247253,
+      "eval_mean_token_accuracy": 0.8262251099843657,
+      "eval_num_tokens": 7366217.0,
+      "eval_runtime": 87.9635,
+      "eval_samples_per_second": 16.143,
+      "eval_steps_per_second": 2.024,
+      "step": 3180
+    },
+    {
+      "entropy": 0.11955609126016498,
+      "epoch": 7.712130356065178,
+      "grad_norm": 0.7005976438522339,
+      "learning_rate": 5.705014101934252e-05,
+      "loss": 0.09319776892662049,
+      "mean_token_accuracy": 0.9674513377249241,
+      "num_tokens": 7410887.0,
+      "step": 3200
+    },
+    {
+      "epoch": 7.712130356065178,
+      "eval_entropy": 0.2626291513442993,
+      "eval_loss": 0.91884446144104,
+      "eval_mean_token_accuracy": 0.827563955207889,
+      "eval_num_tokens": 7410887.0,
+      "eval_runtime": 87.8849,
+      "eval_samples_per_second": 16.157,
+      "eval_steps_per_second": 2.025,
+      "step": 3200
+    },
+    {
+      "entropy": 0.11767532899975777,
+      "epoch": 7.76041038020519,
+      "grad_norm": 0.4675362706184387,
+      "learning_rate": 5.479854558149187e-05,
+      "loss": 0.09038087725639343,
+      "mean_token_accuracy": 0.9669680051505566,
+      "num_tokens": 7457578.0,
+      "step": 3220
+    },
+    {
+      "epoch": 7.76041038020519,
+      "eval_entropy": 0.26581037203582486,
+      "eval_loss": 0.9146883487701416,
+      "eval_mean_token_accuracy": 0.8270345812433222,
+      "eval_num_tokens": 7457578.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 3220
+    },
+    {
+      "entropy": 0.11383061073720455,
+      "epoch": 7.808690404345202,
+      "grad_norm": 0.541247546672821,
+      "learning_rate": 5.258468515101918e-05,
+      "loss": 0.08942405581474304,
+      "mean_token_accuracy": 0.968398705124855,
+      "num_tokens": 7505761.0,
+      "step": 3240
+    },
+    {
+      "epoch": 7.808690404345202,
+      "eval_entropy": 0.26687998960861997,
+      "eval_loss": 0.9037085175514221,
+      "eval_mean_token_accuracy": 0.8274977863504646,
+      "eval_num_tokens": 7505761.0,
+      "eval_runtime": 88.5163,
+      "eval_samples_per_second": 16.042,
+      "eval_steps_per_second": 2.011,
+      "step": 3240
+    },
+    {
+      "entropy": 0.12004058873280883,
+      "epoch": 7.856970428485214,
+      "grad_norm": 0.5289552211761475,
+      "learning_rate": 5.040918622423899e-05,
+      "loss": 0.09064805507659912,
+      "mean_token_accuracy": 0.9673330299556255,
+      "num_tokens": 7552557.0,
+      "step": 3260
+    },
+    {
+      "epoch": 7.856970428485214,
+      "eval_entropy": 0.26288379251622085,
+      "eval_loss": 0.9146178364753723,
+      "eval_mean_token_accuracy": 0.8276662448149049,
+      "eval_num_tokens": 7552557.0,
+      "eval_runtime": 88.0723,
+      "eval_samples_per_second": 16.123,
+      "eval_steps_per_second": 2.021,
+      "step": 3260
+    },
+    {
+      "entropy": 0.1180689456872642,
+      "epoch": 7.905250452625227,
+      "grad_norm": 0.4754861891269684,
+      "learning_rate": 4.827266444161377e-05,
+      "loss": 0.09402655363082886,
+      "mean_token_accuracy": 0.9663361787796021,
+      "num_tokens": 7596998.0,
+      "step": 3280
+    },
+    {
+      "epoch": 7.905250452625227,
+      "eval_entropy": 0.261690996922134,
+      "eval_loss": 0.9120186567306519,
+      "eval_mean_token_accuracy": 0.8278995015647974,
+      "eval_num_tokens": 7596998.0,
+      "eval_runtime": 88.6573,
+      "eval_samples_per_second": 16.017,
+      "eval_steps_per_second": 2.008,
+      "step": 3280
+    },
+    {
+      "entropy": 0.1162164019420743,
+      "epoch": 7.953530476765239,
+      "grad_norm": 0.6203398108482361,
+      "learning_rate": 4.61757244135355e-05,
+      "loss": 0.08832768201828003,
+      "mean_token_accuracy": 0.9694220684468746,
+      "num_tokens": 7643288.0,
+      "step": 3300
+    },
+    {
+      "epoch": 7.953530476765239,
+      "eval_entropy": 0.26158678129817664,
+      "eval_loss": 0.9301361441612244,
+      "eval_mean_token_accuracy": 0.8269603885291668,
+      "eval_num_tokens": 7643288.0,
+      "eval_runtime": 88.7064,
+      "eval_samples_per_second": 16.008,
+      "eval_steps_per_second": 2.007,
+      "step": 3300
+    },
+    {
+      "entropy": 0.10877202204489089,
+      "epoch": 8.0,
+      "grad_norm": 3.058673143386841,
+      "learning_rate": 4.411895954922763e-05,
+      "loss": 0.09159615635871887,
+      "mean_token_accuracy": 0.9686878834451947,
+      "num_tokens": 7688608.0,
+      "step": 3320
+    },
+    {
+      "epoch": 8.0,
+      "eval_entropy": 0.2590438057197614,
+      "eval_loss": 0.9262065887451172,
+      "eval_mean_token_accuracy": 0.8277404546067956,
+      "eval_num_tokens": 7688608.0,
+      "eval_runtime": 87.7557,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 3320
+    },
+    {
+      "entropy": 0.09969092034734786,
+      "epoch": 8.048280024140013,
+      "grad_norm": 0.2561911642551422,
+      "learning_rate": 4.2102951888817644e-05,
+      "loss": 0.07228735685348511,
+      "mean_token_accuracy": 0.9743556551635265,
+      "num_tokens": 7732856.0,
+      "step": 3340
+    },
+    {
+      "epoch": 8.048280024140013,
+      "eval_entropy": 0.2462534454933713,
+      "eval_loss": 0.9755983352661133,
+      "eval_mean_token_accuracy": 0.827510209565752,
+      "eval_num_tokens": 7732856.0,
+      "eval_runtime": 88.0297,
+      "eval_samples_per_second": 16.131,
+      "eval_steps_per_second": 2.022,
+      "step": 3340
+    },
+    {
+      "entropy": 0.09477624730207027,
+      "epoch": 8.096560048280024,
+      "grad_norm": 0.3670356571674347,
+      "learning_rate": 4.012827193862633e-05,
+      "loss": 0.0686036467552185,
+      "mean_token_accuracy": 0.9743670664727688,
+      "num_tokens": 7778585.0,
+      "step": 3360
+    },
+    {
+      "epoch": 8.096560048280024,
+      "eval_entropy": 0.2440882951356052,
+      "eval_loss": 1.0118423700332642,
+      "eval_mean_token_accuracy": 0.8271241070849172,
+      "eval_num_tokens": 7778585.0,
+      "eval_runtime": 88.1364,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 3360
+    },
+    {
+      "entropy": 0.09417980071157217,
+      "epoch": 8.144840072420036,
+      "grad_norm": 0.5785111784934998,
+      "learning_rate": 3.819547850972133e-05,
+      "loss": 0.07100698947906495,
+      "mean_token_accuracy": 0.9737804099917412,
+      "num_tokens": 7824287.0,
+      "step": 3380
+    },
+    {
+      "epoch": 8.144840072420036,
+      "eval_entropy": 0.24603172860453637,
+      "eval_loss": 1.0021991729736328,
+      "eval_mean_token_accuracy": 0.827162122793412,
+      "eval_num_tokens": 7824287.0,
+      "eval_runtime": 88.4534,
+      "eval_samples_per_second": 16.054,
+      "eval_steps_per_second": 2.012,
+      "step": 3380
+    },
+    {
+      "entropy": 0.087222247896716,
+      "epoch": 8.193120096560047,
+      "grad_norm": 0.44848504662513733,
+      "learning_rate": 3.6305118559780393e-05,
+      "loss": 0.06325234174728393,
+      "mean_token_accuracy": 0.9769335232675076,
+      "num_tokens": 7874784.0,
+      "step": 3400
+    },
+    {
+      "epoch": 8.193120096560047,
+      "eval_entropy": 0.24544863905129807,
+      "eval_loss": 1.005008339881897,
+      "eval_mean_token_accuracy": 0.8273035240976998,
+      "eval_num_tokens": 7874784.0,
+      "eval_runtime": 88.6737,
+      "eval_samples_per_second": 16.014,
+      "eval_steps_per_second": 2.007,
+      "step": 3400
+    },
+    {
+      "entropy": 0.09037275062873959,
+      "epoch": 8.24140012070006,
+      "grad_norm": 0.23529155552387238,
+      "learning_rate": 3.4457727038308435e-05,
+      "loss": 0.06589569449424744,
+      "mean_token_accuracy": 0.9751470290124417,
+      "num_tokens": 7925240.0,
+      "step": 3420
+    },
+    {
+      "epoch": 8.24140012070006,
+      "eval_entropy": 0.24203423494368456,
+      "eval_loss": 1.0052824020385742,
+      "eval_mean_token_accuracy": 0.8275575503874361,
+      "eval_num_tokens": 7925240.0,
+      "eval_runtime": 88.3486,
+      "eval_samples_per_second": 16.073,
+      "eval_steps_per_second": 2.015,
+      "step": 3420
+    },
+    {
+      "entropy": 0.10373208015225828,
+      "epoch": 8.289680144840073,
+      "grad_norm": 0.24796408414840698,
+      "learning_rate": 3.2653826735253594e-05,
+      "loss": 0.07714771628379821,
+      "mean_token_accuracy": 0.9722891598939896,
+      "num_tokens": 7966685.0,
+      "step": 3440
+    },
+    {
+      "epoch": 8.289680144840073,
+      "eval_entropy": 0.23461157050025597,
+      "eval_loss": 1.012108325958252,
+      "eval_mean_token_accuracy": 0.82825900564033,
+      "eval_num_tokens": 7966685.0,
+      "eval_runtime": 88.1018,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 3440
+    },
+    {
+      "entropy": 0.08967954274266958,
+      "epoch": 8.337960168980084,
+      "grad_norm": 0.5314674973487854,
+      "learning_rate": 3.08939281330635e-05,
+      "loss": 0.06746266484260559,
+      "mean_token_accuracy": 0.9753133699297905,
+      "num_tokens": 8013656.0,
+      "step": 3460
+    },
+    {
+      "epoch": 8.337960168980084,
+      "eval_entropy": 0.23781998589467468,
+      "eval_loss": 1.0119913816452026,
+      "eval_mean_token_accuracy": 0.8273287157664139,
+      "eval_num_tokens": 8013656.0,
+      "eval_runtime": 88.6038,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3460
+    },
+    {
+      "entropy": 0.09280927330255509,
+      "epoch": 8.386240193120097,
+      "grad_norm": 0.30658936500549316,
+      "learning_rate": 2.9178529262224946e-05,
+      "loss": 0.06936421990394592,
+      "mean_token_accuracy": 0.9742105580866337,
+      "num_tokens": 8061108.0,
+      "step": 3480
+    },
+    {
+      "epoch": 8.386240193120097,
+      "eval_entropy": 0.23672426734747512,
+      "eval_loss": 1.0162041187286377,
+      "eval_mean_token_accuracy": 0.8278394281194451,
+      "eval_num_tokens": 8061108.0,
+      "eval_runtime": 88.6015,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 3480
+    },
+    {
+      "entropy": 0.09195122267119586,
+      "epoch": 8.43452021726011,
+      "grad_norm": 0.24512454867362976,
+      "learning_rate": 2.7508115560327256e-05,
+      "loss": 0.06996064782142639,
+      "mean_token_accuracy": 0.974104854464531,
+      "num_tokens": 8106924.0,
+      "step": 3500
+    },
+    {
+      "epoch": 8.43452021726011,
+      "eval_entropy": 0.23758245828781235,
+      "eval_loss": 1.0157747268676758,
+      "eval_mean_token_accuracy": 0.8278501579600773,
+      "eval_num_tokens": 8106924.0,
+      "eval_runtime": 88.5678,
+      "eval_samples_per_second": 16.033,
+      "eval_steps_per_second": 2.01,
+      "step": 3500
+    },
+    {
+      "entropy": 0.09518477590754629,
+      "epoch": 8.48280024140012,
+      "grad_norm": 0.2373228371143341,
+      "learning_rate": 2.5883159734688868e-05,
+      "loss": 0.06832797527313232,
+      "mean_token_accuracy": 0.9743806682527065,
+      "num_tokens": 8153070.0,
+      "step": 3520
+    },
+    {
+      "epoch": 8.48280024140012,
+      "eval_entropy": 0.2379308074210467,
+      "eval_loss": 1.0180530548095703,
+      "eval_mean_token_accuracy": 0.8282477052024241,
+      "eval_num_tokens": 8153070.0,
+      "eval_runtime": 88.4651,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 3520
+    },
+    {
+      "entropy": 0.0906269280705601,
+      "epoch": 8.531080265540133,
+      "grad_norm": 0.37602657079696655,
+      "learning_rate": 2.4304121628587173e-05,
+      "loss": 0.06921588778495788,
+      "mean_token_accuracy": 0.9745571479201317,
+      "num_tokens": 8200066.0,
+      "step": 3540
+    },
+    {
+      "epoch": 8.531080265540133,
+      "eval_entropy": 0.2357490769263064,
+      "eval_loss": 1.017139196395874,
+      "eval_mean_token_accuracy": 0.828713178299786,
+      "eval_num_tokens": 8200066.0,
+      "eval_runtime": 88.6424,
+      "eval_samples_per_second": 16.019,
+      "eval_steps_per_second": 2.008,
+      "step": 3540
+    },
+    {
+      "entropy": 0.09273334196768701,
+      "epoch": 8.579360289680144,
+      "grad_norm": 0.2978838086128235,
+      "learning_rate": 2.277144809112794e-05,
+      "loss": 0.06970887780189514,
+      "mean_token_accuracy": 0.9751374177634716,
+      "num_tokens": 8246157.0,
+      "step": 3560
+    },
+    {
+      "epoch": 8.579360289680144,
+      "eval_entropy": 0.23590933138065123,
+      "eval_loss": 1.019231915473938,
+      "eval_mean_token_accuracy": 0.8279163412163767,
+      "eval_num_tokens": 8246157.0,
+      "eval_runtime": 88.2691,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 3560
+    },
+    {
+      "entropy": 0.09854201921261847,
+      "epoch": 8.627640313820157,
+      "grad_norm": 0.5381243228912354,
+      "learning_rate": 2.1285572850792413e-05,
+      "loss": 0.07123501300811767,
+      "mean_token_accuracy": 0.9723988175392151,
+      "num_tokens": 8291752.0,
+      "step": 3580
+    },
+    {
+      "epoch": 8.627640313820157,
+      "eval_entropy": 0.23592024949494372,
+      "eval_loss": 1.020250678062439,
+      "eval_mean_token_accuracy": 0.8282466226079491,
+      "eval_num_tokens": 8291752.0,
+      "eval_runtime": 87.8939,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 3580
+    },
+    {
+      "entropy": 0.09092000788077712,
+      "epoch": 8.67592033796017,
+      "grad_norm": 0.29089561104774475,
+      "learning_rate": 1.9846916392697493e-05,
+      "loss": 0.06982720494270325,
+      "mean_token_accuracy": 0.9749995484948158,
+      "num_tokens": 8337651.0,
+      "step": 3600
+    },
+    {
+      "epoch": 8.67592033796017,
+      "eval_entropy": 0.23658913382318583,
+      "eval_loss": 1.019862413406372,
+      "eval_mean_token_accuracy": 0.8280127842104836,
+      "eval_num_tokens": 8337651.0,
+      "eval_runtime": 88.2855,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 3600
+    },
+    {
+      "entropy": 0.08708424470387399,
+      "epoch": 8.72420036210018,
+      "grad_norm": 0.29904839396476746,
+      "learning_rate": 1.8455885839603106e-05,
+      "loss": 0.06801370978355407,
+      "mean_token_accuracy": 0.9755801670253277,
+      "num_tokens": 8385330.0,
+      "step": 3620
+    },
+    {
+      "epoch": 8.72420036210018,
+      "eval_entropy": 0.23627035771862845,
+      "eval_loss": 1.0204569101333618,
+      "eval_mean_token_accuracy": 0.8279967894045155,
+      "eval_num_tokens": 8385330.0,
+      "eval_runtime": 88.1753,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 3620
+    },
+    {
+      "entropy": 0.0916255566291511,
+      "epoch": 8.772480386240193,
+      "grad_norm": 0.20571033656597137,
+      "learning_rate": 1.7112874836701502e-05,
+      "loss": 0.0692249894142151,
+      "mean_token_accuracy": 0.9745148062705994,
+      "num_tokens": 8432134.0,
+      "step": 3640
+    },
+    {
+      "epoch": 8.772480386240193,
+      "eval_entropy": 0.2340821772813797,
+      "eval_loss": 1.0212821960449219,
+      "eval_mean_token_accuracy": 0.828373622358515,
+      "eval_num_tokens": 8432134.0,
+      "eval_runtime": 87.2518,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 3640
+    },
+    {
+      "entropy": 0.0926868049427867,
+      "epoch": 8.820760410380204,
+      "grad_norm": 0.2618265450000763,
+      "learning_rate": 1.5818263440220203e-05,
+      "loss": 0.07338647246360779,
+      "mean_token_accuracy": 0.9724708944559097,
+      "num_tokens": 8476758.0,
+      "step": 3660
+    },
+    {
+      "epoch": 8.820760410380204,
+      "eval_entropy": 0.23637074575330433,
+      "eval_loss": 1.0186725854873657,
+      "eval_mean_token_accuracy": 0.8277707605549459,
+      "eval_num_tokens": 8476758.0,
+      "eval_runtime": 88.4695,
+      "eval_samples_per_second": 16.051,
+      "eval_steps_per_second": 2.012,
+      "step": 3660
+    },
+    {
+      "entropy": 0.09175782897509635,
+      "epoch": 8.869040434520217,
+      "grad_norm": 0.2889852821826935,
+      "learning_rate": 1.4572418009870604e-05,
+      "loss": 0.07066041231155396,
+      "mean_token_accuracy": 0.9725673109292984,
+      "num_tokens": 8523294.0,
+      "step": 3680
+    },
+    {
+      "epoch": 8.869040434520217,
+      "eval_entropy": 0.23945164404223473,
+      "eval_loss": 1.0145231485366821,
+      "eval_mean_token_accuracy": 0.8281202523895864,
+      "eval_num_tokens": 8523294.0,
+      "eval_runtime": 88.368,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 3680
+    },
+    {
+      "entropy": 0.09265611860901117,
+      "epoch": 8.91732045866023,
+      "grad_norm": 0.41092950105667114,
+      "learning_rate": 1.3375691105172625e-05,
+      "loss": 0.06975027322769164,
+      "mean_token_accuracy": 0.974636297672987,
+      "num_tokens": 8570358.0,
+      "step": 3700
+    },
+    {
+      "epoch": 8.91732045866023,
+      "eval_entropy": 0.239588919901446,
+      "eval_loss": 1.014059066772461,
+      "eval_mean_token_accuracy": 0.828028041660116,
+      "eval_num_tokens": 8570358.0,
+      "eval_runtime": 88.1039,
+      "eval_samples_per_second": 16.117,
+      "eval_steps_per_second": 2.02,
+      "step": 3700
+    },
+    {
+      "entropy": 0.0896465003490448,
+      "epoch": 8.965600482800241,
+      "grad_norm": 0.24000254273414612,
+      "learning_rate": 1.2228421385684305e-05,
+      "loss": 0.06743953824043274,
+      "mean_token_accuracy": 0.974006575345993,
+      "num_tokens": 8618246.0,
+      "step": 3720
+    },
+    {
+      "epoch": 8.965600482800241,
+      "eval_entropy": 0.2390461369846644,
+      "eval_loss": 1.011997938156128,
+      "eval_mean_token_accuracy": 0.8284533211354459,
+      "eval_num_tokens": 8618246.0,
+      "eval_runtime": 88.6202,
+      "eval_samples_per_second": 16.023,
+      "eval_steps_per_second": 2.009,
+      "step": 3720
+    },
+    {
+      "entropy": 0.09575109251520851,
+      "epoch": 9.012070006035003,
+      "grad_norm": 0.18088576197624207,
+      "learning_rate": 1.1130933515165429e-05,
+      "loss": 0.07295305132865906,
+      "mean_token_accuracy": 0.9739722857227573,
+      "num_tokens": 8661177.0,
+      "step": 3740
+    },
+    {
+      "epoch": 9.012070006035003,
+      "eval_entropy": 0.23728882387447892,
+      "eval_loss": 1.0134038925170898,
+      "eval_mean_token_accuracy": 0.8286508951294288,
+      "eval_num_tokens": 8661177.0,
+      "eval_runtime": 88.6434,
+      "eval_samples_per_second": 16.019,
+      "eval_steps_per_second": 2.008,
+      "step": 3740
+    },
+    {
+      "entropy": 0.08883201805874705,
+      "epoch": 9.060350030175016,
+      "grad_norm": 0.23120898008346558,
+      "learning_rate": 1.008353806970132e-05,
+      "loss": 0.06326142549514771,
+      "mean_token_accuracy": 0.9780020996928215,
+      "num_tokens": 8707602.0,
+      "step": 3760
+    },
+    {
+      "epoch": 9.060350030175016,
+      "eval_entropy": 0.23416905331142832,
+      "eval_loss": 1.0284100770950317,
+      "eval_mean_token_accuracy": 0.8284083213029283,
+      "eval_num_tokens": 8707602.0,
+      "eval_runtime": 88.2933,
+      "eval_samples_per_second": 16.083,
+      "eval_steps_per_second": 2.016,
+      "step": 3760
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.6810055792911974e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3780/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3780/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3780/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3780/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3780/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3780/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3780/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3780/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3780/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3780/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3780/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3780/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3780/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3780/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..33ceb284384171e1992955899b3b8adc27fb3dc1
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3780/trainer_state.json
@@ -0,0 +1,4003 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 9.108630054315027,
+  "eval_steps": 20,
+  "global_step": 3780,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    },
+    {
+      "entropy": 0.1326121488586068,
+      "epoch": 7.181050090525045,
+      "grad_norm": 0.4843117296695709,
+      "learning_rate": 8.411870386507724e-05,
+      "loss": 0.09537842869758606,
+      "mean_token_accuracy": 0.9645151920616627,
+      "num_tokens": 6898444.0,
+      "step": 2980
+    },
+    {
+      "epoch": 7.181050090525045,
+      "eval_entropy": 0.27999477783280813,
+      "eval_loss": 0.9031037092208862,
+      "eval_mean_token_accuracy": 0.8261882204018282,
+      "eval_num_tokens": 6898444.0,
+      "eval_runtime": 88.5602,
+      "eval_samples_per_second": 16.034,
+      "eval_steps_per_second": 2.01,
+      "step": 2980
+    },
+    {
+      "entropy": 0.1156751093454659,
+      "epoch": 7.229330114665057,
+      "grad_norm": 0.5297922492027283,
+      "learning_rate": 8.149629351430326e-05,
+      "loss": 0.09203402996063233,
+      "mean_token_accuracy": 0.9689733698964119,
+      "num_tokens": 6947106.0,
+      "step": 3000
+    },
+    {
+      "epoch": 7.229330114665057,
+      "eval_entropy": 0.2740225845508361,
+      "eval_loss": 0.8876440525054932,
+      "eval_mean_token_accuracy": 0.8265116857678703,
+      "eval_num_tokens": 6947106.0,
+      "eval_runtime": 87.7971,
+      "eval_samples_per_second": 16.174,
+      "eval_steps_per_second": 2.027,
+      "step": 3000
+    },
+    {
+      "entropy": 0.12116114404052496,
+      "epoch": 7.27761013880507,
+      "grad_norm": 0.5467398166656494,
+      "learning_rate": 7.890406302404618e-05,
+      "loss": 0.0933056652545929,
+      "mean_token_accuracy": 0.9676455594599247,
+      "num_tokens": 6992784.0,
+      "step": 3020
+    },
+    {
+      "epoch": 7.27761013880507,
+      "eval_entropy": 0.27238342793804876,
+      "eval_loss": 0.9166117310523987,
+      "eval_mean_token_accuracy": 0.825499015242866,
+      "eval_num_tokens": 6992784.0,
+      "eval_runtime": 88.3276,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 3020
+    },
+    {
+      "entropy": 0.11715119928121567,
+      "epoch": 7.325890162945082,
+      "grad_norm": 0.4976702332496643,
+      "learning_rate": 7.634274596487176e-05,
+      "loss": 0.09449659585952759,
+      "mean_token_accuracy": 0.966506478190422,
+      "num_tokens": 7040482.0,
+      "step": 3040
+    },
+    {
+      "epoch": 7.325890162945082,
+      "eval_entropy": 0.2788975556077582,
+      "eval_loss": 0.8833065629005432,
+      "eval_mean_token_accuracy": 0.8269920596915684,
+      "eval_num_tokens": 7040482.0,
+      "eval_runtime": 88.9016,
+      "eval_samples_per_second": 15.973,
+      "eval_steps_per_second": 2.002,
+      "step": 3040
+    },
+    {
+      "entropy": 0.12565911412239075,
+      "epoch": 7.3741701870850935,
+      "grad_norm": 0.5728679895401001,
+      "learning_rate": 7.381306715921025e-05,
+      "loss": 0.09626244902610778,
+      "mean_token_accuracy": 0.9661444149911403,
+      "num_tokens": 7086261.0,
+      "step": 3060
+    },
+    {
+      "epoch": 7.3741701870850935,
+      "eval_entropy": 0.2762492011939542,
+      "eval_loss": 0.8961160778999329,
+      "eval_mean_token_accuracy": 0.826369910092836,
+      "eval_num_tokens": 7086261.0,
+      "eval_runtime": 88.5204,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 3060
+    },
+    {
+      "entropy": 0.12258567921817302,
+      "epoch": 7.422450211225105,
+      "grad_norm": 0.468985915184021,
+      "learning_rate": 7.131574247624031e-05,
+      "loss": 0.09120266437530518,
+      "mean_token_accuracy": 0.9674938447773457,
+      "num_tokens": 7133247.0,
+      "step": 3080
+    },
+    {
+      "epoch": 7.422450211225105,
+      "eval_entropy": 0.2688512086366,
+      "eval_loss": 0.9048654437065125,
+      "eval_mean_token_accuracy": 0.8265582302982888,
+      "eval_num_tokens": 7133247.0,
+      "eval_runtime": 88.164,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 3080
+    },
+    {
+      "entropy": 0.12070536343380808,
+      "epoch": 7.470730235365117,
+      "grad_norm": 0.5310841798782349,
+      "learning_rate": 6.885147862930688e-05,
+      "loss": 0.09386247992515565,
+      "mean_token_accuracy": 0.96664222702384,
+      "num_tokens": 7177810.0,
+      "step": 3100
+    },
+    {
+      "epoch": 7.470730235365117,
+      "eval_entropy": 0.26289632874593305,
+      "eval_loss": 0.902837336063385,
+      "eval_mean_token_accuracy": 0.828052184554968,
+      "eval_num_tokens": 7177810.0,
+      "eval_runtime": 88.1201,
+      "eval_samples_per_second": 16.114,
+      "eval_steps_per_second": 2.02,
+      "step": 3100
+    },
+    {
+      "entropy": 0.11322665181942285,
+      "epoch": 7.51901025950513,
+      "grad_norm": 0.6846067905426025,
+      "learning_rate": 6.642097297592908e-05,
+      "loss": 0.08772538304328918,
+      "mean_token_accuracy": 0.9691362135112286,
+      "num_tokens": 7225302.0,
+      "step": 3120
+    },
+    {
+      "epoch": 7.51901025950513,
+      "eval_entropy": 0.2670605613776807,
+      "eval_loss": 0.9130566120147705,
+      "eval_mean_token_accuracy": 0.826366903741708,
+      "eval_num_tokens": 7225302.0,
+      "eval_runtime": 88.6035,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3120
+    },
+    {
+      "entropy": 0.10985926580615342,
+      "epoch": 7.567290283645142,
+      "grad_norm": 0.4799748361110687,
+      "learning_rate": 6.402491332045683e-05,
+      "loss": 0.08672296404838561,
+      "mean_token_accuracy": 0.9698824048042297,
+      "num_tokens": 7274881.0,
+      "step": 3140
+    },
+    {
+      "epoch": 7.567290283645142,
+      "eval_entropy": 0.26657010866015146,
+      "eval_loss": 0.9056380987167358,
+      "eval_mean_token_accuracy": 0.8272326461384806,
+      "eval_num_tokens": 7274881.0,
+      "eval_runtime": 88.2967,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3140
+    },
+    {
+      "entropy": 0.11890473761595785,
+      "epoch": 7.615570307785154,
+      "grad_norm": 0.4081912934780121,
+      "learning_rate": 6.166397771942987e-05,
+      "loss": 0.09396827220916748,
+      "mean_token_accuracy": 0.967512384802103,
+      "num_tokens": 7321163.0,
+      "step": 3160
+    },
+    {
+      "epoch": 7.615570307785154,
+      "eval_entropy": 0.26364776575833226,
+      "eval_loss": 0.9204733967781067,
+      "eval_mean_token_accuracy": 0.827419729714983,
+      "eval_num_tokens": 7321163.0,
+      "eval_runtime": 88.3441,
+      "eval_samples_per_second": 16.074,
+      "eval_steps_per_second": 2.015,
+      "step": 3160
+    },
+    {
+      "entropy": 0.12209134344011545,
+      "epoch": 7.663850331925166,
+      "grad_norm": 0.5159774422645569,
+      "learning_rate": 5.93388342896962e-05,
+      "loss": 0.09292978644371033,
+      "mean_token_accuracy": 0.9658421002328396,
+      "num_tokens": 7366217.0,
+      "step": 3180
+    },
+    {
+      "epoch": 7.663850331925166,
+      "eval_entropy": 0.2714110068725736,
+      "eval_loss": 0.9143092036247253,
+      "eval_mean_token_accuracy": 0.8262251099843657,
+      "eval_num_tokens": 7366217.0,
+      "eval_runtime": 87.9635,
+      "eval_samples_per_second": 16.143,
+      "eval_steps_per_second": 2.024,
+      "step": 3180
+    },
+    {
+      "entropy": 0.11955609126016498,
+      "epoch": 7.712130356065178,
+      "grad_norm": 0.7005976438522339,
+      "learning_rate": 5.705014101934252e-05,
+      "loss": 0.09319776892662049,
+      "mean_token_accuracy": 0.9674513377249241,
+      "num_tokens": 7410887.0,
+      "step": 3200
+    },
+    {
+      "epoch": 7.712130356065178,
+      "eval_entropy": 0.2626291513442993,
+      "eval_loss": 0.91884446144104,
+      "eval_mean_token_accuracy": 0.827563955207889,
+      "eval_num_tokens": 7410887.0,
+      "eval_runtime": 87.8849,
+      "eval_samples_per_second": 16.157,
+      "eval_steps_per_second": 2.025,
+      "step": 3200
+    },
+    {
+      "entropy": 0.11767532899975777,
+      "epoch": 7.76041038020519,
+      "grad_norm": 0.4675362706184387,
+      "learning_rate": 5.479854558149187e-05,
+      "loss": 0.09038087725639343,
+      "mean_token_accuracy": 0.9669680051505566,
+      "num_tokens": 7457578.0,
+      "step": 3220
+    },
+    {
+      "epoch": 7.76041038020519,
+      "eval_entropy": 0.26581037203582486,
+      "eval_loss": 0.9146883487701416,
+      "eval_mean_token_accuracy": 0.8270345812433222,
+      "eval_num_tokens": 7457578.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 3220
+    },
+    {
+      "entropy": 0.11383061073720455,
+      "epoch": 7.808690404345202,
+      "grad_norm": 0.541247546672821,
+      "learning_rate": 5.258468515101918e-05,
+      "loss": 0.08942405581474304,
+      "mean_token_accuracy": 0.968398705124855,
+      "num_tokens": 7505761.0,
+      "step": 3240
+    },
+    {
+      "epoch": 7.808690404345202,
+      "eval_entropy": 0.26687998960861997,
+      "eval_loss": 0.9037085175514221,
+      "eval_mean_token_accuracy": 0.8274977863504646,
+      "eval_num_tokens": 7505761.0,
+      "eval_runtime": 88.5163,
+      "eval_samples_per_second": 16.042,
+      "eval_steps_per_second": 2.011,
+      "step": 3240
+    },
+    {
+      "entropy": 0.12004058873280883,
+      "epoch": 7.856970428485214,
+      "grad_norm": 0.5289552211761475,
+      "learning_rate": 5.040918622423899e-05,
+      "loss": 0.09064805507659912,
+      "mean_token_accuracy": 0.9673330299556255,
+      "num_tokens": 7552557.0,
+      "step": 3260
+    },
+    {
+      "epoch": 7.856970428485214,
+      "eval_entropy": 0.26288379251622085,
+      "eval_loss": 0.9146178364753723,
+      "eval_mean_token_accuracy": 0.8276662448149049,
+      "eval_num_tokens": 7552557.0,
+      "eval_runtime": 88.0723,
+      "eval_samples_per_second": 16.123,
+      "eval_steps_per_second": 2.021,
+      "step": 3260
+    },
+    {
+      "entropy": 0.1180689456872642,
+      "epoch": 7.905250452625227,
+      "grad_norm": 0.4754861891269684,
+      "learning_rate": 4.827266444161377e-05,
+      "loss": 0.09402655363082886,
+      "mean_token_accuracy": 0.9663361787796021,
+      "num_tokens": 7596998.0,
+      "step": 3280
+    },
+    {
+      "epoch": 7.905250452625227,
+      "eval_entropy": 0.261690996922134,
+      "eval_loss": 0.9120186567306519,
+      "eval_mean_token_accuracy": 0.8278995015647974,
+      "eval_num_tokens": 7596998.0,
+      "eval_runtime": 88.6573,
+      "eval_samples_per_second": 16.017,
+      "eval_steps_per_second": 2.008,
+      "step": 3280
+    },
+    {
+      "entropy": 0.1162164019420743,
+      "epoch": 7.953530476765239,
+      "grad_norm": 0.6203398108482361,
+      "learning_rate": 4.61757244135355e-05,
+      "loss": 0.08832768201828003,
+      "mean_token_accuracy": 0.9694220684468746,
+      "num_tokens": 7643288.0,
+      "step": 3300
+    },
+    {
+      "epoch": 7.953530476765239,
+      "eval_entropy": 0.26158678129817664,
+      "eval_loss": 0.9301361441612244,
+      "eval_mean_token_accuracy": 0.8269603885291668,
+      "eval_num_tokens": 7643288.0,
+      "eval_runtime": 88.7064,
+      "eval_samples_per_second": 16.008,
+      "eval_steps_per_second": 2.007,
+      "step": 3300
+    },
+    {
+      "entropy": 0.10877202204489089,
+      "epoch": 8.0,
+      "grad_norm": 3.058673143386841,
+      "learning_rate": 4.411895954922763e-05,
+      "loss": 0.09159615635871887,
+      "mean_token_accuracy": 0.9686878834451947,
+      "num_tokens": 7688608.0,
+      "step": 3320
+    },
+    {
+      "epoch": 8.0,
+      "eval_entropy": 0.2590438057197614,
+      "eval_loss": 0.9262065887451172,
+      "eval_mean_token_accuracy": 0.8277404546067956,
+      "eval_num_tokens": 7688608.0,
+      "eval_runtime": 87.7557,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 3320
+    },
+    {
+      "entropy": 0.09969092034734786,
+      "epoch": 8.048280024140013,
+      "grad_norm": 0.2561911642551422,
+      "learning_rate": 4.2102951888817644e-05,
+      "loss": 0.07228735685348511,
+      "mean_token_accuracy": 0.9743556551635265,
+      "num_tokens": 7732856.0,
+      "step": 3340
+    },
+    {
+      "epoch": 8.048280024140013,
+      "eval_entropy": 0.2462534454933713,
+      "eval_loss": 0.9755983352661133,
+      "eval_mean_token_accuracy": 0.827510209565752,
+      "eval_num_tokens": 7732856.0,
+      "eval_runtime": 88.0297,
+      "eval_samples_per_second": 16.131,
+      "eval_steps_per_second": 2.022,
+      "step": 3340
+    },
+    {
+      "entropy": 0.09477624730207027,
+      "epoch": 8.096560048280024,
+      "grad_norm": 0.3670356571674347,
+      "learning_rate": 4.012827193862633e-05,
+      "loss": 0.0686036467552185,
+      "mean_token_accuracy": 0.9743670664727688,
+      "num_tokens": 7778585.0,
+      "step": 3360
+    },
+    {
+      "epoch": 8.096560048280024,
+      "eval_entropy": 0.2440882951356052,
+      "eval_loss": 1.0118423700332642,
+      "eval_mean_token_accuracy": 0.8271241070849172,
+      "eval_num_tokens": 7778585.0,
+      "eval_runtime": 88.1364,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 3360
+    },
+    {
+      "entropy": 0.09417980071157217,
+      "epoch": 8.144840072420036,
+      "grad_norm": 0.5785111784934998,
+      "learning_rate": 3.819547850972133e-05,
+      "loss": 0.07100698947906495,
+      "mean_token_accuracy": 0.9737804099917412,
+      "num_tokens": 7824287.0,
+      "step": 3380
+    },
+    {
+      "epoch": 8.144840072420036,
+      "eval_entropy": 0.24603172860453637,
+      "eval_loss": 1.0021991729736328,
+      "eval_mean_token_accuracy": 0.827162122793412,
+      "eval_num_tokens": 7824287.0,
+      "eval_runtime": 88.4534,
+      "eval_samples_per_second": 16.054,
+      "eval_steps_per_second": 2.012,
+      "step": 3380
+    },
+    {
+      "entropy": 0.087222247896716,
+      "epoch": 8.193120096560047,
+      "grad_norm": 0.44848504662513733,
+      "learning_rate": 3.6305118559780393e-05,
+      "loss": 0.06325234174728393,
+      "mean_token_accuracy": 0.9769335232675076,
+      "num_tokens": 7874784.0,
+      "step": 3400
+    },
+    {
+      "epoch": 8.193120096560047,
+      "eval_entropy": 0.24544863905129807,
+      "eval_loss": 1.005008339881897,
+      "eval_mean_token_accuracy": 0.8273035240976998,
+      "eval_num_tokens": 7874784.0,
+      "eval_runtime": 88.6737,
+      "eval_samples_per_second": 16.014,
+      "eval_steps_per_second": 2.007,
+      "step": 3400
+    },
+    {
+      "entropy": 0.09037275062873959,
+      "epoch": 8.24140012070006,
+      "grad_norm": 0.23529155552387238,
+      "learning_rate": 3.4457727038308435e-05,
+      "loss": 0.06589569449424744,
+      "mean_token_accuracy": 0.9751470290124417,
+      "num_tokens": 7925240.0,
+      "step": 3420
+    },
+    {
+      "epoch": 8.24140012070006,
+      "eval_entropy": 0.24203423494368456,
+      "eval_loss": 1.0052824020385742,
+      "eval_mean_token_accuracy": 0.8275575503874361,
+      "eval_num_tokens": 7925240.0,
+      "eval_runtime": 88.3486,
+      "eval_samples_per_second": 16.073,
+      "eval_steps_per_second": 2.015,
+      "step": 3420
+    },
+    {
+      "entropy": 0.10373208015225828,
+      "epoch": 8.289680144840073,
+      "grad_norm": 0.24796408414840698,
+      "learning_rate": 3.2653826735253594e-05,
+      "loss": 0.07714771628379821,
+      "mean_token_accuracy": 0.9722891598939896,
+      "num_tokens": 7966685.0,
+      "step": 3440
+    },
+    {
+      "epoch": 8.289680144840073,
+      "eval_entropy": 0.23461157050025597,
+      "eval_loss": 1.012108325958252,
+      "eval_mean_token_accuracy": 0.82825900564033,
+      "eval_num_tokens": 7966685.0,
+      "eval_runtime": 88.1018,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 3440
+    },
+    {
+      "entropy": 0.08967954274266958,
+      "epoch": 8.337960168980084,
+      "grad_norm": 0.5314674973487854,
+      "learning_rate": 3.08939281330635e-05,
+      "loss": 0.06746266484260559,
+      "mean_token_accuracy": 0.9753133699297905,
+      "num_tokens": 8013656.0,
+      "step": 3460
+    },
+    {
+      "epoch": 8.337960168980084,
+      "eval_entropy": 0.23781998589467468,
+      "eval_loss": 1.0119913816452026,
+      "eval_mean_token_accuracy": 0.8273287157664139,
+      "eval_num_tokens": 8013656.0,
+      "eval_runtime": 88.6038,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3460
+    },
+    {
+      "entropy": 0.09280927330255509,
+      "epoch": 8.386240193120097,
+      "grad_norm": 0.30658936500549316,
+      "learning_rate": 2.9178529262224946e-05,
+      "loss": 0.06936421990394592,
+      "mean_token_accuracy": 0.9742105580866337,
+      "num_tokens": 8061108.0,
+      "step": 3480
+    },
+    {
+      "epoch": 8.386240193120097,
+      "eval_entropy": 0.23672426734747512,
+      "eval_loss": 1.0162041187286377,
+      "eval_mean_token_accuracy": 0.8278394281194451,
+      "eval_num_tokens": 8061108.0,
+      "eval_runtime": 88.6015,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 3480
+    },
+    {
+      "entropy": 0.09195122267119586,
+      "epoch": 8.43452021726011,
+      "grad_norm": 0.24512454867362976,
+      "learning_rate": 2.7508115560327256e-05,
+      "loss": 0.06996064782142639,
+      "mean_token_accuracy": 0.974104854464531,
+      "num_tokens": 8106924.0,
+      "step": 3500
+    },
+    {
+      "epoch": 8.43452021726011,
+      "eval_entropy": 0.23758245828781235,
+      "eval_loss": 1.0157747268676758,
+      "eval_mean_token_accuracy": 0.8278501579600773,
+      "eval_num_tokens": 8106924.0,
+      "eval_runtime": 88.5678,
+      "eval_samples_per_second": 16.033,
+      "eval_steps_per_second": 2.01,
+      "step": 3500
+    },
+    {
+      "entropy": 0.09518477590754629,
+      "epoch": 8.48280024140012,
+      "grad_norm": 0.2373228371143341,
+      "learning_rate": 2.5883159734688868e-05,
+      "loss": 0.06832797527313232,
+      "mean_token_accuracy": 0.9743806682527065,
+      "num_tokens": 8153070.0,
+      "step": 3520
+    },
+    {
+      "epoch": 8.48280024140012,
+      "eval_entropy": 0.2379308074210467,
+      "eval_loss": 1.0180530548095703,
+      "eval_mean_token_accuracy": 0.8282477052024241,
+      "eval_num_tokens": 8153070.0,
+      "eval_runtime": 88.4651,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 3520
+    },
+    {
+      "entropy": 0.0906269280705601,
+      "epoch": 8.531080265540133,
+      "grad_norm": 0.37602657079696655,
+      "learning_rate": 2.4304121628587173e-05,
+      "loss": 0.06921588778495788,
+      "mean_token_accuracy": 0.9745571479201317,
+      "num_tokens": 8200066.0,
+      "step": 3540
+    },
+    {
+      "epoch": 8.531080265540133,
+      "eval_entropy": 0.2357490769263064,
+      "eval_loss": 1.017139196395874,
+      "eval_mean_token_accuracy": 0.828713178299786,
+      "eval_num_tokens": 8200066.0,
+      "eval_runtime": 88.6424,
+      "eval_samples_per_second": 16.019,
+      "eval_steps_per_second": 2.008,
+      "step": 3540
+    },
+    {
+      "entropy": 0.09273334196768701,
+      "epoch": 8.579360289680144,
+      "grad_norm": 0.2978838086128235,
+      "learning_rate": 2.277144809112794e-05,
+      "loss": 0.06970887780189514,
+      "mean_token_accuracy": 0.9751374177634716,
+      "num_tokens": 8246157.0,
+      "step": 3560
+    },
+    {
+      "epoch": 8.579360289680144,
+      "eval_entropy": 0.23590933138065123,
+      "eval_loss": 1.019231915473938,
+      "eval_mean_token_accuracy": 0.8279163412163767,
+      "eval_num_tokens": 8246157.0,
+      "eval_runtime": 88.2691,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 3560
+    },
+    {
+      "entropy": 0.09854201921261847,
+      "epoch": 8.627640313820157,
+      "grad_norm": 0.5381243228912354,
+      "learning_rate": 2.1285572850792413e-05,
+      "loss": 0.07123501300811767,
+      "mean_token_accuracy": 0.9723988175392151,
+      "num_tokens": 8291752.0,
+      "step": 3580
+    },
+    {
+      "epoch": 8.627640313820157,
+      "eval_entropy": 0.23592024949494372,
+      "eval_loss": 1.020250678062439,
+      "eval_mean_token_accuracy": 0.8282466226079491,
+      "eval_num_tokens": 8291752.0,
+      "eval_runtime": 87.8939,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 3580
+    },
+    {
+      "entropy": 0.09092000788077712,
+      "epoch": 8.67592033796017,
+      "grad_norm": 0.29089561104774475,
+      "learning_rate": 1.9846916392697493e-05,
+      "loss": 0.06982720494270325,
+      "mean_token_accuracy": 0.9749995484948158,
+      "num_tokens": 8337651.0,
+      "step": 3600
+    },
+    {
+      "epoch": 8.67592033796017,
+      "eval_entropy": 0.23658913382318583,
+      "eval_loss": 1.019862413406372,
+      "eval_mean_token_accuracy": 0.8280127842104836,
+      "eval_num_tokens": 8337651.0,
+      "eval_runtime": 88.2855,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 3600
+    },
+    {
+      "entropy": 0.08708424470387399,
+      "epoch": 8.72420036210018,
+      "grad_norm": 0.29904839396476746,
+      "learning_rate": 1.8455885839603106e-05,
+      "loss": 0.06801370978355407,
+      "mean_token_accuracy": 0.9755801670253277,
+      "num_tokens": 8385330.0,
+      "step": 3620
+    },
+    {
+      "epoch": 8.72420036210018,
+      "eval_entropy": 0.23627035771862845,
+      "eval_loss": 1.0204569101333618,
+      "eval_mean_token_accuracy": 0.8279967894045155,
+      "eval_num_tokens": 8385330.0,
+      "eval_runtime": 88.1753,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 3620
+    },
+    {
+      "entropy": 0.0916255566291511,
+      "epoch": 8.772480386240193,
+      "grad_norm": 0.20571033656597137,
+      "learning_rate": 1.7112874836701502e-05,
+      "loss": 0.0692249894142151,
+      "mean_token_accuracy": 0.9745148062705994,
+      "num_tokens": 8432134.0,
+      "step": 3640
+    },
+    {
+      "epoch": 8.772480386240193,
+      "eval_entropy": 0.2340821772813797,
+      "eval_loss": 1.0212821960449219,
+      "eval_mean_token_accuracy": 0.828373622358515,
+      "eval_num_tokens": 8432134.0,
+      "eval_runtime": 87.2518,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 3640
+    },
+    {
+      "entropy": 0.0926868049427867,
+      "epoch": 8.820760410380204,
+      "grad_norm": 0.2618265450000763,
+      "learning_rate": 1.5818263440220203e-05,
+      "loss": 0.07338647246360779,
+      "mean_token_accuracy": 0.9724708944559097,
+      "num_tokens": 8476758.0,
+      "step": 3660
+    },
+    {
+      "epoch": 8.820760410380204,
+      "eval_entropy": 0.23637074575330433,
+      "eval_loss": 1.0186725854873657,
+      "eval_mean_token_accuracy": 0.8277707605549459,
+      "eval_num_tokens": 8476758.0,
+      "eval_runtime": 88.4695,
+      "eval_samples_per_second": 16.051,
+      "eval_steps_per_second": 2.012,
+      "step": 3660
+    },
+    {
+      "entropy": 0.09175782897509635,
+      "epoch": 8.869040434520217,
+      "grad_norm": 0.2889852821826935,
+      "learning_rate": 1.4572418009870604e-05,
+      "loss": 0.07066041231155396,
+      "mean_token_accuracy": 0.9725673109292984,
+      "num_tokens": 8523294.0,
+      "step": 3680
+    },
+    {
+      "epoch": 8.869040434520217,
+      "eval_entropy": 0.23945164404223473,
+      "eval_loss": 1.0145231485366821,
+      "eval_mean_token_accuracy": 0.8281202523895864,
+      "eval_num_tokens": 8523294.0,
+      "eval_runtime": 88.368,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 3680
+    },
+    {
+      "entropy": 0.09265611860901117,
+      "epoch": 8.91732045866023,
+      "grad_norm": 0.41092950105667114,
+      "learning_rate": 1.3375691105172625e-05,
+      "loss": 0.06975027322769164,
+      "mean_token_accuracy": 0.974636297672987,
+      "num_tokens": 8570358.0,
+      "step": 3700
+    },
+    {
+      "epoch": 8.91732045866023,
+      "eval_entropy": 0.239588919901446,
+      "eval_loss": 1.014059066772461,
+      "eval_mean_token_accuracy": 0.828028041660116,
+      "eval_num_tokens": 8570358.0,
+      "eval_runtime": 88.1039,
+      "eval_samples_per_second": 16.117,
+      "eval_steps_per_second": 2.02,
+      "step": 3700
+    },
+    {
+      "entropy": 0.0896465003490448,
+      "epoch": 8.965600482800241,
+      "grad_norm": 0.24000254273414612,
+      "learning_rate": 1.2228421385684305e-05,
+      "loss": 0.06743953824043274,
+      "mean_token_accuracy": 0.974006575345993,
+      "num_tokens": 8618246.0,
+      "step": 3720
+    },
+    {
+      "epoch": 8.965600482800241,
+      "eval_entropy": 0.2390461369846644,
+      "eval_loss": 1.011997938156128,
+      "eval_mean_token_accuracy": 0.8284533211354459,
+      "eval_num_tokens": 8618246.0,
+      "eval_runtime": 88.6202,
+      "eval_samples_per_second": 16.023,
+      "eval_steps_per_second": 2.009,
+      "step": 3720
+    },
+    {
+      "entropy": 0.09575109251520851,
+      "epoch": 9.012070006035003,
+      "grad_norm": 0.18088576197624207,
+      "learning_rate": 1.1130933515165429e-05,
+      "loss": 0.07295305132865906,
+      "mean_token_accuracy": 0.9739722857227573,
+      "num_tokens": 8661177.0,
+      "step": 3740
+    },
+    {
+      "epoch": 9.012070006035003,
+      "eval_entropy": 0.23728882387447892,
+      "eval_loss": 1.0134038925170898,
+      "eval_mean_token_accuracy": 0.8286508951294288,
+      "eval_num_tokens": 8661177.0,
+      "eval_runtime": 88.6434,
+      "eval_samples_per_second": 16.019,
+      "eval_steps_per_second": 2.008,
+      "step": 3740
+    },
+    {
+      "entropy": 0.08883201805874705,
+      "epoch": 9.060350030175016,
+      "grad_norm": 0.23120898008346558,
+      "learning_rate": 1.008353806970132e-05,
+      "loss": 0.06326142549514771,
+      "mean_token_accuracy": 0.9780020996928215,
+      "num_tokens": 8707602.0,
+      "step": 3760
+    },
+    {
+      "epoch": 9.060350030175016,
+      "eval_entropy": 0.23416905331142832,
+      "eval_loss": 1.0284100770950317,
+      "eval_mean_token_accuracy": 0.8284083213029283,
+      "eval_num_tokens": 8707602.0,
+      "eval_runtime": 88.2933,
+      "eval_samples_per_second": 16.083,
+      "eval_steps_per_second": 2.016,
+      "step": 3760
+    },
+    {
+      "entropy": 0.08861468276008963,
+      "epoch": 9.108630054315027,
+      "grad_norm": 0.17277656495571136,
+      "learning_rate": 9.086531449813489e-06,
+      "loss": 0.06727416515350342,
+      "mean_token_accuracy": 0.9760390736162663,
+      "num_tokens": 8751287.0,
+      "step": 3780
+    },
+    {
+      "epoch": 9.108630054315027,
+      "eval_entropy": 0.23370317386442355,
+      "eval_loss": 1.0381091833114624,
+      "eval_mean_token_accuracy": 0.8282668178670862,
+      "eval_num_tokens": 8751287.0,
+      "eval_runtime": 88.5738,
+      "eval_samples_per_second": 16.032,
+      "eval_steps_per_second": 2.01,
+      "step": 3780
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.6994065369944064e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-380/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-380/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-380/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-380/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-380/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-380/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-380/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-380/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-380/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-380/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-380/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-380/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-380/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-380/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..96c5444d306695bfaadc8bc0a7e5eaa3f67510b7
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-380/trainer_state.json
@@ -0,0 +1,433 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.9173204586602294,
+  "eval_steps": 20,
+  "global_step": 380,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.735218097190502e+16,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3800/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3800/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3800/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3800/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3800/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3800/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3800/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3800/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3800/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3800/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3800/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3800/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3800/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3800/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..718c50884ac8267ab29bedb267772c7c6c8006e7
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3800/trainer_state.json
@@ -0,0 +1,4024 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 9.15691007845504,
+  "eval_steps": 20,
+  "global_step": 3800,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    },
+    {
+      "entropy": 0.1326121488586068,
+      "epoch": 7.181050090525045,
+      "grad_norm": 0.4843117296695709,
+      "learning_rate": 8.411870386507724e-05,
+      "loss": 0.09537842869758606,
+      "mean_token_accuracy": 0.9645151920616627,
+      "num_tokens": 6898444.0,
+      "step": 2980
+    },
+    {
+      "epoch": 7.181050090525045,
+      "eval_entropy": 0.27999477783280813,
+      "eval_loss": 0.9031037092208862,
+      "eval_mean_token_accuracy": 0.8261882204018282,
+      "eval_num_tokens": 6898444.0,
+      "eval_runtime": 88.5602,
+      "eval_samples_per_second": 16.034,
+      "eval_steps_per_second": 2.01,
+      "step": 2980
+    },
+    {
+      "entropy": 0.1156751093454659,
+      "epoch": 7.229330114665057,
+      "grad_norm": 0.5297922492027283,
+      "learning_rate": 8.149629351430326e-05,
+      "loss": 0.09203402996063233,
+      "mean_token_accuracy": 0.9689733698964119,
+      "num_tokens": 6947106.0,
+      "step": 3000
+    },
+    {
+      "epoch": 7.229330114665057,
+      "eval_entropy": 0.2740225845508361,
+      "eval_loss": 0.8876440525054932,
+      "eval_mean_token_accuracy": 0.8265116857678703,
+      "eval_num_tokens": 6947106.0,
+      "eval_runtime": 87.7971,
+      "eval_samples_per_second": 16.174,
+      "eval_steps_per_second": 2.027,
+      "step": 3000
+    },
+    {
+      "entropy": 0.12116114404052496,
+      "epoch": 7.27761013880507,
+      "grad_norm": 0.5467398166656494,
+      "learning_rate": 7.890406302404618e-05,
+      "loss": 0.0933056652545929,
+      "mean_token_accuracy": 0.9676455594599247,
+      "num_tokens": 6992784.0,
+      "step": 3020
+    },
+    {
+      "epoch": 7.27761013880507,
+      "eval_entropy": 0.27238342793804876,
+      "eval_loss": 0.9166117310523987,
+      "eval_mean_token_accuracy": 0.825499015242866,
+      "eval_num_tokens": 6992784.0,
+      "eval_runtime": 88.3276,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 3020
+    },
+    {
+      "entropy": 0.11715119928121567,
+      "epoch": 7.325890162945082,
+      "grad_norm": 0.4976702332496643,
+      "learning_rate": 7.634274596487176e-05,
+      "loss": 0.09449659585952759,
+      "mean_token_accuracy": 0.966506478190422,
+      "num_tokens": 7040482.0,
+      "step": 3040
+    },
+    {
+      "epoch": 7.325890162945082,
+      "eval_entropy": 0.2788975556077582,
+      "eval_loss": 0.8833065629005432,
+      "eval_mean_token_accuracy": 0.8269920596915684,
+      "eval_num_tokens": 7040482.0,
+      "eval_runtime": 88.9016,
+      "eval_samples_per_second": 15.973,
+      "eval_steps_per_second": 2.002,
+      "step": 3040
+    },
+    {
+      "entropy": 0.12565911412239075,
+      "epoch": 7.3741701870850935,
+      "grad_norm": 0.5728679895401001,
+      "learning_rate": 7.381306715921025e-05,
+      "loss": 0.09626244902610778,
+      "mean_token_accuracy": 0.9661444149911403,
+      "num_tokens": 7086261.0,
+      "step": 3060
+    },
+    {
+      "epoch": 7.3741701870850935,
+      "eval_entropy": 0.2762492011939542,
+      "eval_loss": 0.8961160778999329,
+      "eval_mean_token_accuracy": 0.826369910092836,
+      "eval_num_tokens": 7086261.0,
+      "eval_runtime": 88.5204,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 3060
+    },
+    {
+      "entropy": 0.12258567921817302,
+      "epoch": 7.422450211225105,
+      "grad_norm": 0.468985915184021,
+      "learning_rate": 7.131574247624031e-05,
+      "loss": 0.09120266437530518,
+      "mean_token_accuracy": 0.9674938447773457,
+      "num_tokens": 7133247.0,
+      "step": 3080
+    },
+    {
+      "epoch": 7.422450211225105,
+      "eval_entropy": 0.2688512086366,
+      "eval_loss": 0.9048654437065125,
+      "eval_mean_token_accuracy": 0.8265582302982888,
+      "eval_num_tokens": 7133247.0,
+      "eval_runtime": 88.164,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 3080
+    },
+    {
+      "entropy": 0.12070536343380808,
+      "epoch": 7.470730235365117,
+      "grad_norm": 0.5310841798782349,
+      "learning_rate": 6.885147862930688e-05,
+      "loss": 0.09386247992515565,
+      "mean_token_accuracy": 0.96664222702384,
+      "num_tokens": 7177810.0,
+      "step": 3100
+    },
+    {
+      "epoch": 7.470730235365117,
+      "eval_entropy": 0.26289632874593305,
+      "eval_loss": 0.902837336063385,
+      "eval_mean_token_accuracy": 0.828052184554968,
+      "eval_num_tokens": 7177810.0,
+      "eval_runtime": 88.1201,
+      "eval_samples_per_second": 16.114,
+      "eval_steps_per_second": 2.02,
+      "step": 3100
+    },
+    {
+      "entropy": 0.11322665181942285,
+      "epoch": 7.51901025950513,
+      "grad_norm": 0.6846067905426025,
+      "learning_rate": 6.642097297592908e-05,
+      "loss": 0.08772538304328918,
+      "mean_token_accuracy": 0.9691362135112286,
+      "num_tokens": 7225302.0,
+      "step": 3120
+    },
+    {
+      "epoch": 7.51901025950513,
+      "eval_entropy": 0.2670605613776807,
+      "eval_loss": 0.9130566120147705,
+      "eval_mean_token_accuracy": 0.826366903741708,
+      "eval_num_tokens": 7225302.0,
+      "eval_runtime": 88.6035,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3120
+    },
+    {
+      "entropy": 0.10985926580615342,
+      "epoch": 7.567290283645142,
+      "grad_norm": 0.4799748361110687,
+      "learning_rate": 6.402491332045683e-05,
+      "loss": 0.08672296404838561,
+      "mean_token_accuracy": 0.9698824048042297,
+      "num_tokens": 7274881.0,
+      "step": 3140
+    },
+    {
+      "epoch": 7.567290283645142,
+      "eval_entropy": 0.26657010866015146,
+      "eval_loss": 0.9056380987167358,
+      "eval_mean_token_accuracy": 0.8272326461384806,
+      "eval_num_tokens": 7274881.0,
+      "eval_runtime": 88.2967,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3140
+    },
+    {
+      "entropy": 0.11890473761595785,
+      "epoch": 7.615570307785154,
+      "grad_norm": 0.4081912934780121,
+      "learning_rate": 6.166397771942987e-05,
+      "loss": 0.09396827220916748,
+      "mean_token_accuracy": 0.967512384802103,
+      "num_tokens": 7321163.0,
+      "step": 3160
+    },
+    {
+      "epoch": 7.615570307785154,
+      "eval_entropy": 0.26364776575833226,
+      "eval_loss": 0.9204733967781067,
+      "eval_mean_token_accuracy": 0.827419729714983,
+      "eval_num_tokens": 7321163.0,
+      "eval_runtime": 88.3441,
+      "eval_samples_per_second": 16.074,
+      "eval_steps_per_second": 2.015,
+      "step": 3160
+    },
+    {
+      "entropy": 0.12209134344011545,
+      "epoch": 7.663850331925166,
+      "grad_norm": 0.5159774422645569,
+      "learning_rate": 5.93388342896962e-05,
+      "loss": 0.09292978644371033,
+      "mean_token_accuracy": 0.9658421002328396,
+      "num_tokens": 7366217.0,
+      "step": 3180
+    },
+    {
+      "epoch": 7.663850331925166,
+      "eval_entropy": 0.2714110068725736,
+      "eval_loss": 0.9143092036247253,
+      "eval_mean_token_accuracy": 0.8262251099843657,
+      "eval_num_tokens": 7366217.0,
+      "eval_runtime": 87.9635,
+      "eval_samples_per_second": 16.143,
+      "eval_steps_per_second": 2.024,
+      "step": 3180
+    },
+    {
+      "entropy": 0.11955609126016498,
+      "epoch": 7.712130356065178,
+      "grad_norm": 0.7005976438522339,
+      "learning_rate": 5.705014101934252e-05,
+      "loss": 0.09319776892662049,
+      "mean_token_accuracy": 0.9674513377249241,
+      "num_tokens": 7410887.0,
+      "step": 3200
+    },
+    {
+      "epoch": 7.712130356065178,
+      "eval_entropy": 0.2626291513442993,
+      "eval_loss": 0.91884446144104,
+      "eval_mean_token_accuracy": 0.827563955207889,
+      "eval_num_tokens": 7410887.0,
+      "eval_runtime": 87.8849,
+      "eval_samples_per_second": 16.157,
+      "eval_steps_per_second": 2.025,
+      "step": 3200
+    },
+    {
+      "entropy": 0.11767532899975777,
+      "epoch": 7.76041038020519,
+      "grad_norm": 0.4675362706184387,
+      "learning_rate": 5.479854558149187e-05,
+      "loss": 0.09038087725639343,
+      "mean_token_accuracy": 0.9669680051505566,
+      "num_tokens": 7457578.0,
+      "step": 3220
+    },
+    {
+      "epoch": 7.76041038020519,
+      "eval_entropy": 0.26581037203582486,
+      "eval_loss": 0.9146883487701416,
+      "eval_mean_token_accuracy": 0.8270345812433222,
+      "eval_num_tokens": 7457578.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 3220
+    },
+    {
+      "entropy": 0.11383061073720455,
+      "epoch": 7.808690404345202,
+      "grad_norm": 0.541247546672821,
+      "learning_rate": 5.258468515101918e-05,
+      "loss": 0.08942405581474304,
+      "mean_token_accuracy": 0.968398705124855,
+      "num_tokens": 7505761.0,
+      "step": 3240
+    },
+    {
+      "epoch": 7.808690404345202,
+      "eval_entropy": 0.26687998960861997,
+      "eval_loss": 0.9037085175514221,
+      "eval_mean_token_accuracy": 0.8274977863504646,
+      "eval_num_tokens": 7505761.0,
+      "eval_runtime": 88.5163,
+      "eval_samples_per_second": 16.042,
+      "eval_steps_per_second": 2.011,
+      "step": 3240
+    },
+    {
+      "entropy": 0.12004058873280883,
+      "epoch": 7.856970428485214,
+      "grad_norm": 0.5289552211761475,
+      "learning_rate": 5.040918622423899e-05,
+      "loss": 0.09064805507659912,
+      "mean_token_accuracy": 0.9673330299556255,
+      "num_tokens": 7552557.0,
+      "step": 3260
+    },
+    {
+      "epoch": 7.856970428485214,
+      "eval_entropy": 0.26288379251622085,
+      "eval_loss": 0.9146178364753723,
+      "eval_mean_token_accuracy": 0.8276662448149049,
+      "eval_num_tokens": 7552557.0,
+      "eval_runtime": 88.0723,
+      "eval_samples_per_second": 16.123,
+      "eval_steps_per_second": 2.021,
+      "step": 3260
+    },
+    {
+      "entropy": 0.1180689456872642,
+      "epoch": 7.905250452625227,
+      "grad_norm": 0.4754861891269684,
+      "learning_rate": 4.827266444161377e-05,
+      "loss": 0.09402655363082886,
+      "mean_token_accuracy": 0.9663361787796021,
+      "num_tokens": 7596998.0,
+      "step": 3280
+    },
+    {
+      "epoch": 7.905250452625227,
+      "eval_entropy": 0.261690996922134,
+      "eval_loss": 0.9120186567306519,
+      "eval_mean_token_accuracy": 0.8278995015647974,
+      "eval_num_tokens": 7596998.0,
+      "eval_runtime": 88.6573,
+      "eval_samples_per_second": 16.017,
+      "eval_steps_per_second": 2.008,
+      "step": 3280
+    },
+    {
+      "entropy": 0.1162164019420743,
+      "epoch": 7.953530476765239,
+      "grad_norm": 0.6203398108482361,
+      "learning_rate": 4.61757244135355e-05,
+      "loss": 0.08832768201828003,
+      "mean_token_accuracy": 0.9694220684468746,
+      "num_tokens": 7643288.0,
+      "step": 3300
+    },
+    {
+      "epoch": 7.953530476765239,
+      "eval_entropy": 0.26158678129817664,
+      "eval_loss": 0.9301361441612244,
+      "eval_mean_token_accuracy": 0.8269603885291668,
+      "eval_num_tokens": 7643288.0,
+      "eval_runtime": 88.7064,
+      "eval_samples_per_second": 16.008,
+      "eval_steps_per_second": 2.007,
+      "step": 3300
+    },
+    {
+      "entropy": 0.10877202204489089,
+      "epoch": 8.0,
+      "grad_norm": 3.058673143386841,
+      "learning_rate": 4.411895954922763e-05,
+      "loss": 0.09159615635871887,
+      "mean_token_accuracy": 0.9686878834451947,
+      "num_tokens": 7688608.0,
+      "step": 3320
+    },
+    {
+      "epoch": 8.0,
+      "eval_entropy": 0.2590438057197614,
+      "eval_loss": 0.9262065887451172,
+      "eval_mean_token_accuracy": 0.8277404546067956,
+      "eval_num_tokens": 7688608.0,
+      "eval_runtime": 87.7557,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 3320
+    },
+    {
+      "entropy": 0.09969092034734786,
+      "epoch": 8.048280024140013,
+      "grad_norm": 0.2561911642551422,
+      "learning_rate": 4.2102951888817644e-05,
+      "loss": 0.07228735685348511,
+      "mean_token_accuracy": 0.9743556551635265,
+      "num_tokens": 7732856.0,
+      "step": 3340
+    },
+    {
+      "epoch": 8.048280024140013,
+      "eval_entropy": 0.2462534454933713,
+      "eval_loss": 0.9755983352661133,
+      "eval_mean_token_accuracy": 0.827510209565752,
+      "eval_num_tokens": 7732856.0,
+      "eval_runtime": 88.0297,
+      "eval_samples_per_second": 16.131,
+      "eval_steps_per_second": 2.022,
+      "step": 3340
+    },
+    {
+      "entropy": 0.09477624730207027,
+      "epoch": 8.096560048280024,
+      "grad_norm": 0.3670356571674347,
+      "learning_rate": 4.012827193862633e-05,
+      "loss": 0.0686036467552185,
+      "mean_token_accuracy": 0.9743670664727688,
+      "num_tokens": 7778585.0,
+      "step": 3360
+    },
+    {
+      "epoch": 8.096560048280024,
+      "eval_entropy": 0.2440882951356052,
+      "eval_loss": 1.0118423700332642,
+      "eval_mean_token_accuracy": 0.8271241070849172,
+      "eval_num_tokens": 7778585.0,
+      "eval_runtime": 88.1364,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 3360
+    },
+    {
+      "entropy": 0.09417980071157217,
+      "epoch": 8.144840072420036,
+      "grad_norm": 0.5785111784934998,
+      "learning_rate": 3.819547850972133e-05,
+      "loss": 0.07100698947906495,
+      "mean_token_accuracy": 0.9737804099917412,
+      "num_tokens": 7824287.0,
+      "step": 3380
+    },
+    {
+      "epoch": 8.144840072420036,
+      "eval_entropy": 0.24603172860453637,
+      "eval_loss": 1.0021991729736328,
+      "eval_mean_token_accuracy": 0.827162122793412,
+      "eval_num_tokens": 7824287.0,
+      "eval_runtime": 88.4534,
+      "eval_samples_per_second": 16.054,
+      "eval_steps_per_second": 2.012,
+      "step": 3380
+    },
+    {
+      "entropy": 0.087222247896716,
+      "epoch": 8.193120096560047,
+      "grad_norm": 0.44848504662513733,
+      "learning_rate": 3.6305118559780393e-05,
+      "loss": 0.06325234174728393,
+      "mean_token_accuracy": 0.9769335232675076,
+      "num_tokens": 7874784.0,
+      "step": 3400
+    },
+    {
+      "epoch": 8.193120096560047,
+      "eval_entropy": 0.24544863905129807,
+      "eval_loss": 1.005008339881897,
+      "eval_mean_token_accuracy": 0.8273035240976998,
+      "eval_num_tokens": 7874784.0,
+      "eval_runtime": 88.6737,
+      "eval_samples_per_second": 16.014,
+      "eval_steps_per_second": 2.007,
+      "step": 3400
+    },
+    {
+      "entropy": 0.09037275062873959,
+      "epoch": 8.24140012070006,
+      "grad_norm": 0.23529155552387238,
+      "learning_rate": 3.4457727038308435e-05,
+      "loss": 0.06589569449424744,
+      "mean_token_accuracy": 0.9751470290124417,
+      "num_tokens": 7925240.0,
+      "step": 3420
+    },
+    {
+      "epoch": 8.24140012070006,
+      "eval_entropy": 0.24203423494368456,
+      "eval_loss": 1.0052824020385742,
+      "eval_mean_token_accuracy": 0.8275575503874361,
+      "eval_num_tokens": 7925240.0,
+      "eval_runtime": 88.3486,
+      "eval_samples_per_second": 16.073,
+      "eval_steps_per_second": 2.015,
+      "step": 3420
+    },
+    {
+      "entropy": 0.10373208015225828,
+      "epoch": 8.289680144840073,
+      "grad_norm": 0.24796408414840698,
+      "learning_rate": 3.2653826735253594e-05,
+      "loss": 0.07714771628379821,
+      "mean_token_accuracy": 0.9722891598939896,
+      "num_tokens": 7966685.0,
+      "step": 3440
+    },
+    {
+      "epoch": 8.289680144840073,
+      "eval_entropy": 0.23461157050025597,
+      "eval_loss": 1.012108325958252,
+      "eval_mean_token_accuracy": 0.82825900564033,
+      "eval_num_tokens": 7966685.0,
+      "eval_runtime": 88.1018,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 3440
+    },
+    {
+      "entropy": 0.08967954274266958,
+      "epoch": 8.337960168980084,
+      "grad_norm": 0.5314674973487854,
+      "learning_rate": 3.08939281330635e-05,
+      "loss": 0.06746266484260559,
+      "mean_token_accuracy": 0.9753133699297905,
+      "num_tokens": 8013656.0,
+      "step": 3460
+    },
+    {
+      "epoch": 8.337960168980084,
+      "eval_entropy": 0.23781998589467468,
+      "eval_loss": 1.0119913816452026,
+      "eval_mean_token_accuracy": 0.8273287157664139,
+      "eval_num_tokens": 8013656.0,
+      "eval_runtime": 88.6038,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3460
+    },
+    {
+      "entropy": 0.09280927330255509,
+      "epoch": 8.386240193120097,
+      "grad_norm": 0.30658936500549316,
+      "learning_rate": 2.9178529262224946e-05,
+      "loss": 0.06936421990394592,
+      "mean_token_accuracy": 0.9742105580866337,
+      "num_tokens": 8061108.0,
+      "step": 3480
+    },
+    {
+      "epoch": 8.386240193120097,
+      "eval_entropy": 0.23672426734747512,
+      "eval_loss": 1.0162041187286377,
+      "eval_mean_token_accuracy": 0.8278394281194451,
+      "eval_num_tokens": 8061108.0,
+      "eval_runtime": 88.6015,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 3480
+    },
+    {
+      "entropy": 0.09195122267119586,
+      "epoch": 8.43452021726011,
+      "grad_norm": 0.24512454867362976,
+      "learning_rate": 2.7508115560327256e-05,
+      "loss": 0.06996064782142639,
+      "mean_token_accuracy": 0.974104854464531,
+      "num_tokens": 8106924.0,
+      "step": 3500
+    },
+    {
+      "epoch": 8.43452021726011,
+      "eval_entropy": 0.23758245828781235,
+      "eval_loss": 1.0157747268676758,
+      "eval_mean_token_accuracy": 0.8278501579600773,
+      "eval_num_tokens": 8106924.0,
+      "eval_runtime": 88.5678,
+      "eval_samples_per_second": 16.033,
+      "eval_steps_per_second": 2.01,
+      "step": 3500
+    },
+    {
+      "entropy": 0.09518477590754629,
+      "epoch": 8.48280024140012,
+      "grad_norm": 0.2373228371143341,
+      "learning_rate": 2.5883159734688868e-05,
+      "loss": 0.06832797527313232,
+      "mean_token_accuracy": 0.9743806682527065,
+      "num_tokens": 8153070.0,
+      "step": 3520
+    },
+    {
+      "epoch": 8.48280024140012,
+      "eval_entropy": 0.2379308074210467,
+      "eval_loss": 1.0180530548095703,
+      "eval_mean_token_accuracy": 0.8282477052024241,
+      "eval_num_tokens": 8153070.0,
+      "eval_runtime": 88.4651,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 3520
+    },
+    {
+      "entropy": 0.0906269280705601,
+      "epoch": 8.531080265540133,
+      "grad_norm": 0.37602657079696655,
+      "learning_rate": 2.4304121628587173e-05,
+      "loss": 0.06921588778495788,
+      "mean_token_accuracy": 0.9745571479201317,
+      "num_tokens": 8200066.0,
+      "step": 3540
+    },
+    {
+      "epoch": 8.531080265540133,
+      "eval_entropy": 0.2357490769263064,
+      "eval_loss": 1.017139196395874,
+      "eval_mean_token_accuracy": 0.828713178299786,
+      "eval_num_tokens": 8200066.0,
+      "eval_runtime": 88.6424,
+      "eval_samples_per_second": 16.019,
+      "eval_steps_per_second": 2.008,
+      "step": 3540
+    },
+    {
+      "entropy": 0.09273334196768701,
+      "epoch": 8.579360289680144,
+      "grad_norm": 0.2978838086128235,
+      "learning_rate": 2.277144809112794e-05,
+      "loss": 0.06970887780189514,
+      "mean_token_accuracy": 0.9751374177634716,
+      "num_tokens": 8246157.0,
+      "step": 3560
+    },
+    {
+      "epoch": 8.579360289680144,
+      "eval_entropy": 0.23590933138065123,
+      "eval_loss": 1.019231915473938,
+      "eval_mean_token_accuracy": 0.8279163412163767,
+      "eval_num_tokens": 8246157.0,
+      "eval_runtime": 88.2691,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 3560
+    },
+    {
+      "entropy": 0.09854201921261847,
+      "epoch": 8.627640313820157,
+      "grad_norm": 0.5381243228912354,
+      "learning_rate": 2.1285572850792413e-05,
+      "loss": 0.07123501300811767,
+      "mean_token_accuracy": 0.9723988175392151,
+      "num_tokens": 8291752.0,
+      "step": 3580
+    },
+    {
+      "epoch": 8.627640313820157,
+      "eval_entropy": 0.23592024949494372,
+      "eval_loss": 1.020250678062439,
+      "eval_mean_token_accuracy": 0.8282466226079491,
+      "eval_num_tokens": 8291752.0,
+      "eval_runtime": 87.8939,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 3580
+    },
+    {
+      "entropy": 0.09092000788077712,
+      "epoch": 8.67592033796017,
+      "grad_norm": 0.29089561104774475,
+      "learning_rate": 1.9846916392697493e-05,
+      "loss": 0.06982720494270325,
+      "mean_token_accuracy": 0.9749995484948158,
+      "num_tokens": 8337651.0,
+      "step": 3600
+    },
+    {
+      "epoch": 8.67592033796017,
+      "eval_entropy": 0.23658913382318583,
+      "eval_loss": 1.019862413406372,
+      "eval_mean_token_accuracy": 0.8280127842104836,
+      "eval_num_tokens": 8337651.0,
+      "eval_runtime": 88.2855,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 3600
+    },
+    {
+      "entropy": 0.08708424470387399,
+      "epoch": 8.72420036210018,
+      "grad_norm": 0.29904839396476746,
+      "learning_rate": 1.8455885839603106e-05,
+      "loss": 0.06801370978355407,
+      "mean_token_accuracy": 0.9755801670253277,
+      "num_tokens": 8385330.0,
+      "step": 3620
+    },
+    {
+      "epoch": 8.72420036210018,
+      "eval_entropy": 0.23627035771862845,
+      "eval_loss": 1.0204569101333618,
+      "eval_mean_token_accuracy": 0.8279967894045155,
+      "eval_num_tokens": 8385330.0,
+      "eval_runtime": 88.1753,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 3620
+    },
+    {
+      "entropy": 0.0916255566291511,
+      "epoch": 8.772480386240193,
+      "grad_norm": 0.20571033656597137,
+      "learning_rate": 1.7112874836701502e-05,
+      "loss": 0.0692249894142151,
+      "mean_token_accuracy": 0.9745148062705994,
+      "num_tokens": 8432134.0,
+      "step": 3640
+    },
+    {
+      "epoch": 8.772480386240193,
+      "eval_entropy": 0.2340821772813797,
+      "eval_loss": 1.0212821960449219,
+      "eval_mean_token_accuracy": 0.828373622358515,
+      "eval_num_tokens": 8432134.0,
+      "eval_runtime": 87.2518,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 3640
+    },
+    {
+      "entropy": 0.0926868049427867,
+      "epoch": 8.820760410380204,
+      "grad_norm": 0.2618265450000763,
+      "learning_rate": 1.5818263440220203e-05,
+      "loss": 0.07338647246360779,
+      "mean_token_accuracy": 0.9724708944559097,
+      "num_tokens": 8476758.0,
+      "step": 3660
+    },
+    {
+      "epoch": 8.820760410380204,
+      "eval_entropy": 0.23637074575330433,
+      "eval_loss": 1.0186725854873657,
+      "eval_mean_token_accuracy": 0.8277707605549459,
+      "eval_num_tokens": 8476758.0,
+      "eval_runtime": 88.4695,
+      "eval_samples_per_second": 16.051,
+      "eval_steps_per_second": 2.012,
+      "step": 3660
+    },
+    {
+      "entropy": 0.09175782897509635,
+      "epoch": 8.869040434520217,
+      "grad_norm": 0.2889852821826935,
+      "learning_rate": 1.4572418009870604e-05,
+      "loss": 0.07066041231155396,
+      "mean_token_accuracy": 0.9725673109292984,
+      "num_tokens": 8523294.0,
+      "step": 3680
+    },
+    {
+      "epoch": 8.869040434520217,
+      "eval_entropy": 0.23945164404223473,
+      "eval_loss": 1.0145231485366821,
+      "eval_mean_token_accuracy": 0.8281202523895864,
+      "eval_num_tokens": 8523294.0,
+      "eval_runtime": 88.368,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 3680
+    },
+    {
+      "entropy": 0.09265611860901117,
+      "epoch": 8.91732045866023,
+      "grad_norm": 0.41092950105667114,
+      "learning_rate": 1.3375691105172625e-05,
+      "loss": 0.06975027322769164,
+      "mean_token_accuracy": 0.974636297672987,
+      "num_tokens": 8570358.0,
+      "step": 3700
+    },
+    {
+      "epoch": 8.91732045866023,
+      "eval_entropy": 0.239588919901446,
+      "eval_loss": 1.014059066772461,
+      "eval_mean_token_accuracy": 0.828028041660116,
+      "eval_num_tokens": 8570358.0,
+      "eval_runtime": 88.1039,
+      "eval_samples_per_second": 16.117,
+      "eval_steps_per_second": 2.02,
+      "step": 3700
+    },
+    {
+      "entropy": 0.0896465003490448,
+      "epoch": 8.965600482800241,
+      "grad_norm": 0.24000254273414612,
+      "learning_rate": 1.2228421385684305e-05,
+      "loss": 0.06743953824043274,
+      "mean_token_accuracy": 0.974006575345993,
+      "num_tokens": 8618246.0,
+      "step": 3720
+    },
+    {
+      "epoch": 8.965600482800241,
+      "eval_entropy": 0.2390461369846644,
+      "eval_loss": 1.011997938156128,
+      "eval_mean_token_accuracy": 0.8284533211354459,
+      "eval_num_tokens": 8618246.0,
+      "eval_runtime": 88.6202,
+      "eval_samples_per_second": 16.023,
+      "eval_steps_per_second": 2.009,
+      "step": 3720
+    },
+    {
+      "entropy": 0.09575109251520851,
+      "epoch": 9.012070006035003,
+      "grad_norm": 0.18088576197624207,
+      "learning_rate": 1.1130933515165429e-05,
+      "loss": 0.07295305132865906,
+      "mean_token_accuracy": 0.9739722857227573,
+      "num_tokens": 8661177.0,
+      "step": 3740
+    },
+    {
+      "epoch": 9.012070006035003,
+      "eval_entropy": 0.23728882387447892,
+      "eval_loss": 1.0134038925170898,
+      "eval_mean_token_accuracy": 0.8286508951294288,
+      "eval_num_tokens": 8661177.0,
+      "eval_runtime": 88.6434,
+      "eval_samples_per_second": 16.019,
+      "eval_steps_per_second": 2.008,
+      "step": 3740
+    },
+    {
+      "entropy": 0.08883201805874705,
+      "epoch": 9.060350030175016,
+      "grad_norm": 0.23120898008346558,
+      "learning_rate": 1.008353806970132e-05,
+      "loss": 0.06326142549514771,
+      "mean_token_accuracy": 0.9780020996928215,
+      "num_tokens": 8707602.0,
+      "step": 3760
+    },
+    {
+      "epoch": 9.060350030175016,
+      "eval_entropy": 0.23416905331142832,
+      "eval_loss": 1.0284100770950317,
+      "eval_mean_token_accuracy": 0.8284083213029283,
+      "eval_num_tokens": 8707602.0,
+      "eval_runtime": 88.2933,
+      "eval_samples_per_second": 16.083,
+      "eval_steps_per_second": 2.016,
+      "step": 3760
+    },
+    {
+      "entropy": 0.08861468276008963,
+      "epoch": 9.108630054315027,
+      "grad_norm": 0.17277656495571136,
+      "learning_rate": 9.086531449813489e-06,
+      "loss": 0.06727416515350342,
+      "mean_token_accuracy": 0.9760390736162663,
+      "num_tokens": 8751287.0,
+      "step": 3780
+    },
+    {
+      "epoch": 9.108630054315027,
+      "eval_entropy": 0.23370317386442355,
+      "eval_loss": 1.0381091833114624,
+      "eval_mean_token_accuracy": 0.8282668178670862,
+      "eval_num_tokens": 8751287.0,
+      "eval_runtime": 88.5738,
+      "eval_samples_per_second": 16.032,
+      "eval_steps_per_second": 2.01,
+      "step": 3780
+    },
+    {
+      "entropy": 0.09032954541034996,
+      "epoch": 9.15691007845504,
+      "grad_norm": 0.19481085240840912,
+      "learning_rate": 8.140195796581981e-06,
+      "loss": 0.06117786169052124,
+      "mean_token_accuracy": 0.9764454193413258,
+      "num_tokens": 8799982.0,
+      "step": 3800
+    },
+    {
+      "epoch": 9.15691007845504,
+      "eval_entropy": 0.23215936609868254,
+      "eval_loss": 1.0467767715454102,
+      "eval_mean_token_accuracy": 0.828405868806196,
+      "eval_num_tokens": 8799982.0,
+      "eval_runtime": 87.9551,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 3800
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.7204776818191565e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3820/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3820/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3820/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3820/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3820/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3820/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3820/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3820/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3820/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3820/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3820/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3820/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3820/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3820/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..5f76988076a1cc0adc313032629bc17e1e7e2066
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3820/trainer_state.json
@@ -0,0 +1,4045 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 9.20519010259505,
+  "eval_steps": 20,
+  "global_step": 3820,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    },
+    {
+      "entropy": 0.1326121488586068,
+      "epoch": 7.181050090525045,
+      "grad_norm": 0.4843117296695709,
+      "learning_rate": 8.411870386507724e-05,
+      "loss": 0.09537842869758606,
+      "mean_token_accuracy": 0.9645151920616627,
+      "num_tokens": 6898444.0,
+      "step": 2980
+    },
+    {
+      "epoch": 7.181050090525045,
+      "eval_entropy": 0.27999477783280813,
+      "eval_loss": 0.9031037092208862,
+      "eval_mean_token_accuracy": 0.8261882204018282,
+      "eval_num_tokens": 6898444.0,
+      "eval_runtime": 88.5602,
+      "eval_samples_per_second": 16.034,
+      "eval_steps_per_second": 2.01,
+      "step": 2980
+    },
+    {
+      "entropy": 0.1156751093454659,
+      "epoch": 7.229330114665057,
+      "grad_norm": 0.5297922492027283,
+      "learning_rate": 8.149629351430326e-05,
+      "loss": 0.09203402996063233,
+      "mean_token_accuracy": 0.9689733698964119,
+      "num_tokens": 6947106.0,
+      "step": 3000
+    },
+    {
+      "epoch": 7.229330114665057,
+      "eval_entropy": 0.2740225845508361,
+      "eval_loss": 0.8876440525054932,
+      "eval_mean_token_accuracy": 0.8265116857678703,
+      "eval_num_tokens": 6947106.0,
+      "eval_runtime": 87.7971,
+      "eval_samples_per_second": 16.174,
+      "eval_steps_per_second": 2.027,
+      "step": 3000
+    },
+    {
+      "entropy": 0.12116114404052496,
+      "epoch": 7.27761013880507,
+      "grad_norm": 0.5467398166656494,
+      "learning_rate": 7.890406302404618e-05,
+      "loss": 0.0933056652545929,
+      "mean_token_accuracy": 0.9676455594599247,
+      "num_tokens": 6992784.0,
+      "step": 3020
+    },
+    {
+      "epoch": 7.27761013880507,
+      "eval_entropy": 0.27238342793804876,
+      "eval_loss": 0.9166117310523987,
+      "eval_mean_token_accuracy": 0.825499015242866,
+      "eval_num_tokens": 6992784.0,
+      "eval_runtime": 88.3276,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 3020
+    },
+    {
+      "entropy": 0.11715119928121567,
+      "epoch": 7.325890162945082,
+      "grad_norm": 0.4976702332496643,
+      "learning_rate": 7.634274596487176e-05,
+      "loss": 0.09449659585952759,
+      "mean_token_accuracy": 0.966506478190422,
+      "num_tokens": 7040482.0,
+      "step": 3040
+    },
+    {
+      "epoch": 7.325890162945082,
+      "eval_entropy": 0.2788975556077582,
+      "eval_loss": 0.8833065629005432,
+      "eval_mean_token_accuracy": 0.8269920596915684,
+      "eval_num_tokens": 7040482.0,
+      "eval_runtime": 88.9016,
+      "eval_samples_per_second": 15.973,
+      "eval_steps_per_second": 2.002,
+      "step": 3040
+    },
+    {
+      "entropy": 0.12565911412239075,
+      "epoch": 7.3741701870850935,
+      "grad_norm": 0.5728679895401001,
+      "learning_rate": 7.381306715921025e-05,
+      "loss": 0.09626244902610778,
+      "mean_token_accuracy": 0.9661444149911403,
+      "num_tokens": 7086261.0,
+      "step": 3060
+    },
+    {
+      "epoch": 7.3741701870850935,
+      "eval_entropy": 0.2762492011939542,
+      "eval_loss": 0.8961160778999329,
+      "eval_mean_token_accuracy": 0.826369910092836,
+      "eval_num_tokens": 7086261.0,
+      "eval_runtime": 88.5204,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 3060
+    },
+    {
+      "entropy": 0.12258567921817302,
+      "epoch": 7.422450211225105,
+      "grad_norm": 0.468985915184021,
+      "learning_rate": 7.131574247624031e-05,
+      "loss": 0.09120266437530518,
+      "mean_token_accuracy": 0.9674938447773457,
+      "num_tokens": 7133247.0,
+      "step": 3080
+    },
+    {
+      "epoch": 7.422450211225105,
+      "eval_entropy": 0.2688512086366,
+      "eval_loss": 0.9048654437065125,
+      "eval_mean_token_accuracy": 0.8265582302982888,
+      "eval_num_tokens": 7133247.0,
+      "eval_runtime": 88.164,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 3080
+    },
+    {
+      "entropy": 0.12070536343380808,
+      "epoch": 7.470730235365117,
+      "grad_norm": 0.5310841798782349,
+      "learning_rate": 6.885147862930688e-05,
+      "loss": 0.09386247992515565,
+      "mean_token_accuracy": 0.96664222702384,
+      "num_tokens": 7177810.0,
+      "step": 3100
+    },
+    {
+      "epoch": 7.470730235365117,
+      "eval_entropy": 0.26289632874593305,
+      "eval_loss": 0.902837336063385,
+      "eval_mean_token_accuracy": 0.828052184554968,
+      "eval_num_tokens": 7177810.0,
+      "eval_runtime": 88.1201,
+      "eval_samples_per_second": 16.114,
+      "eval_steps_per_second": 2.02,
+      "step": 3100
+    },
+    {
+      "entropy": 0.11322665181942285,
+      "epoch": 7.51901025950513,
+      "grad_norm": 0.6846067905426025,
+      "learning_rate": 6.642097297592908e-05,
+      "loss": 0.08772538304328918,
+      "mean_token_accuracy": 0.9691362135112286,
+      "num_tokens": 7225302.0,
+      "step": 3120
+    },
+    {
+      "epoch": 7.51901025950513,
+      "eval_entropy": 0.2670605613776807,
+      "eval_loss": 0.9130566120147705,
+      "eval_mean_token_accuracy": 0.826366903741708,
+      "eval_num_tokens": 7225302.0,
+      "eval_runtime": 88.6035,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3120
+    },
+    {
+      "entropy": 0.10985926580615342,
+      "epoch": 7.567290283645142,
+      "grad_norm": 0.4799748361110687,
+      "learning_rate": 6.402491332045683e-05,
+      "loss": 0.08672296404838561,
+      "mean_token_accuracy": 0.9698824048042297,
+      "num_tokens": 7274881.0,
+      "step": 3140
+    },
+    {
+      "epoch": 7.567290283645142,
+      "eval_entropy": 0.26657010866015146,
+      "eval_loss": 0.9056380987167358,
+      "eval_mean_token_accuracy": 0.8272326461384806,
+      "eval_num_tokens": 7274881.0,
+      "eval_runtime": 88.2967,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3140
+    },
+    {
+      "entropy": 0.11890473761595785,
+      "epoch": 7.615570307785154,
+      "grad_norm": 0.4081912934780121,
+      "learning_rate": 6.166397771942987e-05,
+      "loss": 0.09396827220916748,
+      "mean_token_accuracy": 0.967512384802103,
+      "num_tokens": 7321163.0,
+      "step": 3160
+    },
+    {
+      "epoch": 7.615570307785154,
+      "eval_entropy": 0.26364776575833226,
+      "eval_loss": 0.9204733967781067,
+      "eval_mean_token_accuracy": 0.827419729714983,
+      "eval_num_tokens": 7321163.0,
+      "eval_runtime": 88.3441,
+      "eval_samples_per_second": 16.074,
+      "eval_steps_per_second": 2.015,
+      "step": 3160
+    },
+    {
+      "entropy": 0.12209134344011545,
+      "epoch": 7.663850331925166,
+      "grad_norm": 0.5159774422645569,
+      "learning_rate": 5.93388342896962e-05,
+      "loss": 0.09292978644371033,
+      "mean_token_accuracy": 0.9658421002328396,
+      "num_tokens": 7366217.0,
+      "step": 3180
+    },
+    {
+      "epoch": 7.663850331925166,
+      "eval_entropy": 0.2714110068725736,
+      "eval_loss": 0.9143092036247253,
+      "eval_mean_token_accuracy": 0.8262251099843657,
+      "eval_num_tokens": 7366217.0,
+      "eval_runtime": 87.9635,
+      "eval_samples_per_second": 16.143,
+      "eval_steps_per_second": 2.024,
+      "step": 3180
+    },
+    {
+      "entropy": 0.11955609126016498,
+      "epoch": 7.712130356065178,
+      "grad_norm": 0.7005976438522339,
+      "learning_rate": 5.705014101934252e-05,
+      "loss": 0.09319776892662049,
+      "mean_token_accuracy": 0.9674513377249241,
+      "num_tokens": 7410887.0,
+      "step": 3200
+    },
+    {
+      "epoch": 7.712130356065178,
+      "eval_entropy": 0.2626291513442993,
+      "eval_loss": 0.91884446144104,
+      "eval_mean_token_accuracy": 0.827563955207889,
+      "eval_num_tokens": 7410887.0,
+      "eval_runtime": 87.8849,
+      "eval_samples_per_second": 16.157,
+      "eval_steps_per_second": 2.025,
+      "step": 3200
+    },
+    {
+      "entropy": 0.11767532899975777,
+      "epoch": 7.76041038020519,
+      "grad_norm": 0.4675362706184387,
+      "learning_rate": 5.479854558149187e-05,
+      "loss": 0.09038087725639343,
+      "mean_token_accuracy": 0.9669680051505566,
+      "num_tokens": 7457578.0,
+      "step": 3220
+    },
+    {
+      "epoch": 7.76041038020519,
+      "eval_entropy": 0.26581037203582486,
+      "eval_loss": 0.9146883487701416,
+      "eval_mean_token_accuracy": 0.8270345812433222,
+      "eval_num_tokens": 7457578.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 3220
+    },
+    {
+      "entropy": 0.11383061073720455,
+      "epoch": 7.808690404345202,
+      "grad_norm": 0.541247546672821,
+      "learning_rate": 5.258468515101918e-05,
+      "loss": 0.08942405581474304,
+      "mean_token_accuracy": 0.968398705124855,
+      "num_tokens": 7505761.0,
+      "step": 3240
+    },
+    {
+      "epoch": 7.808690404345202,
+      "eval_entropy": 0.26687998960861997,
+      "eval_loss": 0.9037085175514221,
+      "eval_mean_token_accuracy": 0.8274977863504646,
+      "eval_num_tokens": 7505761.0,
+      "eval_runtime": 88.5163,
+      "eval_samples_per_second": 16.042,
+      "eval_steps_per_second": 2.011,
+      "step": 3240
+    },
+    {
+      "entropy": 0.12004058873280883,
+      "epoch": 7.856970428485214,
+      "grad_norm": 0.5289552211761475,
+      "learning_rate": 5.040918622423899e-05,
+      "loss": 0.09064805507659912,
+      "mean_token_accuracy": 0.9673330299556255,
+      "num_tokens": 7552557.0,
+      "step": 3260
+    },
+    {
+      "epoch": 7.856970428485214,
+      "eval_entropy": 0.26288379251622085,
+      "eval_loss": 0.9146178364753723,
+      "eval_mean_token_accuracy": 0.8276662448149049,
+      "eval_num_tokens": 7552557.0,
+      "eval_runtime": 88.0723,
+      "eval_samples_per_second": 16.123,
+      "eval_steps_per_second": 2.021,
+      "step": 3260
+    },
+    {
+      "entropy": 0.1180689456872642,
+      "epoch": 7.905250452625227,
+      "grad_norm": 0.4754861891269684,
+      "learning_rate": 4.827266444161377e-05,
+      "loss": 0.09402655363082886,
+      "mean_token_accuracy": 0.9663361787796021,
+      "num_tokens": 7596998.0,
+      "step": 3280
+    },
+    {
+      "epoch": 7.905250452625227,
+      "eval_entropy": 0.261690996922134,
+      "eval_loss": 0.9120186567306519,
+      "eval_mean_token_accuracy": 0.8278995015647974,
+      "eval_num_tokens": 7596998.0,
+      "eval_runtime": 88.6573,
+      "eval_samples_per_second": 16.017,
+      "eval_steps_per_second": 2.008,
+      "step": 3280
+    },
+    {
+      "entropy": 0.1162164019420743,
+      "epoch": 7.953530476765239,
+      "grad_norm": 0.6203398108482361,
+      "learning_rate": 4.61757244135355e-05,
+      "loss": 0.08832768201828003,
+      "mean_token_accuracy": 0.9694220684468746,
+      "num_tokens": 7643288.0,
+      "step": 3300
+    },
+    {
+      "epoch": 7.953530476765239,
+      "eval_entropy": 0.26158678129817664,
+      "eval_loss": 0.9301361441612244,
+      "eval_mean_token_accuracy": 0.8269603885291668,
+      "eval_num_tokens": 7643288.0,
+      "eval_runtime": 88.7064,
+      "eval_samples_per_second": 16.008,
+      "eval_steps_per_second": 2.007,
+      "step": 3300
+    },
+    {
+      "entropy": 0.10877202204489089,
+      "epoch": 8.0,
+      "grad_norm": 3.058673143386841,
+      "learning_rate": 4.411895954922763e-05,
+      "loss": 0.09159615635871887,
+      "mean_token_accuracy": 0.9686878834451947,
+      "num_tokens": 7688608.0,
+      "step": 3320
+    },
+    {
+      "epoch": 8.0,
+      "eval_entropy": 0.2590438057197614,
+      "eval_loss": 0.9262065887451172,
+      "eval_mean_token_accuracy": 0.8277404546067956,
+      "eval_num_tokens": 7688608.0,
+      "eval_runtime": 87.7557,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 3320
+    },
+    {
+      "entropy": 0.09969092034734786,
+      "epoch": 8.048280024140013,
+      "grad_norm": 0.2561911642551422,
+      "learning_rate": 4.2102951888817644e-05,
+      "loss": 0.07228735685348511,
+      "mean_token_accuracy": 0.9743556551635265,
+      "num_tokens": 7732856.0,
+      "step": 3340
+    },
+    {
+      "epoch": 8.048280024140013,
+      "eval_entropy": 0.2462534454933713,
+      "eval_loss": 0.9755983352661133,
+      "eval_mean_token_accuracy": 0.827510209565752,
+      "eval_num_tokens": 7732856.0,
+      "eval_runtime": 88.0297,
+      "eval_samples_per_second": 16.131,
+      "eval_steps_per_second": 2.022,
+      "step": 3340
+    },
+    {
+      "entropy": 0.09477624730207027,
+      "epoch": 8.096560048280024,
+      "grad_norm": 0.3670356571674347,
+      "learning_rate": 4.012827193862633e-05,
+      "loss": 0.0686036467552185,
+      "mean_token_accuracy": 0.9743670664727688,
+      "num_tokens": 7778585.0,
+      "step": 3360
+    },
+    {
+      "epoch": 8.096560048280024,
+      "eval_entropy": 0.2440882951356052,
+      "eval_loss": 1.0118423700332642,
+      "eval_mean_token_accuracy": 0.8271241070849172,
+      "eval_num_tokens": 7778585.0,
+      "eval_runtime": 88.1364,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 3360
+    },
+    {
+      "entropy": 0.09417980071157217,
+      "epoch": 8.144840072420036,
+      "grad_norm": 0.5785111784934998,
+      "learning_rate": 3.819547850972133e-05,
+      "loss": 0.07100698947906495,
+      "mean_token_accuracy": 0.9737804099917412,
+      "num_tokens": 7824287.0,
+      "step": 3380
+    },
+    {
+      "epoch": 8.144840072420036,
+      "eval_entropy": 0.24603172860453637,
+      "eval_loss": 1.0021991729736328,
+      "eval_mean_token_accuracy": 0.827162122793412,
+      "eval_num_tokens": 7824287.0,
+      "eval_runtime": 88.4534,
+      "eval_samples_per_second": 16.054,
+      "eval_steps_per_second": 2.012,
+      "step": 3380
+    },
+    {
+      "entropy": 0.087222247896716,
+      "epoch": 8.193120096560047,
+      "grad_norm": 0.44848504662513733,
+      "learning_rate": 3.6305118559780393e-05,
+      "loss": 0.06325234174728393,
+      "mean_token_accuracy": 0.9769335232675076,
+      "num_tokens": 7874784.0,
+      "step": 3400
+    },
+    {
+      "epoch": 8.193120096560047,
+      "eval_entropy": 0.24544863905129807,
+      "eval_loss": 1.005008339881897,
+      "eval_mean_token_accuracy": 0.8273035240976998,
+      "eval_num_tokens": 7874784.0,
+      "eval_runtime": 88.6737,
+      "eval_samples_per_second": 16.014,
+      "eval_steps_per_second": 2.007,
+      "step": 3400
+    },
+    {
+      "entropy": 0.09037275062873959,
+      "epoch": 8.24140012070006,
+      "grad_norm": 0.23529155552387238,
+      "learning_rate": 3.4457727038308435e-05,
+      "loss": 0.06589569449424744,
+      "mean_token_accuracy": 0.9751470290124417,
+      "num_tokens": 7925240.0,
+      "step": 3420
+    },
+    {
+      "epoch": 8.24140012070006,
+      "eval_entropy": 0.24203423494368456,
+      "eval_loss": 1.0052824020385742,
+      "eval_mean_token_accuracy": 0.8275575503874361,
+      "eval_num_tokens": 7925240.0,
+      "eval_runtime": 88.3486,
+      "eval_samples_per_second": 16.073,
+      "eval_steps_per_second": 2.015,
+      "step": 3420
+    },
+    {
+      "entropy": 0.10373208015225828,
+      "epoch": 8.289680144840073,
+      "grad_norm": 0.24796408414840698,
+      "learning_rate": 3.2653826735253594e-05,
+      "loss": 0.07714771628379821,
+      "mean_token_accuracy": 0.9722891598939896,
+      "num_tokens": 7966685.0,
+      "step": 3440
+    },
+    {
+      "epoch": 8.289680144840073,
+      "eval_entropy": 0.23461157050025597,
+      "eval_loss": 1.012108325958252,
+      "eval_mean_token_accuracy": 0.82825900564033,
+      "eval_num_tokens": 7966685.0,
+      "eval_runtime": 88.1018,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 3440
+    },
+    {
+      "entropy": 0.08967954274266958,
+      "epoch": 8.337960168980084,
+      "grad_norm": 0.5314674973487854,
+      "learning_rate": 3.08939281330635e-05,
+      "loss": 0.06746266484260559,
+      "mean_token_accuracy": 0.9753133699297905,
+      "num_tokens": 8013656.0,
+      "step": 3460
+    },
+    {
+      "epoch": 8.337960168980084,
+      "eval_entropy": 0.23781998589467468,
+      "eval_loss": 1.0119913816452026,
+      "eval_mean_token_accuracy": 0.8273287157664139,
+      "eval_num_tokens": 8013656.0,
+      "eval_runtime": 88.6038,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3460
+    },
+    {
+      "entropy": 0.09280927330255509,
+      "epoch": 8.386240193120097,
+      "grad_norm": 0.30658936500549316,
+      "learning_rate": 2.9178529262224946e-05,
+      "loss": 0.06936421990394592,
+      "mean_token_accuracy": 0.9742105580866337,
+      "num_tokens": 8061108.0,
+      "step": 3480
+    },
+    {
+      "epoch": 8.386240193120097,
+      "eval_entropy": 0.23672426734747512,
+      "eval_loss": 1.0162041187286377,
+      "eval_mean_token_accuracy": 0.8278394281194451,
+      "eval_num_tokens": 8061108.0,
+      "eval_runtime": 88.6015,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 3480
+    },
+    {
+      "entropy": 0.09195122267119586,
+      "epoch": 8.43452021726011,
+      "grad_norm": 0.24512454867362976,
+      "learning_rate": 2.7508115560327256e-05,
+      "loss": 0.06996064782142639,
+      "mean_token_accuracy": 0.974104854464531,
+      "num_tokens": 8106924.0,
+      "step": 3500
+    },
+    {
+      "epoch": 8.43452021726011,
+      "eval_entropy": 0.23758245828781235,
+      "eval_loss": 1.0157747268676758,
+      "eval_mean_token_accuracy": 0.8278501579600773,
+      "eval_num_tokens": 8106924.0,
+      "eval_runtime": 88.5678,
+      "eval_samples_per_second": 16.033,
+      "eval_steps_per_second": 2.01,
+      "step": 3500
+    },
+    {
+      "entropy": 0.09518477590754629,
+      "epoch": 8.48280024140012,
+      "grad_norm": 0.2373228371143341,
+      "learning_rate": 2.5883159734688868e-05,
+      "loss": 0.06832797527313232,
+      "mean_token_accuracy": 0.9743806682527065,
+      "num_tokens": 8153070.0,
+      "step": 3520
+    },
+    {
+      "epoch": 8.48280024140012,
+      "eval_entropy": 0.2379308074210467,
+      "eval_loss": 1.0180530548095703,
+      "eval_mean_token_accuracy": 0.8282477052024241,
+      "eval_num_tokens": 8153070.0,
+      "eval_runtime": 88.4651,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 3520
+    },
+    {
+      "entropy": 0.0906269280705601,
+      "epoch": 8.531080265540133,
+      "grad_norm": 0.37602657079696655,
+      "learning_rate": 2.4304121628587173e-05,
+      "loss": 0.06921588778495788,
+      "mean_token_accuracy": 0.9745571479201317,
+      "num_tokens": 8200066.0,
+      "step": 3540
+    },
+    {
+      "epoch": 8.531080265540133,
+      "eval_entropy": 0.2357490769263064,
+      "eval_loss": 1.017139196395874,
+      "eval_mean_token_accuracy": 0.828713178299786,
+      "eval_num_tokens": 8200066.0,
+      "eval_runtime": 88.6424,
+      "eval_samples_per_second": 16.019,
+      "eval_steps_per_second": 2.008,
+      "step": 3540
+    },
+    {
+      "entropy": 0.09273334196768701,
+      "epoch": 8.579360289680144,
+      "grad_norm": 0.2978838086128235,
+      "learning_rate": 2.277144809112794e-05,
+      "loss": 0.06970887780189514,
+      "mean_token_accuracy": 0.9751374177634716,
+      "num_tokens": 8246157.0,
+      "step": 3560
+    },
+    {
+      "epoch": 8.579360289680144,
+      "eval_entropy": 0.23590933138065123,
+      "eval_loss": 1.019231915473938,
+      "eval_mean_token_accuracy": 0.8279163412163767,
+      "eval_num_tokens": 8246157.0,
+      "eval_runtime": 88.2691,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 3560
+    },
+    {
+      "entropy": 0.09854201921261847,
+      "epoch": 8.627640313820157,
+      "grad_norm": 0.5381243228912354,
+      "learning_rate": 2.1285572850792413e-05,
+      "loss": 0.07123501300811767,
+      "mean_token_accuracy": 0.9723988175392151,
+      "num_tokens": 8291752.0,
+      "step": 3580
+    },
+    {
+      "epoch": 8.627640313820157,
+      "eval_entropy": 0.23592024949494372,
+      "eval_loss": 1.020250678062439,
+      "eval_mean_token_accuracy": 0.8282466226079491,
+      "eval_num_tokens": 8291752.0,
+      "eval_runtime": 87.8939,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 3580
+    },
+    {
+      "entropy": 0.09092000788077712,
+      "epoch": 8.67592033796017,
+      "grad_norm": 0.29089561104774475,
+      "learning_rate": 1.9846916392697493e-05,
+      "loss": 0.06982720494270325,
+      "mean_token_accuracy": 0.9749995484948158,
+      "num_tokens": 8337651.0,
+      "step": 3600
+    },
+    {
+      "epoch": 8.67592033796017,
+      "eval_entropy": 0.23658913382318583,
+      "eval_loss": 1.019862413406372,
+      "eval_mean_token_accuracy": 0.8280127842104836,
+      "eval_num_tokens": 8337651.0,
+      "eval_runtime": 88.2855,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 3600
+    },
+    {
+      "entropy": 0.08708424470387399,
+      "epoch": 8.72420036210018,
+      "grad_norm": 0.29904839396476746,
+      "learning_rate": 1.8455885839603106e-05,
+      "loss": 0.06801370978355407,
+      "mean_token_accuracy": 0.9755801670253277,
+      "num_tokens": 8385330.0,
+      "step": 3620
+    },
+    {
+      "epoch": 8.72420036210018,
+      "eval_entropy": 0.23627035771862845,
+      "eval_loss": 1.0204569101333618,
+      "eval_mean_token_accuracy": 0.8279967894045155,
+      "eval_num_tokens": 8385330.0,
+      "eval_runtime": 88.1753,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 3620
+    },
+    {
+      "entropy": 0.0916255566291511,
+      "epoch": 8.772480386240193,
+      "grad_norm": 0.20571033656597137,
+      "learning_rate": 1.7112874836701502e-05,
+      "loss": 0.0692249894142151,
+      "mean_token_accuracy": 0.9745148062705994,
+      "num_tokens": 8432134.0,
+      "step": 3640
+    },
+    {
+      "epoch": 8.772480386240193,
+      "eval_entropy": 0.2340821772813797,
+      "eval_loss": 1.0212821960449219,
+      "eval_mean_token_accuracy": 0.828373622358515,
+      "eval_num_tokens": 8432134.0,
+      "eval_runtime": 87.2518,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 3640
+    },
+    {
+      "entropy": 0.0926868049427867,
+      "epoch": 8.820760410380204,
+      "grad_norm": 0.2618265450000763,
+      "learning_rate": 1.5818263440220203e-05,
+      "loss": 0.07338647246360779,
+      "mean_token_accuracy": 0.9724708944559097,
+      "num_tokens": 8476758.0,
+      "step": 3660
+    },
+    {
+      "epoch": 8.820760410380204,
+      "eval_entropy": 0.23637074575330433,
+      "eval_loss": 1.0186725854873657,
+      "eval_mean_token_accuracy": 0.8277707605549459,
+      "eval_num_tokens": 8476758.0,
+      "eval_runtime": 88.4695,
+      "eval_samples_per_second": 16.051,
+      "eval_steps_per_second": 2.012,
+      "step": 3660
+    },
+    {
+      "entropy": 0.09175782897509635,
+      "epoch": 8.869040434520217,
+      "grad_norm": 0.2889852821826935,
+      "learning_rate": 1.4572418009870604e-05,
+      "loss": 0.07066041231155396,
+      "mean_token_accuracy": 0.9725673109292984,
+      "num_tokens": 8523294.0,
+      "step": 3680
+    },
+    {
+      "epoch": 8.869040434520217,
+      "eval_entropy": 0.23945164404223473,
+      "eval_loss": 1.0145231485366821,
+      "eval_mean_token_accuracy": 0.8281202523895864,
+      "eval_num_tokens": 8523294.0,
+      "eval_runtime": 88.368,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 3680
+    },
+    {
+      "entropy": 0.09265611860901117,
+      "epoch": 8.91732045866023,
+      "grad_norm": 0.41092950105667114,
+      "learning_rate": 1.3375691105172625e-05,
+      "loss": 0.06975027322769164,
+      "mean_token_accuracy": 0.974636297672987,
+      "num_tokens": 8570358.0,
+      "step": 3700
+    },
+    {
+      "epoch": 8.91732045866023,
+      "eval_entropy": 0.239588919901446,
+      "eval_loss": 1.014059066772461,
+      "eval_mean_token_accuracy": 0.828028041660116,
+      "eval_num_tokens": 8570358.0,
+      "eval_runtime": 88.1039,
+      "eval_samples_per_second": 16.117,
+      "eval_steps_per_second": 2.02,
+      "step": 3700
+    },
+    {
+      "entropy": 0.0896465003490448,
+      "epoch": 8.965600482800241,
+      "grad_norm": 0.24000254273414612,
+      "learning_rate": 1.2228421385684305e-05,
+      "loss": 0.06743953824043274,
+      "mean_token_accuracy": 0.974006575345993,
+      "num_tokens": 8618246.0,
+      "step": 3720
+    },
+    {
+      "epoch": 8.965600482800241,
+      "eval_entropy": 0.2390461369846644,
+      "eval_loss": 1.011997938156128,
+      "eval_mean_token_accuracy": 0.8284533211354459,
+      "eval_num_tokens": 8618246.0,
+      "eval_runtime": 88.6202,
+      "eval_samples_per_second": 16.023,
+      "eval_steps_per_second": 2.009,
+      "step": 3720
+    },
+    {
+      "entropy": 0.09575109251520851,
+      "epoch": 9.012070006035003,
+      "grad_norm": 0.18088576197624207,
+      "learning_rate": 1.1130933515165429e-05,
+      "loss": 0.07295305132865906,
+      "mean_token_accuracy": 0.9739722857227573,
+      "num_tokens": 8661177.0,
+      "step": 3740
+    },
+    {
+      "epoch": 9.012070006035003,
+      "eval_entropy": 0.23728882387447892,
+      "eval_loss": 1.0134038925170898,
+      "eval_mean_token_accuracy": 0.8286508951294288,
+      "eval_num_tokens": 8661177.0,
+      "eval_runtime": 88.6434,
+      "eval_samples_per_second": 16.019,
+      "eval_steps_per_second": 2.008,
+      "step": 3740
+    },
+    {
+      "entropy": 0.08883201805874705,
+      "epoch": 9.060350030175016,
+      "grad_norm": 0.23120898008346558,
+      "learning_rate": 1.008353806970132e-05,
+      "loss": 0.06326142549514771,
+      "mean_token_accuracy": 0.9780020996928215,
+      "num_tokens": 8707602.0,
+      "step": 3760
+    },
+    {
+      "epoch": 9.060350030175016,
+      "eval_entropy": 0.23416905331142832,
+      "eval_loss": 1.0284100770950317,
+      "eval_mean_token_accuracy": 0.8284083213029283,
+      "eval_num_tokens": 8707602.0,
+      "eval_runtime": 88.2933,
+      "eval_samples_per_second": 16.083,
+      "eval_steps_per_second": 2.016,
+      "step": 3760
+    },
+    {
+      "entropy": 0.08861468276008963,
+      "epoch": 9.108630054315027,
+      "grad_norm": 0.17277656495571136,
+      "learning_rate": 9.086531449813489e-06,
+      "loss": 0.06727416515350342,
+      "mean_token_accuracy": 0.9760390736162663,
+      "num_tokens": 8751287.0,
+      "step": 3780
+    },
+    {
+      "epoch": 9.108630054315027,
+      "eval_entropy": 0.23370317386442355,
+      "eval_loss": 1.0381091833114624,
+      "eval_mean_token_accuracy": 0.8282668178670862,
+      "eval_num_tokens": 8751287.0,
+      "eval_runtime": 88.5738,
+      "eval_samples_per_second": 16.032,
+      "eval_steps_per_second": 2.01,
+      "step": 3780
+    },
+    {
+      "entropy": 0.09032954541034996,
+      "epoch": 9.15691007845504,
+      "grad_norm": 0.19481085240840912,
+      "learning_rate": 8.140195796581981e-06,
+      "loss": 0.06117786169052124,
+      "mean_token_accuracy": 0.9764454193413258,
+      "num_tokens": 8799982.0,
+      "step": 3800
+    },
+    {
+      "epoch": 9.15691007845504,
+      "eval_entropy": 0.23215936609868254,
+      "eval_loss": 1.0467767715454102,
+      "eval_mean_token_accuracy": 0.828405868806196,
+      "eval_num_tokens": 8799982.0,
+      "eval_runtime": 87.9551,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 3800
+    },
+    {
+      "entropy": 0.08824449330568314,
+      "epoch": 9.20519010259505,
+      "grad_norm": 0.20382793247699738,
+      "learning_rate": 7.2447989118025455e-06,
+      "loss": 0.06422553658485412,
+      "mean_token_accuracy": 0.9763121999800205,
+      "num_tokens": 8845536.0,
+      "step": 3820
+    },
+    {
+      "epoch": 9.20519010259505,
+      "eval_entropy": 0.23169661472352704,
+      "eval_loss": 1.0529043674468994,
+      "eval_mean_token_accuracy": 0.8282025610463003,
+      "eval_num_tokens": 8845536.0,
+      "eval_runtime": 88.5548,
+      "eval_samples_per_second": 16.035,
+      "eval_steps_per_second": 2.01,
+      "step": 3820
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.7405672434017894e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3840/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3840/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3840/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3840/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3840/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3840/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3840/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3840/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3840/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3840/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3840/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3840/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3840/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3840/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..6c07ec60a405e6c17924fde8c2f29b3c473dfa50
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3840/trainer_state.json
@@ -0,0 +1,4066 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 9.253470126735063,
+  "eval_steps": 20,
+  "global_step": 3840,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    },
+    {
+      "entropy": 0.1326121488586068,
+      "epoch": 7.181050090525045,
+      "grad_norm": 0.4843117296695709,
+      "learning_rate": 8.411870386507724e-05,
+      "loss": 0.09537842869758606,
+      "mean_token_accuracy": 0.9645151920616627,
+      "num_tokens": 6898444.0,
+      "step": 2980
+    },
+    {
+      "epoch": 7.181050090525045,
+      "eval_entropy": 0.27999477783280813,
+      "eval_loss": 0.9031037092208862,
+      "eval_mean_token_accuracy": 0.8261882204018282,
+      "eval_num_tokens": 6898444.0,
+      "eval_runtime": 88.5602,
+      "eval_samples_per_second": 16.034,
+      "eval_steps_per_second": 2.01,
+      "step": 2980
+    },
+    {
+      "entropy": 0.1156751093454659,
+      "epoch": 7.229330114665057,
+      "grad_norm": 0.5297922492027283,
+      "learning_rate": 8.149629351430326e-05,
+      "loss": 0.09203402996063233,
+      "mean_token_accuracy": 0.9689733698964119,
+      "num_tokens": 6947106.0,
+      "step": 3000
+    },
+    {
+      "epoch": 7.229330114665057,
+      "eval_entropy": 0.2740225845508361,
+      "eval_loss": 0.8876440525054932,
+      "eval_mean_token_accuracy": 0.8265116857678703,
+      "eval_num_tokens": 6947106.0,
+      "eval_runtime": 87.7971,
+      "eval_samples_per_second": 16.174,
+      "eval_steps_per_second": 2.027,
+      "step": 3000
+    },
+    {
+      "entropy": 0.12116114404052496,
+      "epoch": 7.27761013880507,
+      "grad_norm": 0.5467398166656494,
+      "learning_rate": 7.890406302404618e-05,
+      "loss": 0.0933056652545929,
+      "mean_token_accuracy": 0.9676455594599247,
+      "num_tokens": 6992784.0,
+      "step": 3020
+    },
+    {
+      "epoch": 7.27761013880507,
+      "eval_entropy": 0.27238342793804876,
+      "eval_loss": 0.9166117310523987,
+      "eval_mean_token_accuracy": 0.825499015242866,
+      "eval_num_tokens": 6992784.0,
+      "eval_runtime": 88.3276,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 3020
+    },
+    {
+      "entropy": 0.11715119928121567,
+      "epoch": 7.325890162945082,
+      "grad_norm": 0.4976702332496643,
+      "learning_rate": 7.634274596487176e-05,
+      "loss": 0.09449659585952759,
+      "mean_token_accuracy": 0.966506478190422,
+      "num_tokens": 7040482.0,
+      "step": 3040
+    },
+    {
+      "epoch": 7.325890162945082,
+      "eval_entropy": 0.2788975556077582,
+      "eval_loss": 0.8833065629005432,
+      "eval_mean_token_accuracy": 0.8269920596915684,
+      "eval_num_tokens": 7040482.0,
+      "eval_runtime": 88.9016,
+      "eval_samples_per_second": 15.973,
+      "eval_steps_per_second": 2.002,
+      "step": 3040
+    },
+    {
+      "entropy": 0.12565911412239075,
+      "epoch": 7.3741701870850935,
+      "grad_norm": 0.5728679895401001,
+      "learning_rate": 7.381306715921025e-05,
+      "loss": 0.09626244902610778,
+      "mean_token_accuracy": 0.9661444149911403,
+      "num_tokens": 7086261.0,
+      "step": 3060
+    },
+    {
+      "epoch": 7.3741701870850935,
+      "eval_entropy": 0.2762492011939542,
+      "eval_loss": 0.8961160778999329,
+      "eval_mean_token_accuracy": 0.826369910092836,
+      "eval_num_tokens": 7086261.0,
+      "eval_runtime": 88.5204,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 3060
+    },
+    {
+      "entropy": 0.12258567921817302,
+      "epoch": 7.422450211225105,
+      "grad_norm": 0.468985915184021,
+      "learning_rate": 7.131574247624031e-05,
+      "loss": 0.09120266437530518,
+      "mean_token_accuracy": 0.9674938447773457,
+      "num_tokens": 7133247.0,
+      "step": 3080
+    },
+    {
+      "epoch": 7.422450211225105,
+      "eval_entropy": 0.2688512086366,
+      "eval_loss": 0.9048654437065125,
+      "eval_mean_token_accuracy": 0.8265582302982888,
+      "eval_num_tokens": 7133247.0,
+      "eval_runtime": 88.164,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 3080
+    },
+    {
+      "entropy": 0.12070536343380808,
+      "epoch": 7.470730235365117,
+      "grad_norm": 0.5310841798782349,
+      "learning_rate": 6.885147862930688e-05,
+      "loss": 0.09386247992515565,
+      "mean_token_accuracy": 0.96664222702384,
+      "num_tokens": 7177810.0,
+      "step": 3100
+    },
+    {
+      "epoch": 7.470730235365117,
+      "eval_entropy": 0.26289632874593305,
+      "eval_loss": 0.902837336063385,
+      "eval_mean_token_accuracy": 0.828052184554968,
+      "eval_num_tokens": 7177810.0,
+      "eval_runtime": 88.1201,
+      "eval_samples_per_second": 16.114,
+      "eval_steps_per_second": 2.02,
+      "step": 3100
+    },
+    {
+      "entropy": 0.11322665181942285,
+      "epoch": 7.51901025950513,
+      "grad_norm": 0.6846067905426025,
+      "learning_rate": 6.642097297592908e-05,
+      "loss": 0.08772538304328918,
+      "mean_token_accuracy": 0.9691362135112286,
+      "num_tokens": 7225302.0,
+      "step": 3120
+    },
+    {
+      "epoch": 7.51901025950513,
+      "eval_entropy": 0.2670605613776807,
+      "eval_loss": 0.9130566120147705,
+      "eval_mean_token_accuracy": 0.826366903741708,
+      "eval_num_tokens": 7225302.0,
+      "eval_runtime": 88.6035,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3120
+    },
+    {
+      "entropy": 0.10985926580615342,
+      "epoch": 7.567290283645142,
+      "grad_norm": 0.4799748361110687,
+      "learning_rate": 6.402491332045683e-05,
+      "loss": 0.08672296404838561,
+      "mean_token_accuracy": 0.9698824048042297,
+      "num_tokens": 7274881.0,
+      "step": 3140
+    },
+    {
+      "epoch": 7.567290283645142,
+      "eval_entropy": 0.26657010866015146,
+      "eval_loss": 0.9056380987167358,
+      "eval_mean_token_accuracy": 0.8272326461384806,
+      "eval_num_tokens": 7274881.0,
+      "eval_runtime": 88.2967,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3140
+    },
+    {
+      "entropy": 0.11890473761595785,
+      "epoch": 7.615570307785154,
+      "grad_norm": 0.4081912934780121,
+      "learning_rate": 6.166397771942987e-05,
+      "loss": 0.09396827220916748,
+      "mean_token_accuracy": 0.967512384802103,
+      "num_tokens": 7321163.0,
+      "step": 3160
+    },
+    {
+      "epoch": 7.615570307785154,
+      "eval_entropy": 0.26364776575833226,
+      "eval_loss": 0.9204733967781067,
+      "eval_mean_token_accuracy": 0.827419729714983,
+      "eval_num_tokens": 7321163.0,
+      "eval_runtime": 88.3441,
+      "eval_samples_per_second": 16.074,
+      "eval_steps_per_second": 2.015,
+      "step": 3160
+    },
+    {
+      "entropy": 0.12209134344011545,
+      "epoch": 7.663850331925166,
+      "grad_norm": 0.5159774422645569,
+      "learning_rate": 5.93388342896962e-05,
+      "loss": 0.09292978644371033,
+      "mean_token_accuracy": 0.9658421002328396,
+      "num_tokens": 7366217.0,
+      "step": 3180
+    },
+    {
+      "epoch": 7.663850331925166,
+      "eval_entropy": 0.2714110068725736,
+      "eval_loss": 0.9143092036247253,
+      "eval_mean_token_accuracy": 0.8262251099843657,
+      "eval_num_tokens": 7366217.0,
+      "eval_runtime": 87.9635,
+      "eval_samples_per_second": 16.143,
+      "eval_steps_per_second": 2.024,
+      "step": 3180
+    },
+    {
+      "entropy": 0.11955609126016498,
+      "epoch": 7.712130356065178,
+      "grad_norm": 0.7005976438522339,
+      "learning_rate": 5.705014101934252e-05,
+      "loss": 0.09319776892662049,
+      "mean_token_accuracy": 0.9674513377249241,
+      "num_tokens": 7410887.0,
+      "step": 3200
+    },
+    {
+      "epoch": 7.712130356065178,
+      "eval_entropy": 0.2626291513442993,
+      "eval_loss": 0.91884446144104,
+      "eval_mean_token_accuracy": 0.827563955207889,
+      "eval_num_tokens": 7410887.0,
+      "eval_runtime": 87.8849,
+      "eval_samples_per_second": 16.157,
+      "eval_steps_per_second": 2.025,
+      "step": 3200
+    },
+    {
+      "entropy": 0.11767532899975777,
+      "epoch": 7.76041038020519,
+      "grad_norm": 0.4675362706184387,
+      "learning_rate": 5.479854558149187e-05,
+      "loss": 0.09038087725639343,
+      "mean_token_accuracy": 0.9669680051505566,
+      "num_tokens": 7457578.0,
+      "step": 3220
+    },
+    {
+      "epoch": 7.76041038020519,
+      "eval_entropy": 0.26581037203582486,
+      "eval_loss": 0.9146883487701416,
+      "eval_mean_token_accuracy": 0.8270345812433222,
+      "eval_num_tokens": 7457578.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 3220
+    },
+    {
+      "entropy": 0.11383061073720455,
+      "epoch": 7.808690404345202,
+      "grad_norm": 0.541247546672821,
+      "learning_rate": 5.258468515101918e-05,
+      "loss": 0.08942405581474304,
+      "mean_token_accuracy": 0.968398705124855,
+      "num_tokens": 7505761.0,
+      "step": 3240
+    },
+    {
+      "epoch": 7.808690404345202,
+      "eval_entropy": 0.26687998960861997,
+      "eval_loss": 0.9037085175514221,
+      "eval_mean_token_accuracy": 0.8274977863504646,
+      "eval_num_tokens": 7505761.0,
+      "eval_runtime": 88.5163,
+      "eval_samples_per_second": 16.042,
+      "eval_steps_per_second": 2.011,
+      "step": 3240
+    },
+    {
+      "entropy": 0.12004058873280883,
+      "epoch": 7.856970428485214,
+      "grad_norm": 0.5289552211761475,
+      "learning_rate": 5.040918622423899e-05,
+      "loss": 0.09064805507659912,
+      "mean_token_accuracy": 0.9673330299556255,
+      "num_tokens": 7552557.0,
+      "step": 3260
+    },
+    {
+      "epoch": 7.856970428485214,
+      "eval_entropy": 0.26288379251622085,
+      "eval_loss": 0.9146178364753723,
+      "eval_mean_token_accuracy": 0.8276662448149049,
+      "eval_num_tokens": 7552557.0,
+      "eval_runtime": 88.0723,
+      "eval_samples_per_second": 16.123,
+      "eval_steps_per_second": 2.021,
+      "step": 3260
+    },
+    {
+      "entropy": 0.1180689456872642,
+      "epoch": 7.905250452625227,
+      "grad_norm": 0.4754861891269684,
+      "learning_rate": 4.827266444161377e-05,
+      "loss": 0.09402655363082886,
+      "mean_token_accuracy": 0.9663361787796021,
+      "num_tokens": 7596998.0,
+      "step": 3280
+    },
+    {
+      "epoch": 7.905250452625227,
+      "eval_entropy": 0.261690996922134,
+      "eval_loss": 0.9120186567306519,
+      "eval_mean_token_accuracy": 0.8278995015647974,
+      "eval_num_tokens": 7596998.0,
+      "eval_runtime": 88.6573,
+      "eval_samples_per_second": 16.017,
+      "eval_steps_per_second": 2.008,
+      "step": 3280
+    },
+    {
+      "entropy": 0.1162164019420743,
+      "epoch": 7.953530476765239,
+      "grad_norm": 0.6203398108482361,
+      "learning_rate": 4.61757244135355e-05,
+      "loss": 0.08832768201828003,
+      "mean_token_accuracy": 0.9694220684468746,
+      "num_tokens": 7643288.0,
+      "step": 3300
+    },
+    {
+      "epoch": 7.953530476765239,
+      "eval_entropy": 0.26158678129817664,
+      "eval_loss": 0.9301361441612244,
+      "eval_mean_token_accuracy": 0.8269603885291668,
+      "eval_num_tokens": 7643288.0,
+      "eval_runtime": 88.7064,
+      "eval_samples_per_second": 16.008,
+      "eval_steps_per_second": 2.007,
+      "step": 3300
+    },
+    {
+      "entropy": 0.10877202204489089,
+      "epoch": 8.0,
+      "grad_norm": 3.058673143386841,
+      "learning_rate": 4.411895954922763e-05,
+      "loss": 0.09159615635871887,
+      "mean_token_accuracy": 0.9686878834451947,
+      "num_tokens": 7688608.0,
+      "step": 3320
+    },
+    {
+      "epoch": 8.0,
+      "eval_entropy": 0.2590438057197614,
+      "eval_loss": 0.9262065887451172,
+      "eval_mean_token_accuracy": 0.8277404546067956,
+      "eval_num_tokens": 7688608.0,
+      "eval_runtime": 87.7557,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 3320
+    },
+    {
+      "entropy": 0.09969092034734786,
+      "epoch": 8.048280024140013,
+      "grad_norm": 0.2561911642551422,
+      "learning_rate": 4.2102951888817644e-05,
+      "loss": 0.07228735685348511,
+      "mean_token_accuracy": 0.9743556551635265,
+      "num_tokens": 7732856.0,
+      "step": 3340
+    },
+    {
+      "epoch": 8.048280024140013,
+      "eval_entropy": 0.2462534454933713,
+      "eval_loss": 0.9755983352661133,
+      "eval_mean_token_accuracy": 0.827510209565752,
+      "eval_num_tokens": 7732856.0,
+      "eval_runtime": 88.0297,
+      "eval_samples_per_second": 16.131,
+      "eval_steps_per_second": 2.022,
+      "step": 3340
+    },
+    {
+      "entropy": 0.09477624730207027,
+      "epoch": 8.096560048280024,
+      "grad_norm": 0.3670356571674347,
+      "learning_rate": 4.012827193862633e-05,
+      "loss": 0.0686036467552185,
+      "mean_token_accuracy": 0.9743670664727688,
+      "num_tokens": 7778585.0,
+      "step": 3360
+    },
+    {
+      "epoch": 8.096560048280024,
+      "eval_entropy": 0.2440882951356052,
+      "eval_loss": 1.0118423700332642,
+      "eval_mean_token_accuracy": 0.8271241070849172,
+      "eval_num_tokens": 7778585.0,
+      "eval_runtime": 88.1364,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 3360
+    },
+    {
+      "entropy": 0.09417980071157217,
+      "epoch": 8.144840072420036,
+      "grad_norm": 0.5785111784934998,
+      "learning_rate": 3.819547850972133e-05,
+      "loss": 0.07100698947906495,
+      "mean_token_accuracy": 0.9737804099917412,
+      "num_tokens": 7824287.0,
+      "step": 3380
+    },
+    {
+      "epoch": 8.144840072420036,
+      "eval_entropy": 0.24603172860453637,
+      "eval_loss": 1.0021991729736328,
+      "eval_mean_token_accuracy": 0.827162122793412,
+      "eval_num_tokens": 7824287.0,
+      "eval_runtime": 88.4534,
+      "eval_samples_per_second": 16.054,
+      "eval_steps_per_second": 2.012,
+      "step": 3380
+    },
+    {
+      "entropy": 0.087222247896716,
+      "epoch": 8.193120096560047,
+      "grad_norm": 0.44848504662513733,
+      "learning_rate": 3.6305118559780393e-05,
+      "loss": 0.06325234174728393,
+      "mean_token_accuracy": 0.9769335232675076,
+      "num_tokens": 7874784.0,
+      "step": 3400
+    },
+    {
+      "epoch": 8.193120096560047,
+      "eval_entropy": 0.24544863905129807,
+      "eval_loss": 1.005008339881897,
+      "eval_mean_token_accuracy": 0.8273035240976998,
+      "eval_num_tokens": 7874784.0,
+      "eval_runtime": 88.6737,
+      "eval_samples_per_second": 16.014,
+      "eval_steps_per_second": 2.007,
+      "step": 3400
+    },
+    {
+      "entropy": 0.09037275062873959,
+      "epoch": 8.24140012070006,
+      "grad_norm": 0.23529155552387238,
+      "learning_rate": 3.4457727038308435e-05,
+      "loss": 0.06589569449424744,
+      "mean_token_accuracy": 0.9751470290124417,
+      "num_tokens": 7925240.0,
+      "step": 3420
+    },
+    {
+      "epoch": 8.24140012070006,
+      "eval_entropy": 0.24203423494368456,
+      "eval_loss": 1.0052824020385742,
+      "eval_mean_token_accuracy": 0.8275575503874361,
+      "eval_num_tokens": 7925240.0,
+      "eval_runtime": 88.3486,
+      "eval_samples_per_second": 16.073,
+      "eval_steps_per_second": 2.015,
+      "step": 3420
+    },
+    {
+      "entropy": 0.10373208015225828,
+      "epoch": 8.289680144840073,
+      "grad_norm": 0.24796408414840698,
+      "learning_rate": 3.2653826735253594e-05,
+      "loss": 0.07714771628379821,
+      "mean_token_accuracy": 0.9722891598939896,
+      "num_tokens": 7966685.0,
+      "step": 3440
+    },
+    {
+      "epoch": 8.289680144840073,
+      "eval_entropy": 0.23461157050025597,
+      "eval_loss": 1.012108325958252,
+      "eval_mean_token_accuracy": 0.82825900564033,
+      "eval_num_tokens": 7966685.0,
+      "eval_runtime": 88.1018,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 3440
+    },
+    {
+      "entropy": 0.08967954274266958,
+      "epoch": 8.337960168980084,
+      "grad_norm": 0.5314674973487854,
+      "learning_rate": 3.08939281330635e-05,
+      "loss": 0.06746266484260559,
+      "mean_token_accuracy": 0.9753133699297905,
+      "num_tokens": 8013656.0,
+      "step": 3460
+    },
+    {
+      "epoch": 8.337960168980084,
+      "eval_entropy": 0.23781998589467468,
+      "eval_loss": 1.0119913816452026,
+      "eval_mean_token_accuracy": 0.8273287157664139,
+      "eval_num_tokens": 8013656.0,
+      "eval_runtime": 88.6038,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3460
+    },
+    {
+      "entropy": 0.09280927330255509,
+      "epoch": 8.386240193120097,
+      "grad_norm": 0.30658936500549316,
+      "learning_rate": 2.9178529262224946e-05,
+      "loss": 0.06936421990394592,
+      "mean_token_accuracy": 0.9742105580866337,
+      "num_tokens": 8061108.0,
+      "step": 3480
+    },
+    {
+      "epoch": 8.386240193120097,
+      "eval_entropy": 0.23672426734747512,
+      "eval_loss": 1.0162041187286377,
+      "eval_mean_token_accuracy": 0.8278394281194451,
+      "eval_num_tokens": 8061108.0,
+      "eval_runtime": 88.6015,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 3480
+    },
+    {
+      "entropy": 0.09195122267119586,
+      "epoch": 8.43452021726011,
+      "grad_norm": 0.24512454867362976,
+      "learning_rate": 2.7508115560327256e-05,
+      "loss": 0.06996064782142639,
+      "mean_token_accuracy": 0.974104854464531,
+      "num_tokens": 8106924.0,
+      "step": 3500
+    },
+    {
+      "epoch": 8.43452021726011,
+      "eval_entropy": 0.23758245828781235,
+      "eval_loss": 1.0157747268676758,
+      "eval_mean_token_accuracy": 0.8278501579600773,
+      "eval_num_tokens": 8106924.0,
+      "eval_runtime": 88.5678,
+      "eval_samples_per_second": 16.033,
+      "eval_steps_per_second": 2.01,
+      "step": 3500
+    },
+    {
+      "entropy": 0.09518477590754629,
+      "epoch": 8.48280024140012,
+      "grad_norm": 0.2373228371143341,
+      "learning_rate": 2.5883159734688868e-05,
+      "loss": 0.06832797527313232,
+      "mean_token_accuracy": 0.9743806682527065,
+      "num_tokens": 8153070.0,
+      "step": 3520
+    },
+    {
+      "epoch": 8.48280024140012,
+      "eval_entropy": 0.2379308074210467,
+      "eval_loss": 1.0180530548095703,
+      "eval_mean_token_accuracy": 0.8282477052024241,
+      "eval_num_tokens": 8153070.0,
+      "eval_runtime": 88.4651,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 3520
+    },
+    {
+      "entropy": 0.0906269280705601,
+      "epoch": 8.531080265540133,
+      "grad_norm": 0.37602657079696655,
+      "learning_rate": 2.4304121628587173e-05,
+      "loss": 0.06921588778495788,
+      "mean_token_accuracy": 0.9745571479201317,
+      "num_tokens": 8200066.0,
+      "step": 3540
+    },
+    {
+      "epoch": 8.531080265540133,
+      "eval_entropy": 0.2357490769263064,
+      "eval_loss": 1.017139196395874,
+      "eval_mean_token_accuracy": 0.828713178299786,
+      "eval_num_tokens": 8200066.0,
+      "eval_runtime": 88.6424,
+      "eval_samples_per_second": 16.019,
+      "eval_steps_per_second": 2.008,
+      "step": 3540
+    },
+    {
+      "entropy": 0.09273334196768701,
+      "epoch": 8.579360289680144,
+      "grad_norm": 0.2978838086128235,
+      "learning_rate": 2.277144809112794e-05,
+      "loss": 0.06970887780189514,
+      "mean_token_accuracy": 0.9751374177634716,
+      "num_tokens": 8246157.0,
+      "step": 3560
+    },
+    {
+      "epoch": 8.579360289680144,
+      "eval_entropy": 0.23590933138065123,
+      "eval_loss": 1.019231915473938,
+      "eval_mean_token_accuracy": 0.8279163412163767,
+      "eval_num_tokens": 8246157.0,
+      "eval_runtime": 88.2691,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 3560
+    },
+    {
+      "entropy": 0.09854201921261847,
+      "epoch": 8.627640313820157,
+      "grad_norm": 0.5381243228912354,
+      "learning_rate": 2.1285572850792413e-05,
+      "loss": 0.07123501300811767,
+      "mean_token_accuracy": 0.9723988175392151,
+      "num_tokens": 8291752.0,
+      "step": 3580
+    },
+    {
+      "epoch": 8.627640313820157,
+      "eval_entropy": 0.23592024949494372,
+      "eval_loss": 1.020250678062439,
+      "eval_mean_token_accuracy": 0.8282466226079491,
+      "eval_num_tokens": 8291752.0,
+      "eval_runtime": 87.8939,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 3580
+    },
+    {
+      "entropy": 0.09092000788077712,
+      "epoch": 8.67592033796017,
+      "grad_norm": 0.29089561104774475,
+      "learning_rate": 1.9846916392697493e-05,
+      "loss": 0.06982720494270325,
+      "mean_token_accuracy": 0.9749995484948158,
+      "num_tokens": 8337651.0,
+      "step": 3600
+    },
+    {
+      "epoch": 8.67592033796017,
+      "eval_entropy": 0.23658913382318583,
+      "eval_loss": 1.019862413406372,
+      "eval_mean_token_accuracy": 0.8280127842104836,
+      "eval_num_tokens": 8337651.0,
+      "eval_runtime": 88.2855,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 3600
+    },
+    {
+      "entropy": 0.08708424470387399,
+      "epoch": 8.72420036210018,
+      "grad_norm": 0.29904839396476746,
+      "learning_rate": 1.8455885839603106e-05,
+      "loss": 0.06801370978355407,
+      "mean_token_accuracy": 0.9755801670253277,
+      "num_tokens": 8385330.0,
+      "step": 3620
+    },
+    {
+      "epoch": 8.72420036210018,
+      "eval_entropy": 0.23627035771862845,
+      "eval_loss": 1.0204569101333618,
+      "eval_mean_token_accuracy": 0.8279967894045155,
+      "eval_num_tokens": 8385330.0,
+      "eval_runtime": 88.1753,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 3620
+    },
+    {
+      "entropy": 0.0916255566291511,
+      "epoch": 8.772480386240193,
+      "grad_norm": 0.20571033656597137,
+      "learning_rate": 1.7112874836701502e-05,
+      "loss": 0.0692249894142151,
+      "mean_token_accuracy": 0.9745148062705994,
+      "num_tokens": 8432134.0,
+      "step": 3640
+    },
+    {
+      "epoch": 8.772480386240193,
+      "eval_entropy": 0.2340821772813797,
+      "eval_loss": 1.0212821960449219,
+      "eval_mean_token_accuracy": 0.828373622358515,
+      "eval_num_tokens": 8432134.0,
+      "eval_runtime": 87.2518,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 3640
+    },
+    {
+      "entropy": 0.0926868049427867,
+      "epoch": 8.820760410380204,
+      "grad_norm": 0.2618265450000763,
+      "learning_rate": 1.5818263440220203e-05,
+      "loss": 0.07338647246360779,
+      "mean_token_accuracy": 0.9724708944559097,
+      "num_tokens": 8476758.0,
+      "step": 3660
+    },
+    {
+      "epoch": 8.820760410380204,
+      "eval_entropy": 0.23637074575330433,
+      "eval_loss": 1.0186725854873657,
+      "eval_mean_token_accuracy": 0.8277707605549459,
+      "eval_num_tokens": 8476758.0,
+      "eval_runtime": 88.4695,
+      "eval_samples_per_second": 16.051,
+      "eval_steps_per_second": 2.012,
+      "step": 3660
+    },
+    {
+      "entropy": 0.09175782897509635,
+      "epoch": 8.869040434520217,
+      "grad_norm": 0.2889852821826935,
+      "learning_rate": 1.4572418009870604e-05,
+      "loss": 0.07066041231155396,
+      "mean_token_accuracy": 0.9725673109292984,
+      "num_tokens": 8523294.0,
+      "step": 3680
+    },
+    {
+      "epoch": 8.869040434520217,
+      "eval_entropy": 0.23945164404223473,
+      "eval_loss": 1.0145231485366821,
+      "eval_mean_token_accuracy": 0.8281202523895864,
+      "eval_num_tokens": 8523294.0,
+      "eval_runtime": 88.368,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 3680
+    },
+    {
+      "entropy": 0.09265611860901117,
+      "epoch": 8.91732045866023,
+      "grad_norm": 0.41092950105667114,
+      "learning_rate": 1.3375691105172625e-05,
+      "loss": 0.06975027322769164,
+      "mean_token_accuracy": 0.974636297672987,
+      "num_tokens": 8570358.0,
+      "step": 3700
+    },
+    {
+      "epoch": 8.91732045866023,
+      "eval_entropy": 0.239588919901446,
+      "eval_loss": 1.014059066772461,
+      "eval_mean_token_accuracy": 0.828028041660116,
+      "eval_num_tokens": 8570358.0,
+      "eval_runtime": 88.1039,
+      "eval_samples_per_second": 16.117,
+      "eval_steps_per_second": 2.02,
+      "step": 3700
+    },
+    {
+      "entropy": 0.0896465003490448,
+      "epoch": 8.965600482800241,
+      "grad_norm": 0.24000254273414612,
+      "learning_rate": 1.2228421385684305e-05,
+      "loss": 0.06743953824043274,
+      "mean_token_accuracy": 0.974006575345993,
+      "num_tokens": 8618246.0,
+      "step": 3720
+    },
+    {
+      "epoch": 8.965600482800241,
+      "eval_entropy": 0.2390461369846644,
+      "eval_loss": 1.011997938156128,
+      "eval_mean_token_accuracy": 0.8284533211354459,
+      "eval_num_tokens": 8618246.0,
+      "eval_runtime": 88.6202,
+      "eval_samples_per_second": 16.023,
+      "eval_steps_per_second": 2.009,
+      "step": 3720
+    },
+    {
+      "entropy": 0.09575109251520851,
+      "epoch": 9.012070006035003,
+      "grad_norm": 0.18088576197624207,
+      "learning_rate": 1.1130933515165429e-05,
+      "loss": 0.07295305132865906,
+      "mean_token_accuracy": 0.9739722857227573,
+      "num_tokens": 8661177.0,
+      "step": 3740
+    },
+    {
+      "epoch": 9.012070006035003,
+      "eval_entropy": 0.23728882387447892,
+      "eval_loss": 1.0134038925170898,
+      "eval_mean_token_accuracy": 0.8286508951294288,
+      "eval_num_tokens": 8661177.0,
+      "eval_runtime": 88.6434,
+      "eval_samples_per_second": 16.019,
+      "eval_steps_per_second": 2.008,
+      "step": 3740
+    },
+    {
+      "entropy": 0.08883201805874705,
+      "epoch": 9.060350030175016,
+      "grad_norm": 0.23120898008346558,
+      "learning_rate": 1.008353806970132e-05,
+      "loss": 0.06326142549514771,
+      "mean_token_accuracy": 0.9780020996928215,
+      "num_tokens": 8707602.0,
+      "step": 3760
+    },
+    {
+      "epoch": 9.060350030175016,
+      "eval_entropy": 0.23416905331142832,
+      "eval_loss": 1.0284100770950317,
+      "eval_mean_token_accuracy": 0.8284083213029283,
+      "eval_num_tokens": 8707602.0,
+      "eval_runtime": 88.2933,
+      "eval_samples_per_second": 16.083,
+      "eval_steps_per_second": 2.016,
+      "step": 3760
+    },
+    {
+      "entropy": 0.08861468276008963,
+      "epoch": 9.108630054315027,
+      "grad_norm": 0.17277656495571136,
+      "learning_rate": 9.086531449813489e-06,
+      "loss": 0.06727416515350342,
+      "mean_token_accuracy": 0.9760390736162663,
+      "num_tokens": 8751287.0,
+      "step": 3780
+    },
+    {
+      "epoch": 9.108630054315027,
+      "eval_entropy": 0.23370317386442355,
+      "eval_loss": 1.0381091833114624,
+      "eval_mean_token_accuracy": 0.8282668178670862,
+      "eval_num_tokens": 8751287.0,
+      "eval_runtime": 88.5738,
+      "eval_samples_per_second": 16.032,
+      "eval_steps_per_second": 2.01,
+      "step": 3780
+    },
+    {
+      "entropy": 0.09032954541034996,
+      "epoch": 9.15691007845504,
+      "grad_norm": 0.19481085240840912,
+      "learning_rate": 8.140195796581981e-06,
+      "loss": 0.06117786169052124,
+      "mean_token_accuracy": 0.9764454193413258,
+      "num_tokens": 8799982.0,
+      "step": 3800
+    },
+    {
+      "epoch": 9.15691007845504,
+      "eval_entropy": 0.23215936609868254,
+      "eval_loss": 1.0467767715454102,
+      "eval_mean_token_accuracy": 0.828405868806196,
+      "eval_num_tokens": 8799982.0,
+      "eval_runtime": 87.9551,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 3800
+    },
+    {
+      "entropy": 0.08824449330568314,
+      "epoch": 9.20519010259505,
+      "grad_norm": 0.20382793247699738,
+      "learning_rate": 7.2447989118025455e-06,
+      "loss": 0.06422553658485412,
+      "mean_token_accuracy": 0.9763121999800205,
+      "num_tokens": 8845536.0,
+      "step": 3820
+    },
+    {
+      "epoch": 9.20519010259505,
+      "eval_entropy": 0.23169661472352704,
+      "eval_loss": 1.0529043674468994,
+      "eval_mean_token_accuracy": 0.8282025610463003,
+      "eval_num_tokens": 8845536.0,
+      "eval_runtime": 88.5548,
+      "eval_samples_per_second": 16.035,
+      "eval_steps_per_second": 2.01,
+      "step": 3820
+    },
+    {
+      "entropy": 0.07768499720841646,
+      "epoch": 9.253470126735063,
+      "grad_norm": 0.11727919429540634,
+      "learning_rate": 6.400594182202084e-06,
+      "loss": 0.057279336452484134,
+      "mean_token_accuracy": 0.9795166745781898,
+      "num_tokens": 8896053.0,
+      "step": 3840
+    },
+    {
+      "epoch": 9.253470126735063,
+      "eval_entropy": 0.2301936373114586,
+      "eval_loss": 1.0576049089431763,
+      "eval_mean_token_accuracy": 0.8283417489421502,
+      "eval_num_tokens": 8896053.0,
+      "eval_runtime": 88.7854,
+      "eval_samples_per_second": 15.994,
+      "eval_steps_per_second": 2.005,
+      "step": 3840
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.761387509935882e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3860/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3860/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3860/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3860/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3860/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3860/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3860/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3860/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3860/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3860/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3860/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3860/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3860/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3860/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..e5fc2e8411c67c126118e479b8d71f3c48e368ff
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3860/trainer_state.json
@@ -0,0 +1,4087 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 9.301750150875076,
+  "eval_steps": 20,
+  "global_step": 3860,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    },
+    {
+      "entropy": 0.1326121488586068,
+      "epoch": 7.181050090525045,
+      "grad_norm": 0.4843117296695709,
+      "learning_rate": 8.411870386507724e-05,
+      "loss": 0.09537842869758606,
+      "mean_token_accuracy": 0.9645151920616627,
+      "num_tokens": 6898444.0,
+      "step": 2980
+    },
+    {
+      "epoch": 7.181050090525045,
+      "eval_entropy": 0.27999477783280813,
+      "eval_loss": 0.9031037092208862,
+      "eval_mean_token_accuracy": 0.8261882204018282,
+      "eval_num_tokens": 6898444.0,
+      "eval_runtime": 88.5602,
+      "eval_samples_per_second": 16.034,
+      "eval_steps_per_second": 2.01,
+      "step": 2980
+    },
+    {
+      "entropy": 0.1156751093454659,
+      "epoch": 7.229330114665057,
+      "grad_norm": 0.5297922492027283,
+      "learning_rate": 8.149629351430326e-05,
+      "loss": 0.09203402996063233,
+      "mean_token_accuracy": 0.9689733698964119,
+      "num_tokens": 6947106.0,
+      "step": 3000
+    },
+    {
+      "epoch": 7.229330114665057,
+      "eval_entropy": 0.2740225845508361,
+      "eval_loss": 0.8876440525054932,
+      "eval_mean_token_accuracy": 0.8265116857678703,
+      "eval_num_tokens": 6947106.0,
+      "eval_runtime": 87.7971,
+      "eval_samples_per_second": 16.174,
+      "eval_steps_per_second": 2.027,
+      "step": 3000
+    },
+    {
+      "entropy": 0.12116114404052496,
+      "epoch": 7.27761013880507,
+      "grad_norm": 0.5467398166656494,
+      "learning_rate": 7.890406302404618e-05,
+      "loss": 0.0933056652545929,
+      "mean_token_accuracy": 0.9676455594599247,
+      "num_tokens": 6992784.0,
+      "step": 3020
+    },
+    {
+      "epoch": 7.27761013880507,
+      "eval_entropy": 0.27238342793804876,
+      "eval_loss": 0.9166117310523987,
+      "eval_mean_token_accuracy": 0.825499015242866,
+      "eval_num_tokens": 6992784.0,
+      "eval_runtime": 88.3276,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 3020
+    },
+    {
+      "entropy": 0.11715119928121567,
+      "epoch": 7.325890162945082,
+      "grad_norm": 0.4976702332496643,
+      "learning_rate": 7.634274596487176e-05,
+      "loss": 0.09449659585952759,
+      "mean_token_accuracy": 0.966506478190422,
+      "num_tokens": 7040482.0,
+      "step": 3040
+    },
+    {
+      "epoch": 7.325890162945082,
+      "eval_entropy": 0.2788975556077582,
+      "eval_loss": 0.8833065629005432,
+      "eval_mean_token_accuracy": 0.8269920596915684,
+      "eval_num_tokens": 7040482.0,
+      "eval_runtime": 88.9016,
+      "eval_samples_per_second": 15.973,
+      "eval_steps_per_second": 2.002,
+      "step": 3040
+    },
+    {
+      "entropy": 0.12565911412239075,
+      "epoch": 7.3741701870850935,
+      "grad_norm": 0.5728679895401001,
+      "learning_rate": 7.381306715921025e-05,
+      "loss": 0.09626244902610778,
+      "mean_token_accuracy": 0.9661444149911403,
+      "num_tokens": 7086261.0,
+      "step": 3060
+    },
+    {
+      "epoch": 7.3741701870850935,
+      "eval_entropy": 0.2762492011939542,
+      "eval_loss": 0.8961160778999329,
+      "eval_mean_token_accuracy": 0.826369910092836,
+      "eval_num_tokens": 7086261.0,
+      "eval_runtime": 88.5204,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 3060
+    },
+    {
+      "entropy": 0.12258567921817302,
+      "epoch": 7.422450211225105,
+      "grad_norm": 0.468985915184021,
+      "learning_rate": 7.131574247624031e-05,
+      "loss": 0.09120266437530518,
+      "mean_token_accuracy": 0.9674938447773457,
+      "num_tokens": 7133247.0,
+      "step": 3080
+    },
+    {
+      "epoch": 7.422450211225105,
+      "eval_entropy": 0.2688512086366,
+      "eval_loss": 0.9048654437065125,
+      "eval_mean_token_accuracy": 0.8265582302982888,
+      "eval_num_tokens": 7133247.0,
+      "eval_runtime": 88.164,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 3080
+    },
+    {
+      "entropy": 0.12070536343380808,
+      "epoch": 7.470730235365117,
+      "grad_norm": 0.5310841798782349,
+      "learning_rate": 6.885147862930688e-05,
+      "loss": 0.09386247992515565,
+      "mean_token_accuracy": 0.96664222702384,
+      "num_tokens": 7177810.0,
+      "step": 3100
+    },
+    {
+      "epoch": 7.470730235365117,
+      "eval_entropy": 0.26289632874593305,
+      "eval_loss": 0.902837336063385,
+      "eval_mean_token_accuracy": 0.828052184554968,
+      "eval_num_tokens": 7177810.0,
+      "eval_runtime": 88.1201,
+      "eval_samples_per_second": 16.114,
+      "eval_steps_per_second": 2.02,
+      "step": 3100
+    },
+    {
+      "entropy": 0.11322665181942285,
+      "epoch": 7.51901025950513,
+      "grad_norm": 0.6846067905426025,
+      "learning_rate": 6.642097297592908e-05,
+      "loss": 0.08772538304328918,
+      "mean_token_accuracy": 0.9691362135112286,
+      "num_tokens": 7225302.0,
+      "step": 3120
+    },
+    {
+      "epoch": 7.51901025950513,
+      "eval_entropy": 0.2670605613776807,
+      "eval_loss": 0.9130566120147705,
+      "eval_mean_token_accuracy": 0.826366903741708,
+      "eval_num_tokens": 7225302.0,
+      "eval_runtime": 88.6035,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3120
+    },
+    {
+      "entropy": 0.10985926580615342,
+      "epoch": 7.567290283645142,
+      "grad_norm": 0.4799748361110687,
+      "learning_rate": 6.402491332045683e-05,
+      "loss": 0.08672296404838561,
+      "mean_token_accuracy": 0.9698824048042297,
+      "num_tokens": 7274881.0,
+      "step": 3140
+    },
+    {
+      "epoch": 7.567290283645142,
+      "eval_entropy": 0.26657010866015146,
+      "eval_loss": 0.9056380987167358,
+      "eval_mean_token_accuracy": 0.8272326461384806,
+      "eval_num_tokens": 7274881.0,
+      "eval_runtime": 88.2967,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3140
+    },
+    {
+      "entropy": 0.11890473761595785,
+      "epoch": 7.615570307785154,
+      "grad_norm": 0.4081912934780121,
+      "learning_rate": 6.166397771942987e-05,
+      "loss": 0.09396827220916748,
+      "mean_token_accuracy": 0.967512384802103,
+      "num_tokens": 7321163.0,
+      "step": 3160
+    },
+    {
+      "epoch": 7.615570307785154,
+      "eval_entropy": 0.26364776575833226,
+      "eval_loss": 0.9204733967781067,
+      "eval_mean_token_accuracy": 0.827419729714983,
+      "eval_num_tokens": 7321163.0,
+      "eval_runtime": 88.3441,
+      "eval_samples_per_second": 16.074,
+      "eval_steps_per_second": 2.015,
+      "step": 3160
+    },
+    {
+      "entropy": 0.12209134344011545,
+      "epoch": 7.663850331925166,
+      "grad_norm": 0.5159774422645569,
+      "learning_rate": 5.93388342896962e-05,
+      "loss": 0.09292978644371033,
+      "mean_token_accuracy": 0.9658421002328396,
+      "num_tokens": 7366217.0,
+      "step": 3180
+    },
+    {
+      "epoch": 7.663850331925166,
+      "eval_entropy": 0.2714110068725736,
+      "eval_loss": 0.9143092036247253,
+      "eval_mean_token_accuracy": 0.8262251099843657,
+      "eval_num_tokens": 7366217.0,
+      "eval_runtime": 87.9635,
+      "eval_samples_per_second": 16.143,
+      "eval_steps_per_second": 2.024,
+      "step": 3180
+    },
+    {
+      "entropy": 0.11955609126016498,
+      "epoch": 7.712130356065178,
+      "grad_norm": 0.7005976438522339,
+      "learning_rate": 5.705014101934252e-05,
+      "loss": 0.09319776892662049,
+      "mean_token_accuracy": 0.9674513377249241,
+      "num_tokens": 7410887.0,
+      "step": 3200
+    },
+    {
+      "epoch": 7.712130356065178,
+      "eval_entropy": 0.2626291513442993,
+      "eval_loss": 0.91884446144104,
+      "eval_mean_token_accuracy": 0.827563955207889,
+      "eval_num_tokens": 7410887.0,
+      "eval_runtime": 87.8849,
+      "eval_samples_per_second": 16.157,
+      "eval_steps_per_second": 2.025,
+      "step": 3200
+    },
+    {
+      "entropy": 0.11767532899975777,
+      "epoch": 7.76041038020519,
+      "grad_norm": 0.4675362706184387,
+      "learning_rate": 5.479854558149187e-05,
+      "loss": 0.09038087725639343,
+      "mean_token_accuracy": 0.9669680051505566,
+      "num_tokens": 7457578.0,
+      "step": 3220
+    },
+    {
+      "epoch": 7.76041038020519,
+      "eval_entropy": 0.26581037203582486,
+      "eval_loss": 0.9146883487701416,
+      "eval_mean_token_accuracy": 0.8270345812433222,
+      "eval_num_tokens": 7457578.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 3220
+    },
+    {
+      "entropy": 0.11383061073720455,
+      "epoch": 7.808690404345202,
+      "grad_norm": 0.541247546672821,
+      "learning_rate": 5.258468515101918e-05,
+      "loss": 0.08942405581474304,
+      "mean_token_accuracy": 0.968398705124855,
+      "num_tokens": 7505761.0,
+      "step": 3240
+    },
+    {
+      "epoch": 7.808690404345202,
+      "eval_entropy": 0.26687998960861997,
+      "eval_loss": 0.9037085175514221,
+      "eval_mean_token_accuracy": 0.8274977863504646,
+      "eval_num_tokens": 7505761.0,
+      "eval_runtime": 88.5163,
+      "eval_samples_per_second": 16.042,
+      "eval_steps_per_second": 2.011,
+      "step": 3240
+    },
+    {
+      "entropy": 0.12004058873280883,
+      "epoch": 7.856970428485214,
+      "grad_norm": 0.5289552211761475,
+      "learning_rate": 5.040918622423899e-05,
+      "loss": 0.09064805507659912,
+      "mean_token_accuracy": 0.9673330299556255,
+      "num_tokens": 7552557.0,
+      "step": 3260
+    },
+    {
+      "epoch": 7.856970428485214,
+      "eval_entropy": 0.26288379251622085,
+      "eval_loss": 0.9146178364753723,
+      "eval_mean_token_accuracy": 0.8276662448149049,
+      "eval_num_tokens": 7552557.0,
+      "eval_runtime": 88.0723,
+      "eval_samples_per_second": 16.123,
+      "eval_steps_per_second": 2.021,
+      "step": 3260
+    },
+    {
+      "entropy": 0.1180689456872642,
+      "epoch": 7.905250452625227,
+      "grad_norm": 0.4754861891269684,
+      "learning_rate": 4.827266444161377e-05,
+      "loss": 0.09402655363082886,
+      "mean_token_accuracy": 0.9663361787796021,
+      "num_tokens": 7596998.0,
+      "step": 3280
+    },
+    {
+      "epoch": 7.905250452625227,
+      "eval_entropy": 0.261690996922134,
+      "eval_loss": 0.9120186567306519,
+      "eval_mean_token_accuracy": 0.8278995015647974,
+      "eval_num_tokens": 7596998.0,
+      "eval_runtime": 88.6573,
+      "eval_samples_per_second": 16.017,
+      "eval_steps_per_second": 2.008,
+      "step": 3280
+    },
+    {
+      "entropy": 0.1162164019420743,
+      "epoch": 7.953530476765239,
+      "grad_norm": 0.6203398108482361,
+      "learning_rate": 4.61757244135355e-05,
+      "loss": 0.08832768201828003,
+      "mean_token_accuracy": 0.9694220684468746,
+      "num_tokens": 7643288.0,
+      "step": 3300
+    },
+    {
+      "epoch": 7.953530476765239,
+      "eval_entropy": 0.26158678129817664,
+      "eval_loss": 0.9301361441612244,
+      "eval_mean_token_accuracy": 0.8269603885291668,
+      "eval_num_tokens": 7643288.0,
+      "eval_runtime": 88.7064,
+      "eval_samples_per_second": 16.008,
+      "eval_steps_per_second": 2.007,
+      "step": 3300
+    },
+    {
+      "entropy": 0.10877202204489089,
+      "epoch": 8.0,
+      "grad_norm": 3.058673143386841,
+      "learning_rate": 4.411895954922763e-05,
+      "loss": 0.09159615635871887,
+      "mean_token_accuracy": 0.9686878834451947,
+      "num_tokens": 7688608.0,
+      "step": 3320
+    },
+    {
+      "epoch": 8.0,
+      "eval_entropy": 0.2590438057197614,
+      "eval_loss": 0.9262065887451172,
+      "eval_mean_token_accuracy": 0.8277404546067956,
+      "eval_num_tokens": 7688608.0,
+      "eval_runtime": 87.7557,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 3320
+    },
+    {
+      "entropy": 0.09969092034734786,
+      "epoch": 8.048280024140013,
+      "grad_norm": 0.2561911642551422,
+      "learning_rate": 4.2102951888817644e-05,
+      "loss": 0.07228735685348511,
+      "mean_token_accuracy": 0.9743556551635265,
+      "num_tokens": 7732856.0,
+      "step": 3340
+    },
+    {
+      "epoch": 8.048280024140013,
+      "eval_entropy": 0.2462534454933713,
+      "eval_loss": 0.9755983352661133,
+      "eval_mean_token_accuracy": 0.827510209565752,
+      "eval_num_tokens": 7732856.0,
+      "eval_runtime": 88.0297,
+      "eval_samples_per_second": 16.131,
+      "eval_steps_per_second": 2.022,
+      "step": 3340
+    },
+    {
+      "entropy": 0.09477624730207027,
+      "epoch": 8.096560048280024,
+      "grad_norm": 0.3670356571674347,
+      "learning_rate": 4.012827193862633e-05,
+      "loss": 0.0686036467552185,
+      "mean_token_accuracy": 0.9743670664727688,
+      "num_tokens": 7778585.0,
+      "step": 3360
+    },
+    {
+      "epoch": 8.096560048280024,
+      "eval_entropy": 0.2440882951356052,
+      "eval_loss": 1.0118423700332642,
+      "eval_mean_token_accuracy": 0.8271241070849172,
+      "eval_num_tokens": 7778585.0,
+      "eval_runtime": 88.1364,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 3360
+    },
+    {
+      "entropy": 0.09417980071157217,
+      "epoch": 8.144840072420036,
+      "grad_norm": 0.5785111784934998,
+      "learning_rate": 3.819547850972133e-05,
+      "loss": 0.07100698947906495,
+      "mean_token_accuracy": 0.9737804099917412,
+      "num_tokens": 7824287.0,
+      "step": 3380
+    },
+    {
+      "epoch": 8.144840072420036,
+      "eval_entropy": 0.24603172860453637,
+      "eval_loss": 1.0021991729736328,
+      "eval_mean_token_accuracy": 0.827162122793412,
+      "eval_num_tokens": 7824287.0,
+      "eval_runtime": 88.4534,
+      "eval_samples_per_second": 16.054,
+      "eval_steps_per_second": 2.012,
+      "step": 3380
+    },
+    {
+      "entropy": 0.087222247896716,
+      "epoch": 8.193120096560047,
+      "grad_norm": 0.44848504662513733,
+      "learning_rate": 3.6305118559780393e-05,
+      "loss": 0.06325234174728393,
+      "mean_token_accuracy": 0.9769335232675076,
+      "num_tokens": 7874784.0,
+      "step": 3400
+    },
+    {
+      "epoch": 8.193120096560047,
+      "eval_entropy": 0.24544863905129807,
+      "eval_loss": 1.005008339881897,
+      "eval_mean_token_accuracy": 0.8273035240976998,
+      "eval_num_tokens": 7874784.0,
+      "eval_runtime": 88.6737,
+      "eval_samples_per_second": 16.014,
+      "eval_steps_per_second": 2.007,
+      "step": 3400
+    },
+    {
+      "entropy": 0.09037275062873959,
+      "epoch": 8.24140012070006,
+      "grad_norm": 0.23529155552387238,
+      "learning_rate": 3.4457727038308435e-05,
+      "loss": 0.06589569449424744,
+      "mean_token_accuracy": 0.9751470290124417,
+      "num_tokens": 7925240.0,
+      "step": 3420
+    },
+    {
+      "epoch": 8.24140012070006,
+      "eval_entropy": 0.24203423494368456,
+      "eval_loss": 1.0052824020385742,
+      "eval_mean_token_accuracy": 0.8275575503874361,
+      "eval_num_tokens": 7925240.0,
+      "eval_runtime": 88.3486,
+      "eval_samples_per_second": 16.073,
+      "eval_steps_per_second": 2.015,
+      "step": 3420
+    },
+    {
+      "entropy": 0.10373208015225828,
+      "epoch": 8.289680144840073,
+      "grad_norm": 0.24796408414840698,
+      "learning_rate": 3.2653826735253594e-05,
+      "loss": 0.07714771628379821,
+      "mean_token_accuracy": 0.9722891598939896,
+      "num_tokens": 7966685.0,
+      "step": 3440
+    },
+    {
+      "epoch": 8.289680144840073,
+      "eval_entropy": 0.23461157050025597,
+      "eval_loss": 1.012108325958252,
+      "eval_mean_token_accuracy": 0.82825900564033,
+      "eval_num_tokens": 7966685.0,
+      "eval_runtime": 88.1018,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 3440
+    },
+    {
+      "entropy": 0.08967954274266958,
+      "epoch": 8.337960168980084,
+      "grad_norm": 0.5314674973487854,
+      "learning_rate": 3.08939281330635e-05,
+      "loss": 0.06746266484260559,
+      "mean_token_accuracy": 0.9753133699297905,
+      "num_tokens": 8013656.0,
+      "step": 3460
+    },
+    {
+      "epoch": 8.337960168980084,
+      "eval_entropy": 0.23781998589467468,
+      "eval_loss": 1.0119913816452026,
+      "eval_mean_token_accuracy": 0.8273287157664139,
+      "eval_num_tokens": 8013656.0,
+      "eval_runtime": 88.6038,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3460
+    },
+    {
+      "entropy": 0.09280927330255509,
+      "epoch": 8.386240193120097,
+      "grad_norm": 0.30658936500549316,
+      "learning_rate": 2.9178529262224946e-05,
+      "loss": 0.06936421990394592,
+      "mean_token_accuracy": 0.9742105580866337,
+      "num_tokens": 8061108.0,
+      "step": 3480
+    },
+    {
+      "epoch": 8.386240193120097,
+      "eval_entropy": 0.23672426734747512,
+      "eval_loss": 1.0162041187286377,
+      "eval_mean_token_accuracy": 0.8278394281194451,
+      "eval_num_tokens": 8061108.0,
+      "eval_runtime": 88.6015,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 3480
+    },
+    {
+      "entropy": 0.09195122267119586,
+      "epoch": 8.43452021726011,
+      "grad_norm": 0.24512454867362976,
+      "learning_rate": 2.7508115560327256e-05,
+      "loss": 0.06996064782142639,
+      "mean_token_accuracy": 0.974104854464531,
+      "num_tokens": 8106924.0,
+      "step": 3500
+    },
+    {
+      "epoch": 8.43452021726011,
+      "eval_entropy": 0.23758245828781235,
+      "eval_loss": 1.0157747268676758,
+      "eval_mean_token_accuracy": 0.8278501579600773,
+      "eval_num_tokens": 8106924.0,
+      "eval_runtime": 88.5678,
+      "eval_samples_per_second": 16.033,
+      "eval_steps_per_second": 2.01,
+      "step": 3500
+    },
+    {
+      "entropy": 0.09518477590754629,
+      "epoch": 8.48280024140012,
+      "grad_norm": 0.2373228371143341,
+      "learning_rate": 2.5883159734688868e-05,
+      "loss": 0.06832797527313232,
+      "mean_token_accuracy": 0.9743806682527065,
+      "num_tokens": 8153070.0,
+      "step": 3520
+    },
+    {
+      "epoch": 8.48280024140012,
+      "eval_entropy": 0.2379308074210467,
+      "eval_loss": 1.0180530548095703,
+      "eval_mean_token_accuracy": 0.8282477052024241,
+      "eval_num_tokens": 8153070.0,
+      "eval_runtime": 88.4651,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 3520
+    },
+    {
+      "entropy": 0.0906269280705601,
+      "epoch": 8.531080265540133,
+      "grad_norm": 0.37602657079696655,
+      "learning_rate": 2.4304121628587173e-05,
+      "loss": 0.06921588778495788,
+      "mean_token_accuracy": 0.9745571479201317,
+      "num_tokens": 8200066.0,
+      "step": 3540
+    },
+    {
+      "epoch": 8.531080265540133,
+      "eval_entropy": 0.2357490769263064,
+      "eval_loss": 1.017139196395874,
+      "eval_mean_token_accuracy": 0.828713178299786,
+      "eval_num_tokens": 8200066.0,
+      "eval_runtime": 88.6424,
+      "eval_samples_per_second": 16.019,
+      "eval_steps_per_second": 2.008,
+      "step": 3540
+    },
+    {
+      "entropy": 0.09273334196768701,
+      "epoch": 8.579360289680144,
+      "grad_norm": 0.2978838086128235,
+      "learning_rate": 2.277144809112794e-05,
+      "loss": 0.06970887780189514,
+      "mean_token_accuracy": 0.9751374177634716,
+      "num_tokens": 8246157.0,
+      "step": 3560
+    },
+    {
+      "epoch": 8.579360289680144,
+      "eval_entropy": 0.23590933138065123,
+      "eval_loss": 1.019231915473938,
+      "eval_mean_token_accuracy": 0.8279163412163767,
+      "eval_num_tokens": 8246157.0,
+      "eval_runtime": 88.2691,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 3560
+    },
+    {
+      "entropy": 0.09854201921261847,
+      "epoch": 8.627640313820157,
+      "grad_norm": 0.5381243228912354,
+      "learning_rate": 2.1285572850792413e-05,
+      "loss": 0.07123501300811767,
+      "mean_token_accuracy": 0.9723988175392151,
+      "num_tokens": 8291752.0,
+      "step": 3580
+    },
+    {
+      "epoch": 8.627640313820157,
+      "eval_entropy": 0.23592024949494372,
+      "eval_loss": 1.020250678062439,
+      "eval_mean_token_accuracy": 0.8282466226079491,
+      "eval_num_tokens": 8291752.0,
+      "eval_runtime": 87.8939,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 3580
+    },
+    {
+      "entropy": 0.09092000788077712,
+      "epoch": 8.67592033796017,
+      "grad_norm": 0.29089561104774475,
+      "learning_rate": 1.9846916392697493e-05,
+      "loss": 0.06982720494270325,
+      "mean_token_accuracy": 0.9749995484948158,
+      "num_tokens": 8337651.0,
+      "step": 3600
+    },
+    {
+      "epoch": 8.67592033796017,
+      "eval_entropy": 0.23658913382318583,
+      "eval_loss": 1.019862413406372,
+      "eval_mean_token_accuracy": 0.8280127842104836,
+      "eval_num_tokens": 8337651.0,
+      "eval_runtime": 88.2855,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 3600
+    },
+    {
+      "entropy": 0.08708424470387399,
+      "epoch": 8.72420036210018,
+      "grad_norm": 0.29904839396476746,
+      "learning_rate": 1.8455885839603106e-05,
+      "loss": 0.06801370978355407,
+      "mean_token_accuracy": 0.9755801670253277,
+      "num_tokens": 8385330.0,
+      "step": 3620
+    },
+    {
+      "epoch": 8.72420036210018,
+      "eval_entropy": 0.23627035771862845,
+      "eval_loss": 1.0204569101333618,
+      "eval_mean_token_accuracy": 0.8279967894045155,
+      "eval_num_tokens": 8385330.0,
+      "eval_runtime": 88.1753,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 3620
+    },
+    {
+      "entropy": 0.0916255566291511,
+      "epoch": 8.772480386240193,
+      "grad_norm": 0.20571033656597137,
+      "learning_rate": 1.7112874836701502e-05,
+      "loss": 0.0692249894142151,
+      "mean_token_accuracy": 0.9745148062705994,
+      "num_tokens": 8432134.0,
+      "step": 3640
+    },
+    {
+      "epoch": 8.772480386240193,
+      "eval_entropy": 0.2340821772813797,
+      "eval_loss": 1.0212821960449219,
+      "eval_mean_token_accuracy": 0.828373622358515,
+      "eval_num_tokens": 8432134.0,
+      "eval_runtime": 87.2518,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 3640
+    },
+    {
+      "entropy": 0.0926868049427867,
+      "epoch": 8.820760410380204,
+      "grad_norm": 0.2618265450000763,
+      "learning_rate": 1.5818263440220203e-05,
+      "loss": 0.07338647246360779,
+      "mean_token_accuracy": 0.9724708944559097,
+      "num_tokens": 8476758.0,
+      "step": 3660
+    },
+    {
+      "epoch": 8.820760410380204,
+      "eval_entropy": 0.23637074575330433,
+      "eval_loss": 1.0186725854873657,
+      "eval_mean_token_accuracy": 0.8277707605549459,
+      "eval_num_tokens": 8476758.0,
+      "eval_runtime": 88.4695,
+      "eval_samples_per_second": 16.051,
+      "eval_steps_per_second": 2.012,
+      "step": 3660
+    },
+    {
+      "entropy": 0.09175782897509635,
+      "epoch": 8.869040434520217,
+      "grad_norm": 0.2889852821826935,
+      "learning_rate": 1.4572418009870604e-05,
+      "loss": 0.07066041231155396,
+      "mean_token_accuracy": 0.9725673109292984,
+      "num_tokens": 8523294.0,
+      "step": 3680
+    },
+    {
+      "epoch": 8.869040434520217,
+      "eval_entropy": 0.23945164404223473,
+      "eval_loss": 1.0145231485366821,
+      "eval_mean_token_accuracy": 0.8281202523895864,
+      "eval_num_tokens": 8523294.0,
+      "eval_runtime": 88.368,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 3680
+    },
+    {
+      "entropy": 0.09265611860901117,
+      "epoch": 8.91732045866023,
+      "grad_norm": 0.41092950105667114,
+      "learning_rate": 1.3375691105172625e-05,
+      "loss": 0.06975027322769164,
+      "mean_token_accuracy": 0.974636297672987,
+      "num_tokens": 8570358.0,
+      "step": 3700
+    },
+    {
+      "epoch": 8.91732045866023,
+      "eval_entropy": 0.239588919901446,
+      "eval_loss": 1.014059066772461,
+      "eval_mean_token_accuracy": 0.828028041660116,
+      "eval_num_tokens": 8570358.0,
+      "eval_runtime": 88.1039,
+      "eval_samples_per_second": 16.117,
+      "eval_steps_per_second": 2.02,
+      "step": 3700
+    },
+    {
+      "entropy": 0.0896465003490448,
+      "epoch": 8.965600482800241,
+      "grad_norm": 0.24000254273414612,
+      "learning_rate": 1.2228421385684305e-05,
+      "loss": 0.06743953824043274,
+      "mean_token_accuracy": 0.974006575345993,
+      "num_tokens": 8618246.0,
+      "step": 3720
+    },
+    {
+      "epoch": 8.965600482800241,
+      "eval_entropy": 0.2390461369846644,
+      "eval_loss": 1.011997938156128,
+      "eval_mean_token_accuracy": 0.8284533211354459,
+      "eval_num_tokens": 8618246.0,
+      "eval_runtime": 88.6202,
+      "eval_samples_per_second": 16.023,
+      "eval_steps_per_second": 2.009,
+      "step": 3720
+    },
+    {
+      "entropy": 0.09575109251520851,
+      "epoch": 9.012070006035003,
+      "grad_norm": 0.18088576197624207,
+      "learning_rate": 1.1130933515165429e-05,
+      "loss": 0.07295305132865906,
+      "mean_token_accuracy": 0.9739722857227573,
+      "num_tokens": 8661177.0,
+      "step": 3740
+    },
+    {
+      "epoch": 9.012070006035003,
+      "eval_entropy": 0.23728882387447892,
+      "eval_loss": 1.0134038925170898,
+      "eval_mean_token_accuracy": 0.8286508951294288,
+      "eval_num_tokens": 8661177.0,
+      "eval_runtime": 88.6434,
+      "eval_samples_per_second": 16.019,
+      "eval_steps_per_second": 2.008,
+      "step": 3740
+    },
+    {
+      "entropy": 0.08883201805874705,
+      "epoch": 9.060350030175016,
+      "grad_norm": 0.23120898008346558,
+      "learning_rate": 1.008353806970132e-05,
+      "loss": 0.06326142549514771,
+      "mean_token_accuracy": 0.9780020996928215,
+      "num_tokens": 8707602.0,
+      "step": 3760
+    },
+    {
+      "epoch": 9.060350030175016,
+      "eval_entropy": 0.23416905331142832,
+      "eval_loss": 1.0284100770950317,
+      "eval_mean_token_accuracy": 0.8284083213029283,
+      "eval_num_tokens": 8707602.0,
+      "eval_runtime": 88.2933,
+      "eval_samples_per_second": 16.083,
+      "eval_steps_per_second": 2.016,
+      "step": 3760
+    },
+    {
+      "entropy": 0.08861468276008963,
+      "epoch": 9.108630054315027,
+      "grad_norm": 0.17277656495571136,
+      "learning_rate": 9.086531449813489e-06,
+      "loss": 0.06727416515350342,
+      "mean_token_accuracy": 0.9760390736162663,
+      "num_tokens": 8751287.0,
+      "step": 3780
+    },
+    {
+      "epoch": 9.108630054315027,
+      "eval_entropy": 0.23370317386442355,
+      "eval_loss": 1.0381091833114624,
+      "eval_mean_token_accuracy": 0.8282668178670862,
+      "eval_num_tokens": 8751287.0,
+      "eval_runtime": 88.5738,
+      "eval_samples_per_second": 16.032,
+      "eval_steps_per_second": 2.01,
+      "step": 3780
+    },
+    {
+      "entropy": 0.09032954541034996,
+      "epoch": 9.15691007845504,
+      "grad_norm": 0.19481085240840912,
+      "learning_rate": 8.140195796581981e-06,
+      "loss": 0.06117786169052124,
+      "mean_token_accuracy": 0.9764454193413258,
+      "num_tokens": 8799982.0,
+      "step": 3800
+    },
+    {
+      "epoch": 9.15691007845504,
+      "eval_entropy": 0.23215936609868254,
+      "eval_loss": 1.0467767715454102,
+      "eval_mean_token_accuracy": 0.828405868806196,
+      "eval_num_tokens": 8799982.0,
+      "eval_runtime": 87.9551,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 3800
+    },
+    {
+      "entropy": 0.08824449330568314,
+      "epoch": 9.20519010259505,
+      "grad_norm": 0.20382793247699738,
+      "learning_rate": 7.2447989118025455e-06,
+      "loss": 0.06422553658485412,
+      "mean_token_accuracy": 0.9763121999800205,
+      "num_tokens": 8845536.0,
+      "step": 3820
+    },
+    {
+      "epoch": 9.20519010259505,
+      "eval_entropy": 0.23169661472352704,
+      "eval_loss": 1.0529043674468994,
+      "eval_mean_token_accuracy": 0.8282025610463003,
+      "eval_num_tokens": 8845536.0,
+      "eval_runtime": 88.5548,
+      "eval_samples_per_second": 16.035,
+      "eval_steps_per_second": 2.01,
+      "step": 3820
+    },
+    {
+      "entropy": 0.07768499720841646,
+      "epoch": 9.253470126735063,
+      "grad_norm": 0.11727919429540634,
+      "learning_rate": 6.400594182202084e-06,
+      "loss": 0.057279336452484134,
+      "mean_token_accuracy": 0.9795166745781898,
+      "num_tokens": 8896053.0,
+      "step": 3840
+    },
+    {
+      "epoch": 9.253470126735063,
+      "eval_entropy": 0.2301936373114586,
+      "eval_loss": 1.0576049089431763,
+      "eval_mean_token_accuracy": 0.8283417489421502,
+      "eval_num_tokens": 8896053.0,
+      "eval_runtime": 88.7854,
+      "eval_samples_per_second": 15.994,
+      "eval_steps_per_second": 2.005,
+      "step": 3840
+    },
+    {
+      "entropy": 0.09015200780704617,
+      "epoch": 9.301750150875076,
+      "grad_norm": 0.2089058756828308,
+      "learning_rate": 5.6078205077330705e-06,
+      "loss": 0.06626194715499878,
+      "mean_token_accuracy": 0.9747506938874722,
+      "num_tokens": 8940591.0,
+      "step": 3860
+    },
+    {
+      "epoch": 9.301750150875076,
+      "eval_entropy": 0.23024900676159377,
+      "eval_loss": 1.0599786043167114,
+      "eval_mean_token_accuracy": 0.8283051915383071,
+      "eval_num_tokens": 8940591.0,
+      "eval_runtime": 88.2963,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3860
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.7804174177593754e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3880/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3880/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3880/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3880/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3880/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3880/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3880/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3880/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3880/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3880/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3880/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3880/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3880/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3880/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..97fa7ff9b2780a2d7b36bcd032956628c61378e7
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3880/trainer_state.json
@@ -0,0 +1,4108 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 9.350030175015087,
+  "eval_steps": 20,
+  "global_step": 3880,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    },
+    {
+      "entropy": 0.1326121488586068,
+      "epoch": 7.181050090525045,
+      "grad_norm": 0.4843117296695709,
+      "learning_rate": 8.411870386507724e-05,
+      "loss": 0.09537842869758606,
+      "mean_token_accuracy": 0.9645151920616627,
+      "num_tokens": 6898444.0,
+      "step": 2980
+    },
+    {
+      "epoch": 7.181050090525045,
+      "eval_entropy": 0.27999477783280813,
+      "eval_loss": 0.9031037092208862,
+      "eval_mean_token_accuracy": 0.8261882204018282,
+      "eval_num_tokens": 6898444.0,
+      "eval_runtime": 88.5602,
+      "eval_samples_per_second": 16.034,
+      "eval_steps_per_second": 2.01,
+      "step": 2980
+    },
+    {
+      "entropy": 0.1156751093454659,
+      "epoch": 7.229330114665057,
+      "grad_norm": 0.5297922492027283,
+      "learning_rate": 8.149629351430326e-05,
+      "loss": 0.09203402996063233,
+      "mean_token_accuracy": 0.9689733698964119,
+      "num_tokens": 6947106.0,
+      "step": 3000
+    },
+    {
+      "epoch": 7.229330114665057,
+      "eval_entropy": 0.2740225845508361,
+      "eval_loss": 0.8876440525054932,
+      "eval_mean_token_accuracy": 0.8265116857678703,
+      "eval_num_tokens": 6947106.0,
+      "eval_runtime": 87.7971,
+      "eval_samples_per_second": 16.174,
+      "eval_steps_per_second": 2.027,
+      "step": 3000
+    },
+    {
+      "entropy": 0.12116114404052496,
+      "epoch": 7.27761013880507,
+      "grad_norm": 0.5467398166656494,
+      "learning_rate": 7.890406302404618e-05,
+      "loss": 0.0933056652545929,
+      "mean_token_accuracy": 0.9676455594599247,
+      "num_tokens": 6992784.0,
+      "step": 3020
+    },
+    {
+      "epoch": 7.27761013880507,
+      "eval_entropy": 0.27238342793804876,
+      "eval_loss": 0.9166117310523987,
+      "eval_mean_token_accuracy": 0.825499015242866,
+      "eval_num_tokens": 6992784.0,
+      "eval_runtime": 88.3276,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 3020
+    },
+    {
+      "entropy": 0.11715119928121567,
+      "epoch": 7.325890162945082,
+      "grad_norm": 0.4976702332496643,
+      "learning_rate": 7.634274596487176e-05,
+      "loss": 0.09449659585952759,
+      "mean_token_accuracy": 0.966506478190422,
+      "num_tokens": 7040482.0,
+      "step": 3040
+    },
+    {
+      "epoch": 7.325890162945082,
+      "eval_entropy": 0.2788975556077582,
+      "eval_loss": 0.8833065629005432,
+      "eval_mean_token_accuracy": 0.8269920596915684,
+      "eval_num_tokens": 7040482.0,
+      "eval_runtime": 88.9016,
+      "eval_samples_per_second": 15.973,
+      "eval_steps_per_second": 2.002,
+      "step": 3040
+    },
+    {
+      "entropy": 0.12565911412239075,
+      "epoch": 7.3741701870850935,
+      "grad_norm": 0.5728679895401001,
+      "learning_rate": 7.381306715921025e-05,
+      "loss": 0.09626244902610778,
+      "mean_token_accuracy": 0.9661444149911403,
+      "num_tokens": 7086261.0,
+      "step": 3060
+    },
+    {
+      "epoch": 7.3741701870850935,
+      "eval_entropy": 0.2762492011939542,
+      "eval_loss": 0.8961160778999329,
+      "eval_mean_token_accuracy": 0.826369910092836,
+      "eval_num_tokens": 7086261.0,
+      "eval_runtime": 88.5204,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 3060
+    },
+    {
+      "entropy": 0.12258567921817302,
+      "epoch": 7.422450211225105,
+      "grad_norm": 0.468985915184021,
+      "learning_rate": 7.131574247624031e-05,
+      "loss": 0.09120266437530518,
+      "mean_token_accuracy": 0.9674938447773457,
+      "num_tokens": 7133247.0,
+      "step": 3080
+    },
+    {
+      "epoch": 7.422450211225105,
+      "eval_entropy": 0.2688512086366,
+      "eval_loss": 0.9048654437065125,
+      "eval_mean_token_accuracy": 0.8265582302982888,
+      "eval_num_tokens": 7133247.0,
+      "eval_runtime": 88.164,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 3080
+    },
+    {
+      "entropy": 0.12070536343380808,
+      "epoch": 7.470730235365117,
+      "grad_norm": 0.5310841798782349,
+      "learning_rate": 6.885147862930688e-05,
+      "loss": 0.09386247992515565,
+      "mean_token_accuracy": 0.96664222702384,
+      "num_tokens": 7177810.0,
+      "step": 3100
+    },
+    {
+      "epoch": 7.470730235365117,
+      "eval_entropy": 0.26289632874593305,
+      "eval_loss": 0.902837336063385,
+      "eval_mean_token_accuracy": 0.828052184554968,
+      "eval_num_tokens": 7177810.0,
+      "eval_runtime": 88.1201,
+      "eval_samples_per_second": 16.114,
+      "eval_steps_per_second": 2.02,
+      "step": 3100
+    },
+    {
+      "entropy": 0.11322665181942285,
+      "epoch": 7.51901025950513,
+      "grad_norm": 0.6846067905426025,
+      "learning_rate": 6.642097297592908e-05,
+      "loss": 0.08772538304328918,
+      "mean_token_accuracy": 0.9691362135112286,
+      "num_tokens": 7225302.0,
+      "step": 3120
+    },
+    {
+      "epoch": 7.51901025950513,
+      "eval_entropy": 0.2670605613776807,
+      "eval_loss": 0.9130566120147705,
+      "eval_mean_token_accuracy": 0.826366903741708,
+      "eval_num_tokens": 7225302.0,
+      "eval_runtime": 88.6035,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3120
+    },
+    {
+      "entropy": 0.10985926580615342,
+      "epoch": 7.567290283645142,
+      "grad_norm": 0.4799748361110687,
+      "learning_rate": 6.402491332045683e-05,
+      "loss": 0.08672296404838561,
+      "mean_token_accuracy": 0.9698824048042297,
+      "num_tokens": 7274881.0,
+      "step": 3140
+    },
+    {
+      "epoch": 7.567290283645142,
+      "eval_entropy": 0.26657010866015146,
+      "eval_loss": 0.9056380987167358,
+      "eval_mean_token_accuracy": 0.8272326461384806,
+      "eval_num_tokens": 7274881.0,
+      "eval_runtime": 88.2967,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3140
+    },
+    {
+      "entropy": 0.11890473761595785,
+      "epoch": 7.615570307785154,
+      "grad_norm": 0.4081912934780121,
+      "learning_rate": 6.166397771942987e-05,
+      "loss": 0.09396827220916748,
+      "mean_token_accuracy": 0.967512384802103,
+      "num_tokens": 7321163.0,
+      "step": 3160
+    },
+    {
+      "epoch": 7.615570307785154,
+      "eval_entropy": 0.26364776575833226,
+      "eval_loss": 0.9204733967781067,
+      "eval_mean_token_accuracy": 0.827419729714983,
+      "eval_num_tokens": 7321163.0,
+      "eval_runtime": 88.3441,
+      "eval_samples_per_second": 16.074,
+      "eval_steps_per_second": 2.015,
+      "step": 3160
+    },
+    {
+      "entropy": 0.12209134344011545,
+      "epoch": 7.663850331925166,
+      "grad_norm": 0.5159774422645569,
+      "learning_rate": 5.93388342896962e-05,
+      "loss": 0.09292978644371033,
+      "mean_token_accuracy": 0.9658421002328396,
+      "num_tokens": 7366217.0,
+      "step": 3180
+    },
+    {
+      "epoch": 7.663850331925166,
+      "eval_entropy": 0.2714110068725736,
+      "eval_loss": 0.9143092036247253,
+      "eval_mean_token_accuracy": 0.8262251099843657,
+      "eval_num_tokens": 7366217.0,
+      "eval_runtime": 87.9635,
+      "eval_samples_per_second": 16.143,
+      "eval_steps_per_second": 2.024,
+      "step": 3180
+    },
+    {
+      "entropy": 0.11955609126016498,
+      "epoch": 7.712130356065178,
+      "grad_norm": 0.7005976438522339,
+      "learning_rate": 5.705014101934252e-05,
+      "loss": 0.09319776892662049,
+      "mean_token_accuracy": 0.9674513377249241,
+      "num_tokens": 7410887.0,
+      "step": 3200
+    },
+    {
+      "epoch": 7.712130356065178,
+      "eval_entropy": 0.2626291513442993,
+      "eval_loss": 0.91884446144104,
+      "eval_mean_token_accuracy": 0.827563955207889,
+      "eval_num_tokens": 7410887.0,
+      "eval_runtime": 87.8849,
+      "eval_samples_per_second": 16.157,
+      "eval_steps_per_second": 2.025,
+      "step": 3200
+    },
+    {
+      "entropy": 0.11767532899975777,
+      "epoch": 7.76041038020519,
+      "grad_norm": 0.4675362706184387,
+      "learning_rate": 5.479854558149187e-05,
+      "loss": 0.09038087725639343,
+      "mean_token_accuracy": 0.9669680051505566,
+      "num_tokens": 7457578.0,
+      "step": 3220
+    },
+    {
+      "epoch": 7.76041038020519,
+      "eval_entropy": 0.26581037203582486,
+      "eval_loss": 0.9146883487701416,
+      "eval_mean_token_accuracy": 0.8270345812433222,
+      "eval_num_tokens": 7457578.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 3220
+    },
+    {
+      "entropy": 0.11383061073720455,
+      "epoch": 7.808690404345202,
+      "grad_norm": 0.541247546672821,
+      "learning_rate": 5.258468515101918e-05,
+      "loss": 0.08942405581474304,
+      "mean_token_accuracy": 0.968398705124855,
+      "num_tokens": 7505761.0,
+      "step": 3240
+    },
+    {
+      "epoch": 7.808690404345202,
+      "eval_entropy": 0.26687998960861997,
+      "eval_loss": 0.9037085175514221,
+      "eval_mean_token_accuracy": 0.8274977863504646,
+      "eval_num_tokens": 7505761.0,
+      "eval_runtime": 88.5163,
+      "eval_samples_per_second": 16.042,
+      "eval_steps_per_second": 2.011,
+      "step": 3240
+    },
+    {
+      "entropy": 0.12004058873280883,
+      "epoch": 7.856970428485214,
+      "grad_norm": 0.5289552211761475,
+      "learning_rate": 5.040918622423899e-05,
+      "loss": 0.09064805507659912,
+      "mean_token_accuracy": 0.9673330299556255,
+      "num_tokens": 7552557.0,
+      "step": 3260
+    },
+    {
+      "epoch": 7.856970428485214,
+      "eval_entropy": 0.26288379251622085,
+      "eval_loss": 0.9146178364753723,
+      "eval_mean_token_accuracy": 0.8276662448149049,
+      "eval_num_tokens": 7552557.0,
+      "eval_runtime": 88.0723,
+      "eval_samples_per_second": 16.123,
+      "eval_steps_per_second": 2.021,
+      "step": 3260
+    },
+    {
+      "entropy": 0.1180689456872642,
+      "epoch": 7.905250452625227,
+      "grad_norm": 0.4754861891269684,
+      "learning_rate": 4.827266444161377e-05,
+      "loss": 0.09402655363082886,
+      "mean_token_accuracy": 0.9663361787796021,
+      "num_tokens": 7596998.0,
+      "step": 3280
+    },
+    {
+      "epoch": 7.905250452625227,
+      "eval_entropy": 0.261690996922134,
+      "eval_loss": 0.9120186567306519,
+      "eval_mean_token_accuracy": 0.8278995015647974,
+      "eval_num_tokens": 7596998.0,
+      "eval_runtime": 88.6573,
+      "eval_samples_per_second": 16.017,
+      "eval_steps_per_second": 2.008,
+      "step": 3280
+    },
+    {
+      "entropy": 0.1162164019420743,
+      "epoch": 7.953530476765239,
+      "grad_norm": 0.6203398108482361,
+      "learning_rate": 4.61757244135355e-05,
+      "loss": 0.08832768201828003,
+      "mean_token_accuracy": 0.9694220684468746,
+      "num_tokens": 7643288.0,
+      "step": 3300
+    },
+    {
+      "epoch": 7.953530476765239,
+      "eval_entropy": 0.26158678129817664,
+      "eval_loss": 0.9301361441612244,
+      "eval_mean_token_accuracy": 0.8269603885291668,
+      "eval_num_tokens": 7643288.0,
+      "eval_runtime": 88.7064,
+      "eval_samples_per_second": 16.008,
+      "eval_steps_per_second": 2.007,
+      "step": 3300
+    },
+    {
+      "entropy": 0.10877202204489089,
+      "epoch": 8.0,
+      "grad_norm": 3.058673143386841,
+      "learning_rate": 4.411895954922763e-05,
+      "loss": 0.09159615635871887,
+      "mean_token_accuracy": 0.9686878834451947,
+      "num_tokens": 7688608.0,
+      "step": 3320
+    },
+    {
+      "epoch": 8.0,
+      "eval_entropy": 0.2590438057197614,
+      "eval_loss": 0.9262065887451172,
+      "eval_mean_token_accuracy": 0.8277404546067956,
+      "eval_num_tokens": 7688608.0,
+      "eval_runtime": 87.7557,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 3320
+    },
+    {
+      "entropy": 0.09969092034734786,
+      "epoch": 8.048280024140013,
+      "grad_norm": 0.2561911642551422,
+      "learning_rate": 4.2102951888817644e-05,
+      "loss": 0.07228735685348511,
+      "mean_token_accuracy": 0.9743556551635265,
+      "num_tokens": 7732856.0,
+      "step": 3340
+    },
+    {
+      "epoch": 8.048280024140013,
+      "eval_entropy": 0.2462534454933713,
+      "eval_loss": 0.9755983352661133,
+      "eval_mean_token_accuracy": 0.827510209565752,
+      "eval_num_tokens": 7732856.0,
+      "eval_runtime": 88.0297,
+      "eval_samples_per_second": 16.131,
+      "eval_steps_per_second": 2.022,
+      "step": 3340
+    },
+    {
+      "entropy": 0.09477624730207027,
+      "epoch": 8.096560048280024,
+      "grad_norm": 0.3670356571674347,
+      "learning_rate": 4.012827193862633e-05,
+      "loss": 0.0686036467552185,
+      "mean_token_accuracy": 0.9743670664727688,
+      "num_tokens": 7778585.0,
+      "step": 3360
+    },
+    {
+      "epoch": 8.096560048280024,
+      "eval_entropy": 0.2440882951356052,
+      "eval_loss": 1.0118423700332642,
+      "eval_mean_token_accuracy": 0.8271241070849172,
+      "eval_num_tokens": 7778585.0,
+      "eval_runtime": 88.1364,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 3360
+    },
+    {
+      "entropy": 0.09417980071157217,
+      "epoch": 8.144840072420036,
+      "grad_norm": 0.5785111784934998,
+      "learning_rate": 3.819547850972133e-05,
+      "loss": 0.07100698947906495,
+      "mean_token_accuracy": 0.9737804099917412,
+      "num_tokens": 7824287.0,
+      "step": 3380
+    },
+    {
+      "epoch": 8.144840072420036,
+      "eval_entropy": 0.24603172860453637,
+      "eval_loss": 1.0021991729736328,
+      "eval_mean_token_accuracy": 0.827162122793412,
+      "eval_num_tokens": 7824287.0,
+      "eval_runtime": 88.4534,
+      "eval_samples_per_second": 16.054,
+      "eval_steps_per_second": 2.012,
+      "step": 3380
+    },
+    {
+      "entropy": 0.087222247896716,
+      "epoch": 8.193120096560047,
+      "grad_norm": 0.44848504662513733,
+      "learning_rate": 3.6305118559780393e-05,
+      "loss": 0.06325234174728393,
+      "mean_token_accuracy": 0.9769335232675076,
+      "num_tokens": 7874784.0,
+      "step": 3400
+    },
+    {
+      "epoch": 8.193120096560047,
+      "eval_entropy": 0.24544863905129807,
+      "eval_loss": 1.005008339881897,
+      "eval_mean_token_accuracy": 0.8273035240976998,
+      "eval_num_tokens": 7874784.0,
+      "eval_runtime": 88.6737,
+      "eval_samples_per_second": 16.014,
+      "eval_steps_per_second": 2.007,
+      "step": 3400
+    },
+    {
+      "entropy": 0.09037275062873959,
+      "epoch": 8.24140012070006,
+      "grad_norm": 0.23529155552387238,
+      "learning_rate": 3.4457727038308435e-05,
+      "loss": 0.06589569449424744,
+      "mean_token_accuracy": 0.9751470290124417,
+      "num_tokens": 7925240.0,
+      "step": 3420
+    },
+    {
+      "epoch": 8.24140012070006,
+      "eval_entropy": 0.24203423494368456,
+      "eval_loss": 1.0052824020385742,
+      "eval_mean_token_accuracy": 0.8275575503874361,
+      "eval_num_tokens": 7925240.0,
+      "eval_runtime": 88.3486,
+      "eval_samples_per_second": 16.073,
+      "eval_steps_per_second": 2.015,
+      "step": 3420
+    },
+    {
+      "entropy": 0.10373208015225828,
+      "epoch": 8.289680144840073,
+      "grad_norm": 0.24796408414840698,
+      "learning_rate": 3.2653826735253594e-05,
+      "loss": 0.07714771628379821,
+      "mean_token_accuracy": 0.9722891598939896,
+      "num_tokens": 7966685.0,
+      "step": 3440
+    },
+    {
+      "epoch": 8.289680144840073,
+      "eval_entropy": 0.23461157050025597,
+      "eval_loss": 1.012108325958252,
+      "eval_mean_token_accuracy": 0.82825900564033,
+      "eval_num_tokens": 7966685.0,
+      "eval_runtime": 88.1018,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 3440
+    },
+    {
+      "entropy": 0.08967954274266958,
+      "epoch": 8.337960168980084,
+      "grad_norm": 0.5314674973487854,
+      "learning_rate": 3.08939281330635e-05,
+      "loss": 0.06746266484260559,
+      "mean_token_accuracy": 0.9753133699297905,
+      "num_tokens": 8013656.0,
+      "step": 3460
+    },
+    {
+      "epoch": 8.337960168980084,
+      "eval_entropy": 0.23781998589467468,
+      "eval_loss": 1.0119913816452026,
+      "eval_mean_token_accuracy": 0.8273287157664139,
+      "eval_num_tokens": 8013656.0,
+      "eval_runtime": 88.6038,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3460
+    },
+    {
+      "entropy": 0.09280927330255509,
+      "epoch": 8.386240193120097,
+      "grad_norm": 0.30658936500549316,
+      "learning_rate": 2.9178529262224946e-05,
+      "loss": 0.06936421990394592,
+      "mean_token_accuracy": 0.9742105580866337,
+      "num_tokens": 8061108.0,
+      "step": 3480
+    },
+    {
+      "epoch": 8.386240193120097,
+      "eval_entropy": 0.23672426734747512,
+      "eval_loss": 1.0162041187286377,
+      "eval_mean_token_accuracy": 0.8278394281194451,
+      "eval_num_tokens": 8061108.0,
+      "eval_runtime": 88.6015,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 3480
+    },
+    {
+      "entropy": 0.09195122267119586,
+      "epoch": 8.43452021726011,
+      "grad_norm": 0.24512454867362976,
+      "learning_rate": 2.7508115560327256e-05,
+      "loss": 0.06996064782142639,
+      "mean_token_accuracy": 0.974104854464531,
+      "num_tokens": 8106924.0,
+      "step": 3500
+    },
+    {
+      "epoch": 8.43452021726011,
+      "eval_entropy": 0.23758245828781235,
+      "eval_loss": 1.0157747268676758,
+      "eval_mean_token_accuracy": 0.8278501579600773,
+      "eval_num_tokens": 8106924.0,
+      "eval_runtime": 88.5678,
+      "eval_samples_per_second": 16.033,
+      "eval_steps_per_second": 2.01,
+      "step": 3500
+    },
+    {
+      "entropy": 0.09518477590754629,
+      "epoch": 8.48280024140012,
+      "grad_norm": 0.2373228371143341,
+      "learning_rate": 2.5883159734688868e-05,
+      "loss": 0.06832797527313232,
+      "mean_token_accuracy": 0.9743806682527065,
+      "num_tokens": 8153070.0,
+      "step": 3520
+    },
+    {
+      "epoch": 8.48280024140012,
+      "eval_entropy": 0.2379308074210467,
+      "eval_loss": 1.0180530548095703,
+      "eval_mean_token_accuracy": 0.8282477052024241,
+      "eval_num_tokens": 8153070.0,
+      "eval_runtime": 88.4651,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 3520
+    },
+    {
+      "entropy": 0.0906269280705601,
+      "epoch": 8.531080265540133,
+      "grad_norm": 0.37602657079696655,
+      "learning_rate": 2.4304121628587173e-05,
+      "loss": 0.06921588778495788,
+      "mean_token_accuracy": 0.9745571479201317,
+      "num_tokens": 8200066.0,
+      "step": 3540
+    },
+    {
+      "epoch": 8.531080265540133,
+      "eval_entropy": 0.2357490769263064,
+      "eval_loss": 1.017139196395874,
+      "eval_mean_token_accuracy": 0.828713178299786,
+      "eval_num_tokens": 8200066.0,
+      "eval_runtime": 88.6424,
+      "eval_samples_per_second": 16.019,
+      "eval_steps_per_second": 2.008,
+      "step": 3540
+    },
+    {
+      "entropy": 0.09273334196768701,
+      "epoch": 8.579360289680144,
+      "grad_norm": 0.2978838086128235,
+      "learning_rate": 2.277144809112794e-05,
+      "loss": 0.06970887780189514,
+      "mean_token_accuracy": 0.9751374177634716,
+      "num_tokens": 8246157.0,
+      "step": 3560
+    },
+    {
+      "epoch": 8.579360289680144,
+      "eval_entropy": 0.23590933138065123,
+      "eval_loss": 1.019231915473938,
+      "eval_mean_token_accuracy": 0.8279163412163767,
+      "eval_num_tokens": 8246157.0,
+      "eval_runtime": 88.2691,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 3560
+    },
+    {
+      "entropy": 0.09854201921261847,
+      "epoch": 8.627640313820157,
+      "grad_norm": 0.5381243228912354,
+      "learning_rate": 2.1285572850792413e-05,
+      "loss": 0.07123501300811767,
+      "mean_token_accuracy": 0.9723988175392151,
+      "num_tokens": 8291752.0,
+      "step": 3580
+    },
+    {
+      "epoch": 8.627640313820157,
+      "eval_entropy": 0.23592024949494372,
+      "eval_loss": 1.020250678062439,
+      "eval_mean_token_accuracy": 0.8282466226079491,
+      "eval_num_tokens": 8291752.0,
+      "eval_runtime": 87.8939,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 3580
+    },
+    {
+      "entropy": 0.09092000788077712,
+      "epoch": 8.67592033796017,
+      "grad_norm": 0.29089561104774475,
+      "learning_rate": 1.9846916392697493e-05,
+      "loss": 0.06982720494270325,
+      "mean_token_accuracy": 0.9749995484948158,
+      "num_tokens": 8337651.0,
+      "step": 3600
+    },
+    {
+      "epoch": 8.67592033796017,
+      "eval_entropy": 0.23658913382318583,
+      "eval_loss": 1.019862413406372,
+      "eval_mean_token_accuracy": 0.8280127842104836,
+      "eval_num_tokens": 8337651.0,
+      "eval_runtime": 88.2855,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 3600
+    },
+    {
+      "entropy": 0.08708424470387399,
+      "epoch": 8.72420036210018,
+      "grad_norm": 0.29904839396476746,
+      "learning_rate": 1.8455885839603106e-05,
+      "loss": 0.06801370978355407,
+      "mean_token_accuracy": 0.9755801670253277,
+      "num_tokens": 8385330.0,
+      "step": 3620
+    },
+    {
+      "epoch": 8.72420036210018,
+      "eval_entropy": 0.23627035771862845,
+      "eval_loss": 1.0204569101333618,
+      "eval_mean_token_accuracy": 0.8279967894045155,
+      "eval_num_tokens": 8385330.0,
+      "eval_runtime": 88.1753,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 3620
+    },
+    {
+      "entropy": 0.0916255566291511,
+      "epoch": 8.772480386240193,
+      "grad_norm": 0.20571033656597137,
+      "learning_rate": 1.7112874836701502e-05,
+      "loss": 0.0692249894142151,
+      "mean_token_accuracy": 0.9745148062705994,
+      "num_tokens": 8432134.0,
+      "step": 3640
+    },
+    {
+      "epoch": 8.772480386240193,
+      "eval_entropy": 0.2340821772813797,
+      "eval_loss": 1.0212821960449219,
+      "eval_mean_token_accuracy": 0.828373622358515,
+      "eval_num_tokens": 8432134.0,
+      "eval_runtime": 87.2518,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 3640
+    },
+    {
+      "entropy": 0.0926868049427867,
+      "epoch": 8.820760410380204,
+      "grad_norm": 0.2618265450000763,
+      "learning_rate": 1.5818263440220203e-05,
+      "loss": 0.07338647246360779,
+      "mean_token_accuracy": 0.9724708944559097,
+      "num_tokens": 8476758.0,
+      "step": 3660
+    },
+    {
+      "epoch": 8.820760410380204,
+      "eval_entropy": 0.23637074575330433,
+      "eval_loss": 1.0186725854873657,
+      "eval_mean_token_accuracy": 0.8277707605549459,
+      "eval_num_tokens": 8476758.0,
+      "eval_runtime": 88.4695,
+      "eval_samples_per_second": 16.051,
+      "eval_steps_per_second": 2.012,
+      "step": 3660
+    },
+    {
+      "entropy": 0.09175782897509635,
+      "epoch": 8.869040434520217,
+      "grad_norm": 0.2889852821826935,
+      "learning_rate": 1.4572418009870604e-05,
+      "loss": 0.07066041231155396,
+      "mean_token_accuracy": 0.9725673109292984,
+      "num_tokens": 8523294.0,
+      "step": 3680
+    },
+    {
+      "epoch": 8.869040434520217,
+      "eval_entropy": 0.23945164404223473,
+      "eval_loss": 1.0145231485366821,
+      "eval_mean_token_accuracy": 0.8281202523895864,
+      "eval_num_tokens": 8523294.0,
+      "eval_runtime": 88.368,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 3680
+    },
+    {
+      "entropy": 0.09265611860901117,
+      "epoch": 8.91732045866023,
+      "grad_norm": 0.41092950105667114,
+      "learning_rate": 1.3375691105172625e-05,
+      "loss": 0.06975027322769164,
+      "mean_token_accuracy": 0.974636297672987,
+      "num_tokens": 8570358.0,
+      "step": 3700
+    },
+    {
+      "epoch": 8.91732045866023,
+      "eval_entropy": 0.239588919901446,
+      "eval_loss": 1.014059066772461,
+      "eval_mean_token_accuracy": 0.828028041660116,
+      "eval_num_tokens": 8570358.0,
+      "eval_runtime": 88.1039,
+      "eval_samples_per_second": 16.117,
+      "eval_steps_per_second": 2.02,
+      "step": 3700
+    },
+    {
+      "entropy": 0.0896465003490448,
+      "epoch": 8.965600482800241,
+      "grad_norm": 0.24000254273414612,
+      "learning_rate": 1.2228421385684305e-05,
+      "loss": 0.06743953824043274,
+      "mean_token_accuracy": 0.974006575345993,
+      "num_tokens": 8618246.0,
+      "step": 3720
+    },
+    {
+      "epoch": 8.965600482800241,
+      "eval_entropy": 0.2390461369846644,
+      "eval_loss": 1.011997938156128,
+      "eval_mean_token_accuracy": 0.8284533211354459,
+      "eval_num_tokens": 8618246.0,
+      "eval_runtime": 88.6202,
+      "eval_samples_per_second": 16.023,
+      "eval_steps_per_second": 2.009,
+      "step": 3720
+    },
+    {
+      "entropy": 0.09575109251520851,
+      "epoch": 9.012070006035003,
+      "grad_norm": 0.18088576197624207,
+      "learning_rate": 1.1130933515165429e-05,
+      "loss": 0.07295305132865906,
+      "mean_token_accuracy": 0.9739722857227573,
+      "num_tokens": 8661177.0,
+      "step": 3740
+    },
+    {
+      "epoch": 9.012070006035003,
+      "eval_entropy": 0.23728882387447892,
+      "eval_loss": 1.0134038925170898,
+      "eval_mean_token_accuracy": 0.8286508951294288,
+      "eval_num_tokens": 8661177.0,
+      "eval_runtime": 88.6434,
+      "eval_samples_per_second": 16.019,
+      "eval_steps_per_second": 2.008,
+      "step": 3740
+    },
+    {
+      "entropy": 0.08883201805874705,
+      "epoch": 9.060350030175016,
+      "grad_norm": 0.23120898008346558,
+      "learning_rate": 1.008353806970132e-05,
+      "loss": 0.06326142549514771,
+      "mean_token_accuracy": 0.9780020996928215,
+      "num_tokens": 8707602.0,
+      "step": 3760
+    },
+    {
+      "epoch": 9.060350030175016,
+      "eval_entropy": 0.23416905331142832,
+      "eval_loss": 1.0284100770950317,
+      "eval_mean_token_accuracy": 0.8284083213029283,
+      "eval_num_tokens": 8707602.0,
+      "eval_runtime": 88.2933,
+      "eval_samples_per_second": 16.083,
+      "eval_steps_per_second": 2.016,
+      "step": 3760
+    },
+    {
+      "entropy": 0.08861468276008963,
+      "epoch": 9.108630054315027,
+      "grad_norm": 0.17277656495571136,
+      "learning_rate": 9.086531449813489e-06,
+      "loss": 0.06727416515350342,
+      "mean_token_accuracy": 0.9760390736162663,
+      "num_tokens": 8751287.0,
+      "step": 3780
+    },
+    {
+      "epoch": 9.108630054315027,
+      "eval_entropy": 0.23370317386442355,
+      "eval_loss": 1.0381091833114624,
+      "eval_mean_token_accuracy": 0.8282668178670862,
+      "eval_num_tokens": 8751287.0,
+      "eval_runtime": 88.5738,
+      "eval_samples_per_second": 16.032,
+      "eval_steps_per_second": 2.01,
+      "step": 3780
+    },
+    {
+      "entropy": 0.09032954541034996,
+      "epoch": 9.15691007845504,
+      "grad_norm": 0.19481085240840912,
+      "learning_rate": 8.140195796581981e-06,
+      "loss": 0.06117786169052124,
+      "mean_token_accuracy": 0.9764454193413258,
+      "num_tokens": 8799982.0,
+      "step": 3800
+    },
+    {
+      "epoch": 9.15691007845504,
+      "eval_entropy": 0.23215936609868254,
+      "eval_loss": 1.0467767715454102,
+      "eval_mean_token_accuracy": 0.828405868806196,
+      "eval_num_tokens": 8799982.0,
+      "eval_runtime": 87.9551,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 3800
+    },
+    {
+      "entropy": 0.08824449330568314,
+      "epoch": 9.20519010259505,
+      "grad_norm": 0.20382793247699738,
+      "learning_rate": 7.2447989118025455e-06,
+      "loss": 0.06422553658485412,
+      "mean_token_accuracy": 0.9763121999800205,
+      "num_tokens": 8845536.0,
+      "step": 3820
+    },
+    {
+      "epoch": 9.20519010259505,
+      "eval_entropy": 0.23169661472352704,
+      "eval_loss": 1.0529043674468994,
+      "eval_mean_token_accuracy": 0.8282025610463003,
+      "eval_num_tokens": 8845536.0,
+      "eval_runtime": 88.5548,
+      "eval_samples_per_second": 16.035,
+      "eval_steps_per_second": 2.01,
+      "step": 3820
+    },
+    {
+      "entropy": 0.07768499720841646,
+      "epoch": 9.253470126735063,
+      "grad_norm": 0.11727919429540634,
+      "learning_rate": 6.400594182202084e-06,
+      "loss": 0.057279336452484134,
+      "mean_token_accuracy": 0.9795166745781898,
+      "num_tokens": 8896053.0,
+      "step": 3840
+    },
+    {
+      "epoch": 9.253470126735063,
+      "eval_entropy": 0.2301936373114586,
+      "eval_loss": 1.0576049089431763,
+      "eval_mean_token_accuracy": 0.8283417489421502,
+      "eval_num_tokens": 8896053.0,
+      "eval_runtime": 88.7854,
+      "eval_samples_per_second": 15.994,
+      "eval_steps_per_second": 2.005,
+      "step": 3840
+    },
+    {
+      "entropy": 0.09015200780704617,
+      "epoch": 9.301750150875076,
+      "grad_norm": 0.2089058756828308,
+      "learning_rate": 5.6078205077330705e-06,
+      "loss": 0.06626194715499878,
+      "mean_token_accuracy": 0.9747506938874722,
+      "num_tokens": 8940591.0,
+      "step": 3860
+    },
+    {
+      "epoch": 9.301750150875076,
+      "eval_entropy": 0.23024900676159377,
+      "eval_loss": 1.0599786043167114,
+      "eval_mean_token_accuracy": 0.8283051915383071,
+      "eval_num_tokens": 8940591.0,
+      "eval_runtime": 88.2963,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3860
+    },
+    {
+      "entropy": 0.08484094133600592,
+      "epoch": 9.350030175015087,
+      "grad_norm": 0.21646778285503387,
+      "learning_rate": 4.866702233967785e-06,
+      "loss": 0.05998039245605469,
+      "mean_token_accuracy": 0.9771868668496608,
+      "num_tokens": 8989035.0,
+      "step": 3880
+    },
+    {
+      "epoch": 9.350030175015087,
+      "eval_entropy": 0.2300723305578982,
+      "eval_loss": 1.0633587837219238,
+      "eval_mean_token_accuracy": 0.828371500701047,
+      "eval_num_tokens": 8989035.0,
+      "eval_runtime": 88.0233,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 3880
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.801052595764347e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3900/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3900/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3900/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3900/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3900/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3900/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3900/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3900/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3900/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3900/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3900/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3900/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3900/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3900/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..b2b935637affd61973a50c17c877c09a3e3dba40
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3900/trainer_state.json
@@ -0,0 +1,4129 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 9.3983101991551,
+  "eval_steps": 20,
+  "global_step": 3900,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    },
+    {
+      "entropy": 0.1326121488586068,
+      "epoch": 7.181050090525045,
+      "grad_norm": 0.4843117296695709,
+      "learning_rate": 8.411870386507724e-05,
+      "loss": 0.09537842869758606,
+      "mean_token_accuracy": 0.9645151920616627,
+      "num_tokens": 6898444.0,
+      "step": 2980
+    },
+    {
+      "epoch": 7.181050090525045,
+      "eval_entropy": 0.27999477783280813,
+      "eval_loss": 0.9031037092208862,
+      "eval_mean_token_accuracy": 0.8261882204018282,
+      "eval_num_tokens": 6898444.0,
+      "eval_runtime": 88.5602,
+      "eval_samples_per_second": 16.034,
+      "eval_steps_per_second": 2.01,
+      "step": 2980
+    },
+    {
+      "entropy": 0.1156751093454659,
+      "epoch": 7.229330114665057,
+      "grad_norm": 0.5297922492027283,
+      "learning_rate": 8.149629351430326e-05,
+      "loss": 0.09203402996063233,
+      "mean_token_accuracy": 0.9689733698964119,
+      "num_tokens": 6947106.0,
+      "step": 3000
+    },
+    {
+      "epoch": 7.229330114665057,
+      "eval_entropy": 0.2740225845508361,
+      "eval_loss": 0.8876440525054932,
+      "eval_mean_token_accuracy": 0.8265116857678703,
+      "eval_num_tokens": 6947106.0,
+      "eval_runtime": 87.7971,
+      "eval_samples_per_second": 16.174,
+      "eval_steps_per_second": 2.027,
+      "step": 3000
+    },
+    {
+      "entropy": 0.12116114404052496,
+      "epoch": 7.27761013880507,
+      "grad_norm": 0.5467398166656494,
+      "learning_rate": 7.890406302404618e-05,
+      "loss": 0.0933056652545929,
+      "mean_token_accuracy": 0.9676455594599247,
+      "num_tokens": 6992784.0,
+      "step": 3020
+    },
+    {
+      "epoch": 7.27761013880507,
+      "eval_entropy": 0.27238342793804876,
+      "eval_loss": 0.9166117310523987,
+      "eval_mean_token_accuracy": 0.825499015242866,
+      "eval_num_tokens": 6992784.0,
+      "eval_runtime": 88.3276,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 3020
+    },
+    {
+      "entropy": 0.11715119928121567,
+      "epoch": 7.325890162945082,
+      "grad_norm": 0.4976702332496643,
+      "learning_rate": 7.634274596487176e-05,
+      "loss": 0.09449659585952759,
+      "mean_token_accuracy": 0.966506478190422,
+      "num_tokens": 7040482.0,
+      "step": 3040
+    },
+    {
+      "epoch": 7.325890162945082,
+      "eval_entropy": 0.2788975556077582,
+      "eval_loss": 0.8833065629005432,
+      "eval_mean_token_accuracy": 0.8269920596915684,
+      "eval_num_tokens": 7040482.0,
+      "eval_runtime": 88.9016,
+      "eval_samples_per_second": 15.973,
+      "eval_steps_per_second": 2.002,
+      "step": 3040
+    },
+    {
+      "entropy": 0.12565911412239075,
+      "epoch": 7.3741701870850935,
+      "grad_norm": 0.5728679895401001,
+      "learning_rate": 7.381306715921025e-05,
+      "loss": 0.09626244902610778,
+      "mean_token_accuracy": 0.9661444149911403,
+      "num_tokens": 7086261.0,
+      "step": 3060
+    },
+    {
+      "epoch": 7.3741701870850935,
+      "eval_entropy": 0.2762492011939542,
+      "eval_loss": 0.8961160778999329,
+      "eval_mean_token_accuracy": 0.826369910092836,
+      "eval_num_tokens": 7086261.0,
+      "eval_runtime": 88.5204,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 3060
+    },
+    {
+      "entropy": 0.12258567921817302,
+      "epoch": 7.422450211225105,
+      "grad_norm": 0.468985915184021,
+      "learning_rate": 7.131574247624031e-05,
+      "loss": 0.09120266437530518,
+      "mean_token_accuracy": 0.9674938447773457,
+      "num_tokens": 7133247.0,
+      "step": 3080
+    },
+    {
+      "epoch": 7.422450211225105,
+      "eval_entropy": 0.2688512086366,
+      "eval_loss": 0.9048654437065125,
+      "eval_mean_token_accuracy": 0.8265582302982888,
+      "eval_num_tokens": 7133247.0,
+      "eval_runtime": 88.164,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 3080
+    },
+    {
+      "entropy": 0.12070536343380808,
+      "epoch": 7.470730235365117,
+      "grad_norm": 0.5310841798782349,
+      "learning_rate": 6.885147862930688e-05,
+      "loss": 0.09386247992515565,
+      "mean_token_accuracy": 0.96664222702384,
+      "num_tokens": 7177810.0,
+      "step": 3100
+    },
+    {
+      "epoch": 7.470730235365117,
+      "eval_entropy": 0.26289632874593305,
+      "eval_loss": 0.902837336063385,
+      "eval_mean_token_accuracy": 0.828052184554968,
+      "eval_num_tokens": 7177810.0,
+      "eval_runtime": 88.1201,
+      "eval_samples_per_second": 16.114,
+      "eval_steps_per_second": 2.02,
+      "step": 3100
+    },
+    {
+      "entropy": 0.11322665181942285,
+      "epoch": 7.51901025950513,
+      "grad_norm": 0.6846067905426025,
+      "learning_rate": 6.642097297592908e-05,
+      "loss": 0.08772538304328918,
+      "mean_token_accuracy": 0.9691362135112286,
+      "num_tokens": 7225302.0,
+      "step": 3120
+    },
+    {
+      "epoch": 7.51901025950513,
+      "eval_entropy": 0.2670605613776807,
+      "eval_loss": 0.9130566120147705,
+      "eval_mean_token_accuracy": 0.826366903741708,
+      "eval_num_tokens": 7225302.0,
+      "eval_runtime": 88.6035,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3120
+    },
+    {
+      "entropy": 0.10985926580615342,
+      "epoch": 7.567290283645142,
+      "grad_norm": 0.4799748361110687,
+      "learning_rate": 6.402491332045683e-05,
+      "loss": 0.08672296404838561,
+      "mean_token_accuracy": 0.9698824048042297,
+      "num_tokens": 7274881.0,
+      "step": 3140
+    },
+    {
+      "epoch": 7.567290283645142,
+      "eval_entropy": 0.26657010866015146,
+      "eval_loss": 0.9056380987167358,
+      "eval_mean_token_accuracy": 0.8272326461384806,
+      "eval_num_tokens": 7274881.0,
+      "eval_runtime": 88.2967,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3140
+    },
+    {
+      "entropy": 0.11890473761595785,
+      "epoch": 7.615570307785154,
+      "grad_norm": 0.4081912934780121,
+      "learning_rate": 6.166397771942987e-05,
+      "loss": 0.09396827220916748,
+      "mean_token_accuracy": 0.967512384802103,
+      "num_tokens": 7321163.0,
+      "step": 3160
+    },
+    {
+      "epoch": 7.615570307785154,
+      "eval_entropy": 0.26364776575833226,
+      "eval_loss": 0.9204733967781067,
+      "eval_mean_token_accuracy": 0.827419729714983,
+      "eval_num_tokens": 7321163.0,
+      "eval_runtime": 88.3441,
+      "eval_samples_per_second": 16.074,
+      "eval_steps_per_second": 2.015,
+      "step": 3160
+    },
+    {
+      "entropy": 0.12209134344011545,
+      "epoch": 7.663850331925166,
+      "grad_norm": 0.5159774422645569,
+      "learning_rate": 5.93388342896962e-05,
+      "loss": 0.09292978644371033,
+      "mean_token_accuracy": 0.9658421002328396,
+      "num_tokens": 7366217.0,
+      "step": 3180
+    },
+    {
+      "epoch": 7.663850331925166,
+      "eval_entropy": 0.2714110068725736,
+      "eval_loss": 0.9143092036247253,
+      "eval_mean_token_accuracy": 0.8262251099843657,
+      "eval_num_tokens": 7366217.0,
+      "eval_runtime": 87.9635,
+      "eval_samples_per_second": 16.143,
+      "eval_steps_per_second": 2.024,
+      "step": 3180
+    },
+    {
+      "entropy": 0.11955609126016498,
+      "epoch": 7.712130356065178,
+      "grad_norm": 0.7005976438522339,
+      "learning_rate": 5.705014101934252e-05,
+      "loss": 0.09319776892662049,
+      "mean_token_accuracy": 0.9674513377249241,
+      "num_tokens": 7410887.0,
+      "step": 3200
+    },
+    {
+      "epoch": 7.712130356065178,
+      "eval_entropy": 0.2626291513442993,
+      "eval_loss": 0.91884446144104,
+      "eval_mean_token_accuracy": 0.827563955207889,
+      "eval_num_tokens": 7410887.0,
+      "eval_runtime": 87.8849,
+      "eval_samples_per_second": 16.157,
+      "eval_steps_per_second": 2.025,
+      "step": 3200
+    },
+    {
+      "entropy": 0.11767532899975777,
+      "epoch": 7.76041038020519,
+      "grad_norm": 0.4675362706184387,
+      "learning_rate": 5.479854558149187e-05,
+      "loss": 0.09038087725639343,
+      "mean_token_accuracy": 0.9669680051505566,
+      "num_tokens": 7457578.0,
+      "step": 3220
+    },
+    {
+      "epoch": 7.76041038020519,
+      "eval_entropy": 0.26581037203582486,
+      "eval_loss": 0.9146883487701416,
+      "eval_mean_token_accuracy": 0.8270345812433222,
+      "eval_num_tokens": 7457578.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 3220
+    },
+    {
+      "entropy": 0.11383061073720455,
+      "epoch": 7.808690404345202,
+      "grad_norm": 0.541247546672821,
+      "learning_rate": 5.258468515101918e-05,
+      "loss": 0.08942405581474304,
+      "mean_token_accuracy": 0.968398705124855,
+      "num_tokens": 7505761.0,
+      "step": 3240
+    },
+    {
+      "epoch": 7.808690404345202,
+      "eval_entropy": 0.26687998960861997,
+      "eval_loss": 0.9037085175514221,
+      "eval_mean_token_accuracy": 0.8274977863504646,
+      "eval_num_tokens": 7505761.0,
+      "eval_runtime": 88.5163,
+      "eval_samples_per_second": 16.042,
+      "eval_steps_per_second": 2.011,
+      "step": 3240
+    },
+    {
+      "entropy": 0.12004058873280883,
+      "epoch": 7.856970428485214,
+      "grad_norm": 0.5289552211761475,
+      "learning_rate": 5.040918622423899e-05,
+      "loss": 0.09064805507659912,
+      "mean_token_accuracy": 0.9673330299556255,
+      "num_tokens": 7552557.0,
+      "step": 3260
+    },
+    {
+      "epoch": 7.856970428485214,
+      "eval_entropy": 0.26288379251622085,
+      "eval_loss": 0.9146178364753723,
+      "eval_mean_token_accuracy": 0.8276662448149049,
+      "eval_num_tokens": 7552557.0,
+      "eval_runtime": 88.0723,
+      "eval_samples_per_second": 16.123,
+      "eval_steps_per_second": 2.021,
+      "step": 3260
+    },
+    {
+      "entropy": 0.1180689456872642,
+      "epoch": 7.905250452625227,
+      "grad_norm": 0.4754861891269684,
+      "learning_rate": 4.827266444161377e-05,
+      "loss": 0.09402655363082886,
+      "mean_token_accuracy": 0.9663361787796021,
+      "num_tokens": 7596998.0,
+      "step": 3280
+    },
+    {
+      "epoch": 7.905250452625227,
+      "eval_entropy": 0.261690996922134,
+      "eval_loss": 0.9120186567306519,
+      "eval_mean_token_accuracy": 0.8278995015647974,
+      "eval_num_tokens": 7596998.0,
+      "eval_runtime": 88.6573,
+      "eval_samples_per_second": 16.017,
+      "eval_steps_per_second": 2.008,
+      "step": 3280
+    },
+    {
+      "entropy": 0.1162164019420743,
+      "epoch": 7.953530476765239,
+      "grad_norm": 0.6203398108482361,
+      "learning_rate": 4.61757244135355e-05,
+      "loss": 0.08832768201828003,
+      "mean_token_accuracy": 0.9694220684468746,
+      "num_tokens": 7643288.0,
+      "step": 3300
+    },
+    {
+      "epoch": 7.953530476765239,
+      "eval_entropy": 0.26158678129817664,
+      "eval_loss": 0.9301361441612244,
+      "eval_mean_token_accuracy": 0.8269603885291668,
+      "eval_num_tokens": 7643288.0,
+      "eval_runtime": 88.7064,
+      "eval_samples_per_second": 16.008,
+      "eval_steps_per_second": 2.007,
+      "step": 3300
+    },
+    {
+      "entropy": 0.10877202204489089,
+      "epoch": 8.0,
+      "grad_norm": 3.058673143386841,
+      "learning_rate": 4.411895954922763e-05,
+      "loss": 0.09159615635871887,
+      "mean_token_accuracy": 0.9686878834451947,
+      "num_tokens": 7688608.0,
+      "step": 3320
+    },
+    {
+      "epoch": 8.0,
+      "eval_entropy": 0.2590438057197614,
+      "eval_loss": 0.9262065887451172,
+      "eval_mean_token_accuracy": 0.8277404546067956,
+      "eval_num_tokens": 7688608.0,
+      "eval_runtime": 87.7557,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 3320
+    },
+    {
+      "entropy": 0.09969092034734786,
+      "epoch": 8.048280024140013,
+      "grad_norm": 0.2561911642551422,
+      "learning_rate": 4.2102951888817644e-05,
+      "loss": 0.07228735685348511,
+      "mean_token_accuracy": 0.9743556551635265,
+      "num_tokens": 7732856.0,
+      "step": 3340
+    },
+    {
+      "epoch": 8.048280024140013,
+      "eval_entropy": 0.2462534454933713,
+      "eval_loss": 0.9755983352661133,
+      "eval_mean_token_accuracy": 0.827510209565752,
+      "eval_num_tokens": 7732856.0,
+      "eval_runtime": 88.0297,
+      "eval_samples_per_second": 16.131,
+      "eval_steps_per_second": 2.022,
+      "step": 3340
+    },
+    {
+      "entropy": 0.09477624730207027,
+      "epoch": 8.096560048280024,
+      "grad_norm": 0.3670356571674347,
+      "learning_rate": 4.012827193862633e-05,
+      "loss": 0.0686036467552185,
+      "mean_token_accuracy": 0.9743670664727688,
+      "num_tokens": 7778585.0,
+      "step": 3360
+    },
+    {
+      "epoch": 8.096560048280024,
+      "eval_entropy": 0.2440882951356052,
+      "eval_loss": 1.0118423700332642,
+      "eval_mean_token_accuracy": 0.8271241070849172,
+      "eval_num_tokens": 7778585.0,
+      "eval_runtime": 88.1364,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 3360
+    },
+    {
+      "entropy": 0.09417980071157217,
+      "epoch": 8.144840072420036,
+      "grad_norm": 0.5785111784934998,
+      "learning_rate": 3.819547850972133e-05,
+      "loss": 0.07100698947906495,
+      "mean_token_accuracy": 0.9737804099917412,
+      "num_tokens": 7824287.0,
+      "step": 3380
+    },
+    {
+      "epoch": 8.144840072420036,
+      "eval_entropy": 0.24603172860453637,
+      "eval_loss": 1.0021991729736328,
+      "eval_mean_token_accuracy": 0.827162122793412,
+      "eval_num_tokens": 7824287.0,
+      "eval_runtime": 88.4534,
+      "eval_samples_per_second": 16.054,
+      "eval_steps_per_second": 2.012,
+      "step": 3380
+    },
+    {
+      "entropy": 0.087222247896716,
+      "epoch": 8.193120096560047,
+      "grad_norm": 0.44848504662513733,
+      "learning_rate": 3.6305118559780393e-05,
+      "loss": 0.06325234174728393,
+      "mean_token_accuracy": 0.9769335232675076,
+      "num_tokens": 7874784.0,
+      "step": 3400
+    },
+    {
+      "epoch": 8.193120096560047,
+      "eval_entropy": 0.24544863905129807,
+      "eval_loss": 1.005008339881897,
+      "eval_mean_token_accuracy": 0.8273035240976998,
+      "eval_num_tokens": 7874784.0,
+      "eval_runtime": 88.6737,
+      "eval_samples_per_second": 16.014,
+      "eval_steps_per_second": 2.007,
+      "step": 3400
+    },
+    {
+      "entropy": 0.09037275062873959,
+      "epoch": 8.24140012070006,
+      "grad_norm": 0.23529155552387238,
+      "learning_rate": 3.4457727038308435e-05,
+      "loss": 0.06589569449424744,
+      "mean_token_accuracy": 0.9751470290124417,
+      "num_tokens": 7925240.0,
+      "step": 3420
+    },
+    {
+      "epoch": 8.24140012070006,
+      "eval_entropy": 0.24203423494368456,
+      "eval_loss": 1.0052824020385742,
+      "eval_mean_token_accuracy": 0.8275575503874361,
+      "eval_num_tokens": 7925240.0,
+      "eval_runtime": 88.3486,
+      "eval_samples_per_second": 16.073,
+      "eval_steps_per_second": 2.015,
+      "step": 3420
+    },
+    {
+      "entropy": 0.10373208015225828,
+      "epoch": 8.289680144840073,
+      "grad_norm": 0.24796408414840698,
+      "learning_rate": 3.2653826735253594e-05,
+      "loss": 0.07714771628379821,
+      "mean_token_accuracy": 0.9722891598939896,
+      "num_tokens": 7966685.0,
+      "step": 3440
+    },
+    {
+      "epoch": 8.289680144840073,
+      "eval_entropy": 0.23461157050025597,
+      "eval_loss": 1.012108325958252,
+      "eval_mean_token_accuracy": 0.82825900564033,
+      "eval_num_tokens": 7966685.0,
+      "eval_runtime": 88.1018,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 3440
+    },
+    {
+      "entropy": 0.08967954274266958,
+      "epoch": 8.337960168980084,
+      "grad_norm": 0.5314674973487854,
+      "learning_rate": 3.08939281330635e-05,
+      "loss": 0.06746266484260559,
+      "mean_token_accuracy": 0.9753133699297905,
+      "num_tokens": 8013656.0,
+      "step": 3460
+    },
+    {
+      "epoch": 8.337960168980084,
+      "eval_entropy": 0.23781998589467468,
+      "eval_loss": 1.0119913816452026,
+      "eval_mean_token_accuracy": 0.8273287157664139,
+      "eval_num_tokens": 8013656.0,
+      "eval_runtime": 88.6038,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3460
+    },
+    {
+      "entropy": 0.09280927330255509,
+      "epoch": 8.386240193120097,
+      "grad_norm": 0.30658936500549316,
+      "learning_rate": 2.9178529262224946e-05,
+      "loss": 0.06936421990394592,
+      "mean_token_accuracy": 0.9742105580866337,
+      "num_tokens": 8061108.0,
+      "step": 3480
+    },
+    {
+      "epoch": 8.386240193120097,
+      "eval_entropy": 0.23672426734747512,
+      "eval_loss": 1.0162041187286377,
+      "eval_mean_token_accuracy": 0.8278394281194451,
+      "eval_num_tokens": 8061108.0,
+      "eval_runtime": 88.6015,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 3480
+    },
+    {
+      "entropy": 0.09195122267119586,
+      "epoch": 8.43452021726011,
+      "grad_norm": 0.24512454867362976,
+      "learning_rate": 2.7508115560327256e-05,
+      "loss": 0.06996064782142639,
+      "mean_token_accuracy": 0.974104854464531,
+      "num_tokens": 8106924.0,
+      "step": 3500
+    },
+    {
+      "epoch": 8.43452021726011,
+      "eval_entropy": 0.23758245828781235,
+      "eval_loss": 1.0157747268676758,
+      "eval_mean_token_accuracy": 0.8278501579600773,
+      "eval_num_tokens": 8106924.0,
+      "eval_runtime": 88.5678,
+      "eval_samples_per_second": 16.033,
+      "eval_steps_per_second": 2.01,
+      "step": 3500
+    },
+    {
+      "entropy": 0.09518477590754629,
+      "epoch": 8.48280024140012,
+      "grad_norm": 0.2373228371143341,
+      "learning_rate": 2.5883159734688868e-05,
+      "loss": 0.06832797527313232,
+      "mean_token_accuracy": 0.9743806682527065,
+      "num_tokens": 8153070.0,
+      "step": 3520
+    },
+    {
+      "epoch": 8.48280024140012,
+      "eval_entropy": 0.2379308074210467,
+      "eval_loss": 1.0180530548095703,
+      "eval_mean_token_accuracy": 0.8282477052024241,
+      "eval_num_tokens": 8153070.0,
+      "eval_runtime": 88.4651,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 3520
+    },
+    {
+      "entropy": 0.0906269280705601,
+      "epoch": 8.531080265540133,
+      "grad_norm": 0.37602657079696655,
+      "learning_rate": 2.4304121628587173e-05,
+      "loss": 0.06921588778495788,
+      "mean_token_accuracy": 0.9745571479201317,
+      "num_tokens": 8200066.0,
+      "step": 3540
+    },
+    {
+      "epoch": 8.531080265540133,
+      "eval_entropy": 0.2357490769263064,
+      "eval_loss": 1.017139196395874,
+      "eval_mean_token_accuracy": 0.828713178299786,
+      "eval_num_tokens": 8200066.0,
+      "eval_runtime": 88.6424,
+      "eval_samples_per_second": 16.019,
+      "eval_steps_per_second": 2.008,
+      "step": 3540
+    },
+    {
+      "entropy": 0.09273334196768701,
+      "epoch": 8.579360289680144,
+      "grad_norm": 0.2978838086128235,
+      "learning_rate": 2.277144809112794e-05,
+      "loss": 0.06970887780189514,
+      "mean_token_accuracy": 0.9751374177634716,
+      "num_tokens": 8246157.0,
+      "step": 3560
+    },
+    {
+      "epoch": 8.579360289680144,
+      "eval_entropy": 0.23590933138065123,
+      "eval_loss": 1.019231915473938,
+      "eval_mean_token_accuracy": 0.8279163412163767,
+      "eval_num_tokens": 8246157.0,
+      "eval_runtime": 88.2691,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 3560
+    },
+    {
+      "entropy": 0.09854201921261847,
+      "epoch": 8.627640313820157,
+      "grad_norm": 0.5381243228912354,
+      "learning_rate": 2.1285572850792413e-05,
+      "loss": 0.07123501300811767,
+      "mean_token_accuracy": 0.9723988175392151,
+      "num_tokens": 8291752.0,
+      "step": 3580
+    },
+    {
+      "epoch": 8.627640313820157,
+      "eval_entropy": 0.23592024949494372,
+      "eval_loss": 1.020250678062439,
+      "eval_mean_token_accuracy": 0.8282466226079491,
+      "eval_num_tokens": 8291752.0,
+      "eval_runtime": 87.8939,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 3580
+    },
+    {
+      "entropy": 0.09092000788077712,
+      "epoch": 8.67592033796017,
+      "grad_norm": 0.29089561104774475,
+      "learning_rate": 1.9846916392697493e-05,
+      "loss": 0.06982720494270325,
+      "mean_token_accuracy": 0.9749995484948158,
+      "num_tokens": 8337651.0,
+      "step": 3600
+    },
+    {
+      "epoch": 8.67592033796017,
+      "eval_entropy": 0.23658913382318583,
+      "eval_loss": 1.019862413406372,
+      "eval_mean_token_accuracy": 0.8280127842104836,
+      "eval_num_tokens": 8337651.0,
+      "eval_runtime": 88.2855,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 3600
+    },
+    {
+      "entropy": 0.08708424470387399,
+      "epoch": 8.72420036210018,
+      "grad_norm": 0.29904839396476746,
+      "learning_rate": 1.8455885839603106e-05,
+      "loss": 0.06801370978355407,
+      "mean_token_accuracy": 0.9755801670253277,
+      "num_tokens": 8385330.0,
+      "step": 3620
+    },
+    {
+      "epoch": 8.72420036210018,
+      "eval_entropy": 0.23627035771862845,
+      "eval_loss": 1.0204569101333618,
+      "eval_mean_token_accuracy": 0.8279967894045155,
+      "eval_num_tokens": 8385330.0,
+      "eval_runtime": 88.1753,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 3620
+    },
+    {
+      "entropy": 0.0916255566291511,
+      "epoch": 8.772480386240193,
+      "grad_norm": 0.20571033656597137,
+      "learning_rate": 1.7112874836701502e-05,
+      "loss": 0.0692249894142151,
+      "mean_token_accuracy": 0.9745148062705994,
+      "num_tokens": 8432134.0,
+      "step": 3640
+    },
+    {
+      "epoch": 8.772480386240193,
+      "eval_entropy": 0.2340821772813797,
+      "eval_loss": 1.0212821960449219,
+      "eval_mean_token_accuracy": 0.828373622358515,
+      "eval_num_tokens": 8432134.0,
+      "eval_runtime": 87.2518,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 3640
+    },
+    {
+      "entropy": 0.0926868049427867,
+      "epoch": 8.820760410380204,
+      "grad_norm": 0.2618265450000763,
+      "learning_rate": 1.5818263440220203e-05,
+      "loss": 0.07338647246360779,
+      "mean_token_accuracy": 0.9724708944559097,
+      "num_tokens": 8476758.0,
+      "step": 3660
+    },
+    {
+      "epoch": 8.820760410380204,
+      "eval_entropy": 0.23637074575330433,
+      "eval_loss": 1.0186725854873657,
+      "eval_mean_token_accuracy": 0.8277707605549459,
+      "eval_num_tokens": 8476758.0,
+      "eval_runtime": 88.4695,
+      "eval_samples_per_second": 16.051,
+      "eval_steps_per_second": 2.012,
+      "step": 3660
+    },
+    {
+      "entropy": 0.09175782897509635,
+      "epoch": 8.869040434520217,
+      "grad_norm": 0.2889852821826935,
+      "learning_rate": 1.4572418009870604e-05,
+      "loss": 0.07066041231155396,
+      "mean_token_accuracy": 0.9725673109292984,
+      "num_tokens": 8523294.0,
+      "step": 3680
+    },
+    {
+      "epoch": 8.869040434520217,
+      "eval_entropy": 0.23945164404223473,
+      "eval_loss": 1.0145231485366821,
+      "eval_mean_token_accuracy": 0.8281202523895864,
+      "eval_num_tokens": 8523294.0,
+      "eval_runtime": 88.368,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 3680
+    },
+    {
+      "entropy": 0.09265611860901117,
+      "epoch": 8.91732045866023,
+      "grad_norm": 0.41092950105667114,
+      "learning_rate": 1.3375691105172625e-05,
+      "loss": 0.06975027322769164,
+      "mean_token_accuracy": 0.974636297672987,
+      "num_tokens": 8570358.0,
+      "step": 3700
+    },
+    {
+      "epoch": 8.91732045866023,
+      "eval_entropy": 0.239588919901446,
+      "eval_loss": 1.014059066772461,
+      "eval_mean_token_accuracy": 0.828028041660116,
+      "eval_num_tokens": 8570358.0,
+      "eval_runtime": 88.1039,
+      "eval_samples_per_second": 16.117,
+      "eval_steps_per_second": 2.02,
+      "step": 3700
+    },
+    {
+      "entropy": 0.0896465003490448,
+      "epoch": 8.965600482800241,
+      "grad_norm": 0.24000254273414612,
+      "learning_rate": 1.2228421385684305e-05,
+      "loss": 0.06743953824043274,
+      "mean_token_accuracy": 0.974006575345993,
+      "num_tokens": 8618246.0,
+      "step": 3720
+    },
+    {
+      "epoch": 8.965600482800241,
+      "eval_entropy": 0.2390461369846644,
+      "eval_loss": 1.011997938156128,
+      "eval_mean_token_accuracy": 0.8284533211354459,
+      "eval_num_tokens": 8618246.0,
+      "eval_runtime": 88.6202,
+      "eval_samples_per_second": 16.023,
+      "eval_steps_per_second": 2.009,
+      "step": 3720
+    },
+    {
+      "entropy": 0.09575109251520851,
+      "epoch": 9.012070006035003,
+      "grad_norm": 0.18088576197624207,
+      "learning_rate": 1.1130933515165429e-05,
+      "loss": 0.07295305132865906,
+      "mean_token_accuracy": 0.9739722857227573,
+      "num_tokens": 8661177.0,
+      "step": 3740
+    },
+    {
+      "epoch": 9.012070006035003,
+      "eval_entropy": 0.23728882387447892,
+      "eval_loss": 1.0134038925170898,
+      "eval_mean_token_accuracy": 0.8286508951294288,
+      "eval_num_tokens": 8661177.0,
+      "eval_runtime": 88.6434,
+      "eval_samples_per_second": 16.019,
+      "eval_steps_per_second": 2.008,
+      "step": 3740
+    },
+    {
+      "entropy": 0.08883201805874705,
+      "epoch": 9.060350030175016,
+      "grad_norm": 0.23120898008346558,
+      "learning_rate": 1.008353806970132e-05,
+      "loss": 0.06326142549514771,
+      "mean_token_accuracy": 0.9780020996928215,
+      "num_tokens": 8707602.0,
+      "step": 3760
+    },
+    {
+      "epoch": 9.060350030175016,
+      "eval_entropy": 0.23416905331142832,
+      "eval_loss": 1.0284100770950317,
+      "eval_mean_token_accuracy": 0.8284083213029283,
+      "eval_num_tokens": 8707602.0,
+      "eval_runtime": 88.2933,
+      "eval_samples_per_second": 16.083,
+      "eval_steps_per_second": 2.016,
+      "step": 3760
+    },
+    {
+      "entropy": 0.08861468276008963,
+      "epoch": 9.108630054315027,
+      "grad_norm": 0.17277656495571136,
+      "learning_rate": 9.086531449813489e-06,
+      "loss": 0.06727416515350342,
+      "mean_token_accuracy": 0.9760390736162663,
+      "num_tokens": 8751287.0,
+      "step": 3780
+    },
+    {
+      "epoch": 9.108630054315027,
+      "eval_entropy": 0.23370317386442355,
+      "eval_loss": 1.0381091833114624,
+      "eval_mean_token_accuracy": 0.8282668178670862,
+      "eval_num_tokens": 8751287.0,
+      "eval_runtime": 88.5738,
+      "eval_samples_per_second": 16.032,
+      "eval_steps_per_second": 2.01,
+      "step": 3780
+    },
+    {
+      "entropy": 0.09032954541034996,
+      "epoch": 9.15691007845504,
+      "grad_norm": 0.19481085240840912,
+      "learning_rate": 8.140195796581981e-06,
+      "loss": 0.06117786169052124,
+      "mean_token_accuracy": 0.9764454193413258,
+      "num_tokens": 8799982.0,
+      "step": 3800
+    },
+    {
+      "epoch": 9.15691007845504,
+      "eval_entropy": 0.23215936609868254,
+      "eval_loss": 1.0467767715454102,
+      "eval_mean_token_accuracy": 0.828405868806196,
+      "eval_num_tokens": 8799982.0,
+      "eval_runtime": 87.9551,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 3800
+    },
+    {
+      "entropy": 0.08824449330568314,
+      "epoch": 9.20519010259505,
+      "grad_norm": 0.20382793247699738,
+      "learning_rate": 7.2447989118025455e-06,
+      "loss": 0.06422553658485412,
+      "mean_token_accuracy": 0.9763121999800205,
+      "num_tokens": 8845536.0,
+      "step": 3820
+    },
+    {
+      "epoch": 9.20519010259505,
+      "eval_entropy": 0.23169661472352704,
+      "eval_loss": 1.0529043674468994,
+      "eval_mean_token_accuracy": 0.8282025610463003,
+      "eval_num_tokens": 8845536.0,
+      "eval_runtime": 88.5548,
+      "eval_samples_per_second": 16.035,
+      "eval_steps_per_second": 2.01,
+      "step": 3820
+    },
+    {
+      "entropy": 0.07768499720841646,
+      "epoch": 9.253470126735063,
+      "grad_norm": 0.11727919429540634,
+      "learning_rate": 6.400594182202084e-06,
+      "loss": 0.057279336452484134,
+      "mean_token_accuracy": 0.9795166745781898,
+      "num_tokens": 8896053.0,
+      "step": 3840
+    },
+    {
+      "epoch": 9.253470126735063,
+      "eval_entropy": 0.2301936373114586,
+      "eval_loss": 1.0576049089431763,
+      "eval_mean_token_accuracy": 0.8283417489421502,
+      "eval_num_tokens": 8896053.0,
+      "eval_runtime": 88.7854,
+      "eval_samples_per_second": 15.994,
+      "eval_steps_per_second": 2.005,
+      "step": 3840
+    },
+    {
+      "entropy": 0.09015200780704617,
+      "epoch": 9.301750150875076,
+      "grad_norm": 0.2089058756828308,
+      "learning_rate": 5.6078205077330705e-06,
+      "loss": 0.06626194715499878,
+      "mean_token_accuracy": 0.9747506938874722,
+      "num_tokens": 8940591.0,
+      "step": 3860
+    },
+    {
+      "epoch": 9.301750150875076,
+      "eval_entropy": 0.23024900676159377,
+      "eval_loss": 1.0599786043167114,
+      "eval_mean_token_accuracy": 0.8283051915383071,
+      "eval_num_tokens": 8940591.0,
+      "eval_runtime": 88.2963,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3860
+    },
+    {
+      "entropy": 0.08484094133600592,
+      "epoch": 9.350030175015087,
+      "grad_norm": 0.21646778285503387,
+      "learning_rate": 4.866702233967785e-06,
+      "loss": 0.05998039245605469,
+      "mean_token_accuracy": 0.9771868668496608,
+      "num_tokens": 8989035.0,
+      "step": 3880
+    },
+    {
+      "epoch": 9.350030175015087,
+      "eval_entropy": 0.2300723305578982,
+      "eval_loss": 1.0633587837219238,
+      "eval_mean_token_accuracy": 0.828371500701047,
+      "eval_num_tokens": 8989035.0,
+      "eval_runtime": 88.0233,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 3880
+    },
+    {
+      "entropy": 0.08905019680969417,
+      "epoch": 9.3983101991551,
+      "grad_norm": 0.21842406690120697,
+      "learning_rate": 4.177449088611095e-06,
+      "loss": 0.06666865348815917,
+      "mean_token_accuracy": 0.9753445640206337,
+      "num_tokens": 9033383.0,
+      "step": 3900
+    },
+    {
+      "epoch": 9.3983101991551,
+      "eval_entropy": 0.2295504841911659,
+      "eval_loss": 1.0662461519241333,
+      "eval_mean_token_accuracy": 0.8284122297603093,
+      "eval_num_tokens": 9033383.0,
+      "eval_runtime": 88.5525,
+      "eval_samples_per_second": 16.036,
+      "eval_steps_per_second": 2.01,
+      "step": 3900
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.819766712732467e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3920/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3920/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3920/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3920/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3920/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3920/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3920/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3920/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3920/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3920/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3920/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3920/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3920/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3920/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..e326084479ff5f78f5a9e96bd3232ad4b8b29d7a
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3920/trainer_state.json
@@ -0,0 +1,4150 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 9.446590223295111,
+  "eval_steps": 20,
+  "global_step": 3920,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    },
+    {
+      "entropy": 0.1326121488586068,
+      "epoch": 7.181050090525045,
+      "grad_norm": 0.4843117296695709,
+      "learning_rate": 8.411870386507724e-05,
+      "loss": 0.09537842869758606,
+      "mean_token_accuracy": 0.9645151920616627,
+      "num_tokens": 6898444.0,
+      "step": 2980
+    },
+    {
+      "epoch": 7.181050090525045,
+      "eval_entropy": 0.27999477783280813,
+      "eval_loss": 0.9031037092208862,
+      "eval_mean_token_accuracy": 0.8261882204018282,
+      "eval_num_tokens": 6898444.0,
+      "eval_runtime": 88.5602,
+      "eval_samples_per_second": 16.034,
+      "eval_steps_per_second": 2.01,
+      "step": 2980
+    },
+    {
+      "entropy": 0.1156751093454659,
+      "epoch": 7.229330114665057,
+      "grad_norm": 0.5297922492027283,
+      "learning_rate": 8.149629351430326e-05,
+      "loss": 0.09203402996063233,
+      "mean_token_accuracy": 0.9689733698964119,
+      "num_tokens": 6947106.0,
+      "step": 3000
+    },
+    {
+      "epoch": 7.229330114665057,
+      "eval_entropy": 0.2740225845508361,
+      "eval_loss": 0.8876440525054932,
+      "eval_mean_token_accuracy": 0.8265116857678703,
+      "eval_num_tokens": 6947106.0,
+      "eval_runtime": 87.7971,
+      "eval_samples_per_second": 16.174,
+      "eval_steps_per_second": 2.027,
+      "step": 3000
+    },
+    {
+      "entropy": 0.12116114404052496,
+      "epoch": 7.27761013880507,
+      "grad_norm": 0.5467398166656494,
+      "learning_rate": 7.890406302404618e-05,
+      "loss": 0.0933056652545929,
+      "mean_token_accuracy": 0.9676455594599247,
+      "num_tokens": 6992784.0,
+      "step": 3020
+    },
+    {
+      "epoch": 7.27761013880507,
+      "eval_entropy": 0.27238342793804876,
+      "eval_loss": 0.9166117310523987,
+      "eval_mean_token_accuracy": 0.825499015242866,
+      "eval_num_tokens": 6992784.0,
+      "eval_runtime": 88.3276,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 3020
+    },
+    {
+      "entropy": 0.11715119928121567,
+      "epoch": 7.325890162945082,
+      "grad_norm": 0.4976702332496643,
+      "learning_rate": 7.634274596487176e-05,
+      "loss": 0.09449659585952759,
+      "mean_token_accuracy": 0.966506478190422,
+      "num_tokens": 7040482.0,
+      "step": 3040
+    },
+    {
+      "epoch": 7.325890162945082,
+      "eval_entropy": 0.2788975556077582,
+      "eval_loss": 0.8833065629005432,
+      "eval_mean_token_accuracy": 0.8269920596915684,
+      "eval_num_tokens": 7040482.0,
+      "eval_runtime": 88.9016,
+      "eval_samples_per_second": 15.973,
+      "eval_steps_per_second": 2.002,
+      "step": 3040
+    },
+    {
+      "entropy": 0.12565911412239075,
+      "epoch": 7.3741701870850935,
+      "grad_norm": 0.5728679895401001,
+      "learning_rate": 7.381306715921025e-05,
+      "loss": 0.09626244902610778,
+      "mean_token_accuracy": 0.9661444149911403,
+      "num_tokens": 7086261.0,
+      "step": 3060
+    },
+    {
+      "epoch": 7.3741701870850935,
+      "eval_entropy": 0.2762492011939542,
+      "eval_loss": 0.8961160778999329,
+      "eval_mean_token_accuracy": 0.826369910092836,
+      "eval_num_tokens": 7086261.0,
+      "eval_runtime": 88.5204,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 3060
+    },
+    {
+      "entropy": 0.12258567921817302,
+      "epoch": 7.422450211225105,
+      "grad_norm": 0.468985915184021,
+      "learning_rate": 7.131574247624031e-05,
+      "loss": 0.09120266437530518,
+      "mean_token_accuracy": 0.9674938447773457,
+      "num_tokens": 7133247.0,
+      "step": 3080
+    },
+    {
+      "epoch": 7.422450211225105,
+      "eval_entropy": 0.2688512086366,
+      "eval_loss": 0.9048654437065125,
+      "eval_mean_token_accuracy": 0.8265582302982888,
+      "eval_num_tokens": 7133247.0,
+      "eval_runtime": 88.164,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 3080
+    },
+    {
+      "entropy": 0.12070536343380808,
+      "epoch": 7.470730235365117,
+      "grad_norm": 0.5310841798782349,
+      "learning_rate": 6.885147862930688e-05,
+      "loss": 0.09386247992515565,
+      "mean_token_accuracy": 0.96664222702384,
+      "num_tokens": 7177810.0,
+      "step": 3100
+    },
+    {
+      "epoch": 7.470730235365117,
+      "eval_entropy": 0.26289632874593305,
+      "eval_loss": 0.902837336063385,
+      "eval_mean_token_accuracy": 0.828052184554968,
+      "eval_num_tokens": 7177810.0,
+      "eval_runtime": 88.1201,
+      "eval_samples_per_second": 16.114,
+      "eval_steps_per_second": 2.02,
+      "step": 3100
+    },
+    {
+      "entropy": 0.11322665181942285,
+      "epoch": 7.51901025950513,
+      "grad_norm": 0.6846067905426025,
+      "learning_rate": 6.642097297592908e-05,
+      "loss": 0.08772538304328918,
+      "mean_token_accuracy": 0.9691362135112286,
+      "num_tokens": 7225302.0,
+      "step": 3120
+    },
+    {
+      "epoch": 7.51901025950513,
+      "eval_entropy": 0.2670605613776807,
+      "eval_loss": 0.9130566120147705,
+      "eval_mean_token_accuracy": 0.826366903741708,
+      "eval_num_tokens": 7225302.0,
+      "eval_runtime": 88.6035,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3120
+    },
+    {
+      "entropy": 0.10985926580615342,
+      "epoch": 7.567290283645142,
+      "grad_norm": 0.4799748361110687,
+      "learning_rate": 6.402491332045683e-05,
+      "loss": 0.08672296404838561,
+      "mean_token_accuracy": 0.9698824048042297,
+      "num_tokens": 7274881.0,
+      "step": 3140
+    },
+    {
+      "epoch": 7.567290283645142,
+      "eval_entropy": 0.26657010866015146,
+      "eval_loss": 0.9056380987167358,
+      "eval_mean_token_accuracy": 0.8272326461384806,
+      "eval_num_tokens": 7274881.0,
+      "eval_runtime": 88.2967,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3140
+    },
+    {
+      "entropy": 0.11890473761595785,
+      "epoch": 7.615570307785154,
+      "grad_norm": 0.4081912934780121,
+      "learning_rate": 6.166397771942987e-05,
+      "loss": 0.09396827220916748,
+      "mean_token_accuracy": 0.967512384802103,
+      "num_tokens": 7321163.0,
+      "step": 3160
+    },
+    {
+      "epoch": 7.615570307785154,
+      "eval_entropy": 0.26364776575833226,
+      "eval_loss": 0.9204733967781067,
+      "eval_mean_token_accuracy": 0.827419729714983,
+      "eval_num_tokens": 7321163.0,
+      "eval_runtime": 88.3441,
+      "eval_samples_per_second": 16.074,
+      "eval_steps_per_second": 2.015,
+      "step": 3160
+    },
+    {
+      "entropy": 0.12209134344011545,
+      "epoch": 7.663850331925166,
+      "grad_norm": 0.5159774422645569,
+      "learning_rate": 5.93388342896962e-05,
+      "loss": 0.09292978644371033,
+      "mean_token_accuracy": 0.9658421002328396,
+      "num_tokens": 7366217.0,
+      "step": 3180
+    },
+    {
+      "epoch": 7.663850331925166,
+      "eval_entropy": 0.2714110068725736,
+      "eval_loss": 0.9143092036247253,
+      "eval_mean_token_accuracy": 0.8262251099843657,
+      "eval_num_tokens": 7366217.0,
+      "eval_runtime": 87.9635,
+      "eval_samples_per_second": 16.143,
+      "eval_steps_per_second": 2.024,
+      "step": 3180
+    },
+    {
+      "entropy": 0.11955609126016498,
+      "epoch": 7.712130356065178,
+      "grad_norm": 0.7005976438522339,
+      "learning_rate": 5.705014101934252e-05,
+      "loss": 0.09319776892662049,
+      "mean_token_accuracy": 0.9674513377249241,
+      "num_tokens": 7410887.0,
+      "step": 3200
+    },
+    {
+      "epoch": 7.712130356065178,
+      "eval_entropy": 0.2626291513442993,
+      "eval_loss": 0.91884446144104,
+      "eval_mean_token_accuracy": 0.827563955207889,
+      "eval_num_tokens": 7410887.0,
+      "eval_runtime": 87.8849,
+      "eval_samples_per_second": 16.157,
+      "eval_steps_per_second": 2.025,
+      "step": 3200
+    },
+    {
+      "entropy": 0.11767532899975777,
+      "epoch": 7.76041038020519,
+      "grad_norm": 0.4675362706184387,
+      "learning_rate": 5.479854558149187e-05,
+      "loss": 0.09038087725639343,
+      "mean_token_accuracy": 0.9669680051505566,
+      "num_tokens": 7457578.0,
+      "step": 3220
+    },
+    {
+      "epoch": 7.76041038020519,
+      "eval_entropy": 0.26581037203582486,
+      "eval_loss": 0.9146883487701416,
+      "eval_mean_token_accuracy": 0.8270345812433222,
+      "eval_num_tokens": 7457578.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 3220
+    },
+    {
+      "entropy": 0.11383061073720455,
+      "epoch": 7.808690404345202,
+      "grad_norm": 0.541247546672821,
+      "learning_rate": 5.258468515101918e-05,
+      "loss": 0.08942405581474304,
+      "mean_token_accuracy": 0.968398705124855,
+      "num_tokens": 7505761.0,
+      "step": 3240
+    },
+    {
+      "epoch": 7.808690404345202,
+      "eval_entropy": 0.26687998960861997,
+      "eval_loss": 0.9037085175514221,
+      "eval_mean_token_accuracy": 0.8274977863504646,
+      "eval_num_tokens": 7505761.0,
+      "eval_runtime": 88.5163,
+      "eval_samples_per_second": 16.042,
+      "eval_steps_per_second": 2.011,
+      "step": 3240
+    },
+    {
+      "entropy": 0.12004058873280883,
+      "epoch": 7.856970428485214,
+      "grad_norm": 0.5289552211761475,
+      "learning_rate": 5.040918622423899e-05,
+      "loss": 0.09064805507659912,
+      "mean_token_accuracy": 0.9673330299556255,
+      "num_tokens": 7552557.0,
+      "step": 3260
+    },
+    {
+      "epoch": 7.856970428485214,
+      "eval_entropy": 0.26288379251622085,
+      "eval_loss": 0.9146178364753723,
+      "eval_mean_token_accuracy": 0.8276662448149049,
+      "eval_num_tokens": 7552557.0,
+      "eval_runtime": 88.0723,
+      "eval_samples_per_second": 16.123,
+      "eval_steps_per_second": 2.021,
+      "step": 3260
+    },
+    {
+      "entropy": 0.1180689456872642,
+      "epoch": 7.905250452625227,
+      "grad_norm": 0.4754861891269684,
+      "learning_rate": 4.827266444161377e-05,
+      "loss": 0.09402655363082886,
+      "mean_token_accuracy": 0.9663361787796021,
+      "num_tokens": 7596998.0,
+      "step": 3280
+    },
+    {
+      "epoch": 7.905250452625227,
+      "eval_entropy": 0.261690996922134,
+      "eval_loss": 0.9120186567306519,
+      "eval_mean_token_accuracy": 0.8278995015647974,
+      "eval_num_tokens": 7596998.0,
+      "eval_runtime": 88.6573,
+      "eval_samples_per_second": 16.017,
+      "eval_steps_per_second": 2.008,
+      "step": 3280
+    },
+    {
+      "entropy": 0.1162164019420743,
+      "epoch": 7.953530476765239,
+      "grad_norm": 0.6203398108482361,
+      "learning_rate": 4.61757244135355e-05,
+      "loss": 0.08832768201828003,
+      "mean_token_accuracy": 0.9694220684468746,
+      "num_tokens": 7643288.0,
+      "step": 3300
+    },
+    {
+      "epoch": 7.953530476765239,
+      "eval_entropy": 0.26158678129817664,
+      "eval_loss": 0.9301361441612244,
+      "eval_mean_token_accuracy": 0.8269603885291668,
+      "eval_num_tokens": 7643288.0,
+      "eval_runtime": 88.7064,
+      "eval_samples_per_second": 16.008,
+      "eval_steps_per_second": 2.007,
+      "step": 3300
+    },
+    {
+      "entropy": 0.10877202204489089,
+      "epoch": 8.0,
+      "grad_norm": 3.058673143386841,
+      "learning_rate": 4.411895954922763e-05,
+      "loss": 0.09159615635871887,
+      "mean_token_accuracy": 0.9686878834451947,
+      "num_tokens": 7688608.0,
+      "step": 3320
+    },
+    {
+      "epoch": 8.0,
+      "eval_entropy": 0.2590438057197614,
+      "eval_loss": 0.9262065887451172,
+      "eval_mean_token_accuracy": 0.8277404546067956,
+      "eval_num_tokens": 7688608.0,
+      "eval_runtime": 87.7557,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 3320
+    },
+    {
+      "entropy": 0.09969092034734786,
+      "epoch": 8.048280024140013,
+      "grad_norm": 0.2561911642551422,
+      "learning_rate": 4.2102951888817644e-05,
+      "loss": 0.07228735685348511,
+      "mean_token_accuracy": 0.9743556551635265,
+      "num_tokens": 7732856.0,
+      "step": 3340
+    },
+    {
+      "epoch": 8.048280024140013,
+      "eval_entropy": 0.2462534454933713,
+      "eval_loss": 0.9755983352661133,
+      "eval_mean_token_accuracy": 0.827510209565752,
+      "eval_num_tokens": 7732856.0,
+      "eval_runtime": 88.0297,
+      "eval_samples_per_second": 16.131,
+      "eval_steps_per_second": 2.022,
+      "step": 3340
+    },
+    {
+      "entropy": 0.09477624730207027,
+      "epoch": 8.096560048280024,
+      "grad_norm": 0.3670356571674347,
+      "learning_rate": 4.012827193862633e-05,
+      "loss": 0.0686036467552185,
+      "mean_token_accuracy": 0.9743670664727688,
+      "num_tokens": 7778585.0,
+      "step": 3360
+    },
+    {
+      "epoch": 8.096560048280024,
+      "eval_entropy": 0.2440882951356052,
+      "eval_loss": 1.0118423700332642,
+      "eval_mean_token_accuracy": 0.8271241070849172,
+      "eval_num_tokens": 7778585.0,
+      "eval_runtime": 88.1364,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 3360
+    },
+    {
+      "entropy": 0.09417980071157217,
+      "epoch": 8.144840072420036,
+      "grad_norm": 0.5785111784934998,
+      "learning_rate": 3.819547850972133e-05,
+      "loss": 0.07100698947906495,
+      "mean_token_accuracy": 0.9737804099917412,
+      "num_tokens": 7824287.0,
+      "step": 3380
+    },
+    {
+      "epoch": 8.144840072420036,
+      "eval_entropy": 0.24603172860453637,
+      "eval_loss": 1.0021991729736328,
+      "eval_mean_token_accuracy": 0.827162122793412,
+      "eval_num_tokens": 7824287.0,
+      "eval_runtime": 88.4534,
+      "eval_samples_per_second": 16.054,
+      "eval_steps_per_second": 2.012,
+      "step": 3380
+    },
+    {
+      "entropy": 0.087222247896716,
+      "epoch": 8.193120096560047,
+      "grad_norm": 0.44848504662513733,
+      "learning_rate": 3.6305118559780393e-05,
+      "loss": 0.06325234174728393,
+      "mean_token_accuracy": 0.9769335232675076,
+      "num_tokens": 7874784.0,
+      "step": 3400
+    },
+    {
+      "epoch": 8.193120096560047,
+      "eval_entropy": 0.24544863905129807,
+      "eval_loss": 1.005008339881897,
+      "eval_mean_token_accuracy": 0.8273035240976998,
+      "eval_num_tokens": 7874784.0,
+      "eval_runtime": 88.6737,
+      "eval_samples_per_second": 16.014,
+      "eval_steps_per_second": 2.007,
+      "step": 3400
+    },
+    {
+      "entropy": 0.09037275062873959,
+      "epoch": 8.24140012070006,
+      "grad_norm": 0.23529155552387238,
+      "learning_rate": 3.4457727038308435e-05,
+      "loss": 0.06589569449424744,
+      "mean_token_accuracy": 0.9751470290124417,
+      "num_tokens": 7925240.0,
+      "step": 3420
+    },
+    {
+      "epoch": 8.24140012070006,
+      "eval_entropy": 0.24203423494368456,
+      "eval_loss": 1.0052824020385742,
+      "eval_mean_token_accuracy": 0.8275575503874361,
+      "eval_num_tokens": 7925240.0,
+      "eval_runtime": 88.3486,
+      "eval_samples_per_second": 16.073,
+      "eval_steps_per_second": 2.015,
+      "step": 3420
+    },
+    {
+      "entropy": 0.10373208015225828,
+      "epoch": 8.289680144840073,
+      "grad_norm": 0.24796408414840698,
+      "learning_rate": 3.2653826735253594e-05,
+      "loss": 0.07714771628379821,
+      "mean_token_accuracy": 0.9722891598939896,
+      "num_tokens": 7966685.0,
+      "step": 3440
+    },
+    {
+      "epoch": 8.289680144840073,
+      "eval_entropy": 0.23461157050025597,
+      "eval_loss": 1.012108325958252,
+      "eval_mean_token_accuracy": 0.82825900564033,
+      "eval_num_tokens": 7966685.0,
+      "eval_runtime": 88.1018,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 3440
+    },
+    {
+      "entropy": 0.08967954274266958,
+      "epoch": 8.337960168980084,
+      "grad_norm": 0.5314674973487854,
+      "learning_rate": 3.08939281330635e-05,
+      "loss": 0.06746266484260559,
+      "mean_token_accuracy": 0.9753133699297905,
+      "num_tokens": 8013656.0,
+      "step": 3460
+    },
+    {
+      "epoch": 8.337960168980084,
+      "eval_entropy": 0.23781998589467468,
+      "eval_loss": 1.0119913816452026,
+      "eval_mean_token_accuracy": 0.8273287157664139,
+      "eval_num_tokens": 8013656.0,
+      "eval_runtime": 88.6038,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3460
+    },
+    {
+      "entropy": 0.09280927330255509,
+      "epoch": 8.386240193120097,
+      "grad_norm": 0.30658936500549316,
+      "learning_rate": 2.9178529262224946e-05,
+      "loss": 0.06936421990394592,
+      "mean_token_accuracy": 0.9742105580866337,
+      "num_tokens": 8061108.0,
+      "step": 3480
+    },
+    {
+      "epoch": 8.386240193120097,
+      "eval_entropy": 0.23672426734747512,
+      "eval_loss": 1.0162041187286377,
+      "eval_mean_token_accuracy": 0.8278394281194451,
+      "eval_num_tokens": 8061108.0,
+      "eval_runtime": 88.6015,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 3480
+    },
+    {
+      "entropy": 0.09195122267119586,
+      "epoch": 8.43452021726011,
+      "grad_norm": 0.24512454867362976,
+      "learning_rate": 2.7508115560327256e-05,
+      "loss": 0.06996064782142639,
+      "mean_token_accuracy": 0.974104854464531,
+      "num_tokens": 8106924.0,
+      "step": 3500
+    },
+    {
+      "epoch": 8.43452021726011,
+      "eval_entropy": 0.23758245828781235,
+      "eval_loss": 1.0157747268676758,
+      "eval_mean_token_accuracy": 0.8278501579600773,
+      "eval_num_tokens": 8106924.0,
+      "eval_runtime": 88.5678,
+      "eval_samples_per_second": 16.033,
+      "eval_steps_per_second": 2.01,
+      "step": 3500
+    },
+    {
+      "entropy": 0.09518477590754629,
+      "epoch": 8.48280024140012,
+      "grad_norm": 0.2373228371143341,
+      "learning_rate": 2.5883159734688868e-05,
+      "loss": 0.06832797527313232,
+      "mean_token_accuracy": 0.9743806682527065,
+      "num_tokens": 8153070.0,
+      "step": 3520
+    },
+    {
+      "epoch": 8.48280024140012,
+      "eval_entropy": 0.2379308074210467,
+      "eval_loss": 1.0180530548095703,
+      "eval_mean_token_accuracy": 0.8282477052024241,
+      "eval_num_tokens": 8153070.0,
+      "eval_runtime": 88.4651,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 3520
+    },
+    {
+      "entropy": 0.0906269280705601,
+      "epoch": 8.531080265540133,
+      "grad_norm": 0.37602657079696655,
+      "learning_rate": 2.4304121628587173e-05,
+      "loss": 0.06921588778495788,
+      "mean_token_accuracy": 0.9745571479201317,
+      "num_tokens": 8200066.0,
+      "step": 3540
+    },
+    {
+      "epoch": 8.531080265540133,
+      "eval_entropy": 0.2357490769263064,
+      "eval_loss": 1.017139196395874,
+      "eval_mean_token_accuracy": 0.828713178299786,
+      "eval_num_tokens": 8200066.0,
+      "eval_runtime": 88.6424,
+      "eval_samples_per_second": 16.019,
+      "eval_steps_per_second": 2.008,
+      "step": 3540
+    },
+    {
+      "entropy": 0.09273334196768701,
+      "epoch": 8.579360289680144,
+      "grad_norm": 0.2978838086128235,
+      "learning_rate": 2.277144809112794e-05,
+      "loss": 0.06970887780189514,
+      "mean_token_accuracy": 0.9751374177634716,
+      "num_tokens": 8246157.0,
+      "step": 3560
+    },
+    {
+      "epoch": 8.579360289680144,
+      "eval_entropy": 0.23590933138065123,
+      "eval_loss": 1.019231915473938,
+      "eval_mean_token_accuracy": 0.8279163412163767,
+      "eval_num_tokens": 8246157.0,
+      "eval_runtime": 88.2691,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 3560
+    },
+    {
+      "entropy": 0.09854201921261847,
+      "epoch": 8.627640313820157,
+      "grad_norm": 0.5381243228912354,
+      "learning_rate": 2.1285572850792413e-05,
+      "loss": 0.07123501300811767,
+      "mean_token_accuracy": 0.9723988175392151,
+      "num_tokens": 8291752.0,
+      "step": 3580
+    },
+    {
+      "epoch": 8.627640313820157,
+      "eval_entropy": 0.23592024949494372,
+      "eval_loss": 1.020250678062439,
+      "eval_mean_token_accuracy": 0.8282466226079491,
+      "eval_num_tokens": 8291752.0,
+      "eval_runtime": 87.8939,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 3580
+    },
+    {
+      "entropy": 0.09092000788077712,
+      "epoch": 8.67592033796017,
+      "grad_norm": 0.29089561104774475,
+      "learning_rate": 1.9846916392697493e-05,
+      "loss": 0.06982720494270325,
+      "mean_token_accuracy": 0.9749995484948158,
+      "num_tokens": 8337651.0,
+      "step": 3600
+    },
+    {
+      "epoch": 8.67592033796017,
+      "eval_entropy": 0.23658913382318583,
+      "eval_loss": 1.019862413406372,
+      "eval_mean_token_accuracy": 0.8280127842104836,
+      "eval_num_tokens": 8337651.0,
+      "eval_runtime": 88.2855,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 3600
+    },
+    {
+      "entropy": 0.08708424470387399,
+      "epoch": 8.72420036210018,
+      "grad_norm": 0.29904839396476746,
+      "learning_rate": 1.8455885839603106e-05,
+      "loss": 0.06801370978355407,
+      "mean_token_accuracy": 0.9755801670253277,
+      "num_tokens": 8385330.0,
+      "step": 3620
+    },
+    {
+      "epoch": 8.72420036210018,
+      "eval_entropy": 0.23627035771862845,
+      "eval_loss": 1.0204569101333618,
+      "eval_mean_token_accuracy": 0.8279967894045155,
+      "eval_num_tokens": 8385330.0,
+      "eval_runtime": 88.1753,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 3620
+    },
+    {
+      "entropy": 0.0916255566291511,
+      "epoch": 8.772480386240193,
+      "grad_norm": 0.20571033656597137,
+      "learning_rate": 1.7112874836701502e-05,
+      "loss": 0.0692249894142151,
+      "mean_token_accuracy": 0.9745148062705994,
+      "num_tokens": 8432134.0,
+      "step": 3640
+    },
+    {
+      "epoch": 8.772480386240193,
+      "eval_entropy": 0.2340821772813797,
+      "eval_loss": 1.0212821960449219,
+      "eval_mean_token_accuracy": 0.828373622358515,
+      "eval_num_tokens": 8432134.0,
+      "eval_runtime": 87.2518,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 3640
+    },
+    {
+      "entropy": 0.0926868049427867,
+      "epoch": 8.820760410380204,
+      "grad_norm": 0.2618265450000763,
+      "learning_rate": 1.5818263440220203e-05,
+      "loss": 0.07338647246360779,
+      "mean_token_accuracy": 0.9724708944559097,
+      "num_tokens": 8476758.0,
+      "step": 3660
+    },
+    {
+      "epoch": 8.820760410380204,
+      "eval_entropy": 0.23637074575330433,
+      "eval_loss": 1.0186725854873657,
+      "eval_mean_token_accuracy": 0.8277707605549459,
+      "eval_num_tokens": 8476758.0,
+      "eval_runtime": 88.4695,
+      "eval_samples_per_second": 16.051,
+      "eval_steps_per_second": 2.012,
+      "step": 3660
+    },
+    {
+      "entropy": 0.09175782897509635,
+      "epoch": 8.869040434520217,
+      "grad_norm": 0.2889852821826935,
+      "learning_rate": 1.4572418009870604e-05,
+      "loss": 0.07066041231155396,
+      "mean_token_accuracy": 0.9725673109292984,
+      "num_tokens": 8523294.0,
+      "step": 3680
+    },
+    {
+      "epoch": 8.869040434520217,
+      "eval_entropy": 0.23945164404223473,
+      "eval_loss": 1.0145231485366821,
+      "eval_mean_token_accuracy": 0.8281202523895864,
+      "eval_num_tokens": 8523294.0,
+      "eval_runtime": 88.368,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 3680
+    },
+    {
+      "entropy": 0.09265611860901117,
+      "epoch": 8.91732045866023,
+      "grad_norm": 0.41092950105667114,
+      "learning_rate": 1.3375691105172625e-05,
+      "loss": 0.06975027322769164,
+      "mean_token_accuracy": 0.974636297672987,
+      "num_tokens": 8570358.0,
+      "step": 3700
+    },
+    {
+      "epoch": 8.91732045866023,
+      "eval_entropy": 0.239588919901446,
+      "eval_loss": 1.014059066772461,
+      "eval_mean_token_accuracy": 0.828028041660116,
+      "eval_num_tokens": 8570358.0,
+      "eval_runtime": 88.1039,
+      "eval_samples_per_second": 16.117,
+      "eval_steps_per_second": 2.02,
+      "step": 3700
+    },
+    {
+      "entropy": 0.0896465003490448,
+      "epoch": 8.965600482800241,
+      "grad_norm": 0.24000254273414612,
+      "learning_rate": 1.2228421385684305e-05,
+      "loss": 0.06743953824043274,
+      "mean_token_accuracy": 0.974006575345993,
+      "num_tokens": 8618246.0,
+      "step": 3720
+    },
+    {
+      "epoch": 8.965600482800241,
+      "eval_entropy": 0.2390461369846644,
+      "eval_loss": 1.011997938156128,
+      "eval_mean_token_accuracy": 0.8284533211354459,
+      "eval_num_tokens": 8618246.0,
+      "eval_runtime": 88.6202,
+      "eval_samples_per_second": 16.023,
+      "eval_steps_per_second": 2.009,
+      "step": 3720
+    },
+    {
+      "entropy": 0.09575109251520851,
+      "epoch": 9.012070006035003,
+      "grad_norm": 0.18088576197624207,
+      "learning_rate": 1.1130933515165429e-05,
+      "loss": 0.07295305132865906,
+      "mean_token_accuracy": 0.9739722857227573,
+      "num_tokens": 8661177.0,
+      "step": 3740
+    },
+    {
+      "epoch": 9.012070006035003,
+      "eval_entropy": 0.23728882387447892,
+      "eval_loss": 1.0134038925170898,
+      "eval_mean_token_accuracy": 0.8286508951294288,
+      "eval_num_tokens": 8661177.0,
+      "eval_runtime": 88.6434,
+      "eval_samples_per_second": 16.019,
+      "eval_steps_per_second": 2.008,
+      "step": 3740
+    },
+    {
+      "entropy": 0.08883201805874705,
+      "epoch": 9.060350030175016,
+      "grad_norm": 0.23120898008346558,
+      "learning_rate": 1.008353806970132e-05,
+      "loss": 0.06326142549514771,
+      "mean_token_accuracy": 0.9780020996928215,
+      "num_tokens": 8707602.0,
+      "step": 3760
+    },
+    {
+      "epoch": 9.060350030175016,
+      "eval_entropy": 0.23416905331142832,
+      "eval_loss": 1.0284100770950317,
+      "eval_mean_token_accuracy": 0.8284083213029283,
+      "eval_num_tokens": 8707602.0,
+      "eval_runtime": 88.2933,
+      "eval_samples_per_second": 16.083,
+      "eval_steps_per_second": 2.016,
+      "step": 3760
+    },
+    {
+      "entropy": 0.08861468276008963,
+      "epoch": 9.108630054315027,
+      "grad_norm": 0.17277656495571136,
+      "learning_rate": 9.086531449813489e-06,
+      "loss": 0.06727416515350342,
+      "mean_token_accuracy": 0.9760390736162663,
+      "num_tokens": 8751287.0,
+      "step": 3780
+    },
+    {
+      "epoch": 9.108630054315027,
+      "eval_entropy": 0.23370317386442355,
+      "eval_loss": 1.0381091833114624,
+      "eval_mean_token_accuracy": 0.8282668178670862,
+      "eval_num_tokens": 8751287.0,
+      "eval_runtime": 88.5738,
+      "eval_samples_per_second": 16.032,
+      "eval_steps_per_second": 2.01,
+      "step": 3780
+    },
+    {
+      "entropy": 0.09032954541034996,
+      "epoch": 9.15691007845504,
+      "grad_norm": 0.19481085240840912,
+      "learning_rate": 8.140195796581981e-06,
+      "loss": 0.06117786169052124,
+      "mean_token_accuracy": 0.9764454193413258,
+      "num_tokens": 8799982.0,
+      "step": 3800
+    },
+    {
+      "epoch": 9.15691007845504,
+      "eval_entropy": 0.23215936609868254,
+      "eval_loss": 1.0467767715454102,
+      "eval_mean_token_accuracy": 0.828405868806196,
+      "eval_num_tokens": 8799982.0,
+      "eval_runtime": 87.9551,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 3800
+    },
+    {
+      "entropy": 0.08824449330568314,
+      "epoch": 9.20519010259505,
+      "grad_norm": 0.20382793247699738,
+      "learning_rate": 7.2447989118025455e-06,
+      "loss": 0.06422553658485412,
+      "mean_token_accuracy": 0.9763121999800205,
+      "num_tokens": 8845536.0,
+      "step": 3820
+    },
+    {
+      "epoch": 9.20519010259505,
+      "eval_entropy": 0.23169661472352704,
+      "eval_loss": 1.0529043674468994,
+      "eval_mean_token_accuracy": 0.8282025610463003,
+      "eval_num_tokens": 8845536.0,
+      "eval_runtime": 88.5548,
+      "eval_samples_per_second": 16.035,
+      "eval_steps_per_second": 2.01,
+      "step": 3820
+    },
+    {
+      "entropy": 0.07768499720841646,
+      "epoch": 9.253470126735063,
+      "grad_norm": 0.11727919429540634,
+      "learning_rate": 6.400594182202084e-06,
+      "loss": 0.057279336452484134,
+      "mean_token_accuracy": 0.9795166745781898,
+      "num_tokens": 8896053.0,
+      "step": 3840
+    },
+    {
+      "epoch": 9.253470126735063,
+      "eval_entropy": 0.2301936373114586,
+      "eval_loss": 1.0576049089431763,
+      "eval_mean_token_accuracy": 0.8283417489421502,
+      "eval_num_tokens": 8896053.0,
+      "eval_runtime": 88.7854,
+      "eval_samples_per_second": 15.994,
+      "eval_steps_per_second": 2.005,
+      "step": 3840
+    },
+    {
+      "entropy": 0.09015200780704617,
+      "epoch": 9.301750150875076,
+      "grad_norm": 0.2089058756828308,
+      "learning_rate": 5.6078205077330705e-06,
+      "loss": 0.06626194715499878,
+      "mean_token_accuracy": 0.9747506938874722,
+      "num_tokens": 8940591.0,
+      "step": 3860
+    },
+    {
+      "epoch": 9.301750150875076,
+      "eval_entropy": 0.23024900676159377,
+      "eval_loss": 1.0599786043167114,
+      "eval_mean_token_accuracy": 0.8283051915383071,
+      "eval_num_tokens": 8940591.0,
+      "eval_runtime": 88.2963,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3860
+    },
+    {
+      "entropy": 0.08484094133600592,
+      "epoch": 9.350030175015087,
+      "grad_norm": 0.21646778285503387,
+      "learning_rate": 4.866702233967785e-06,
+      "loss": 0.05998039245605469,
+      "mean_token_accuracy": 0.9771868668496608,
+      "num_tokens": 8989035.0,
+      "step": 3880
+    },
+    {
+      "epoch": 9.350030175015087,
+      "eval_entropy": 0.2300723305578982,
+      "eval_loss": 1.0633587837219238,
+      "eval_mean_token_accuracy": 0.828371500701047,
+      "eval_num_tokens": 8989035.0,
+      "eval_runtime": 88.0233,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 3880
+    },
+    {
+      "entropy": 0.08905019680969417,
+      "epoch": 9.3983101991551,
+      "grad_norm": 0.21842406690120697,
+      "learning_rate": 4.177449088611095e-06,
+      "loss": 0.06666865348815917,
+      "mean_token_accuracy": 0.9753445640206337,
+      "num_tokens": 9033383.0,
+      "step": 3900
+    },
+    {
+      "epoch": 9.3983101991551,
+      "eval_entropy": 0.2295504841911659,
+      "eval_loss": 1.0662461519241333,
+      "eval_mean_token_accuracy": 0.8284122297603093,
+      "eval_num_tokens": 9033383.0,
+      "eval_runtime": 88.5525,
+      "eval_samples_per_second": 16.036,
+      "eval_steps_per_second": 2.01,
+      "step": 3900
+    },
+    {
+      "entropy": 0.08582521872594953,
+      "epoch": 9.446590223295111,
+      "grad_norm": 0.2000240683555603,
+      "learning_rate": 3.540256122150088e-06,
+      "loss": 0.06327944993972778,
+      "mean_token_accuracy": 0.9765432856976985,
+      "num_tokens": 9079672.0,
+      "step": 3920
+    },
+    {
+      "epoch": 9.446590223295111,
+      "eval_entropy": 0.2297678625818049,
+      "eval_loss": 1.0676169395446777,
+      "eval_mean_token_accuracy": 0.8283703879693921,
+      "eval_num_tokens": 9079672.0,
+      "eval_runtime": 87.6897,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 3920
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.8395843433007514e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3940/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3940/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3940/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3940/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3940/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3940/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3940/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3940/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3940/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3940/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3940/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3940/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3940/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3940/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..659ace35f2cef18fe37a3476dec1ba5a47377f86
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3940/trainer_state.json
@@ -0,0 +1,4171 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 9.494870247435124,
+  "eval_steps": 20,
+  "global_step": 3940,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    },
+    {
+      "entropy": 0.1326121488586068,
+      "epoch": 7.181050090525045,
+      "grad_norm": 0.4843117296695709,
+      "learning_rate": 8.411870386507724e-05,
+      "loss": 0.09537842869758606,
+      "mean_token_accuracy": 0.9645151920616627,
+      "num_tokens": 6898444.0,
+      "step": 2980
+    },
+    {
+      "epoch": 7.181050090525045,
+      "eval_entropy": 0.27999477783280813,
+      "eval_loss": 0.9031037092208862,
+      "eval_mean_token_accuracy": 0.8261882204018282,
+      "eval_num_tokens": 6898444.0,
+      "eval_runtime": 88.5602,
+      "eval_samples_per_second": 16.034,
+      "eval_steps_per_second": 2.01,
+      "step": 2980
+    },
+    {
+      "entropy": 0.1156751093454659,
+      "epoch": 7.229330114665057,
+      "grad_norm": 0.5297922492027283,
+      "learning_rate": 8.149629351430326e-05,
+      "loss": 0.09203402996063233,
+      "mean_token_accuracy": 0.9689733698964119,
+      "num_tokens": 6947106.0,
+      "step": 3000
+    },
+    {
+      "epoch": 7.229330114665057,
+      "eval_entropy": 0.2740225845508361,
+      "eval_loss": 0.8876440525054932,
+      "eval_mean_token_accuracy": 0.8265116857678703,
+      "eval_num_tokens": 6947106.0,
+      "eval_runtime": 87.7971,
+      "eval_samples_per_second": 16.174,
+      "eval_steps_per_second": 2.027,
+      "step": 3000
+    },
+    {
+      "entropy": 0.12116114404052496,
+      "epoch": 7.27761013880507,
+      "grad_norm": 0.5467398166656494,
+      "learning_rate": 7.890406302404618e-05,
+      "loss": 0.0933056652545929,
+      "mean_token_accuracy": 0.9676455594599247,
+      "num_tokens": 6992784.0,
+      "step": 3020
+    },
+    {
+      "epoch": 7.27761013880507,
+      "eval_entropy": 0.27238342793804876,
+      "eval_loss": 0.9166117310523987,
+      "eval_mean_token_accuracy": 0.825499015242866,
+      "eval_num_tokens": 6992784.0,
+      "eval_runtime": 88.3276,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 3020
+    },
+    {
+      "entropy": 0.11715119928121567,
+      "epoch": 7.325890162945082,
+      "grad_norm": 0.4976702332496643,
+      "learning_rate": 7.634274596487176e-05,
+      "loss": 0.09449659585952759,
+      "mean_token_accuracy": 0.966506478190422,
+      "num_tokens": 7040482.0,
+      "step": 3040
+    },
+    {
+      "epoch": 7.325890162945082,
+      "eval_entropy": 0.2788975556077582,
+      "eval_loss": 0.8833065629005432,
+      "eval_mean_token_accuracy": 0.8269920596915684,
+      "eval_num_tokens": 7040482.0,
+      "eval_runtime": 88.9016,
+      "eval_samples_per_second": 15.973,
+      "eval_steps_per_second": 2.002,
+      "step": 3040
+    },
+    {
+      "entropy": 0.12565911412239075,
+      "epoch": 7.3741701870850935,
+      "grad_norm": 0.5728679895401001,
+      "learning_rate": 7.381306715921025e-05,
+      "loss": 0.09626244902610778,
+      "mean_token_accuracy": 0.9661444149911403,
+      "num_tokens": 7086261.0,
+      "step": 3060
+    },
+    {
+      "epoch": 7.3741701870850935,
+      "eval_entropy": 0.2762492011939542,
+      "eval_loss": 0.8961160778999329,
+      "eval_mean_token_accuracy": 0.826369910092836,
+      "eval_num_tokens": 7086261.0,
+      "eval_runtime": 88.5204,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 3060
+    },
+    {
+      "entropy": 0.12258567921817302,
+      "epoch": 7.422450211225105,
+      "grad_norm": 0.468985915184021,
+      "learning_rate": 7.131574247624031e-05,
+      "loss": 0.09120266437530518,
+      "mean_token_accuracy": 0.9674938447773457,
+      "num_tokens": 7133247.0,
+      "step": 3080
+    },
+    {
+      "epoch": 7.422450211225105,
+      "eval_entropy": 0.2688512086366,
+      "eval_loss": 0.9048654437065125,
+      "eval_mean_token_accuracy": 0.8265582302982888,
+      "eval_num_tokens": 7133247.0,
+      "eval_runtime": 88.164,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 3080
+    },
+    {
+      "entropy": 0.12070536343380808,
+      "epoch": 7.470730235365117,
+      "grad_norm": 0.5310841798782349,
+      "learning_rate": 6.885147862930688e-05,
+      "loss": 0.09386247992515565,
+      "mean_token_accuracy": 0.96664222702384,
+      "num_tokens": 7177810.0,
+      "step": 3100
+    },
+    {
+      "epoch": 7.470730235365117,
+      "eval_entropy": 0.26289632874593305,
+      "eval_loss": 0.902837336063385,
+      "eval_mean_token_accuracy": 0.828052184554968,
+      "eval_num_tokens": 7177810.0,
+      "eval_runtime": 88.1201,
+      "eval_samples_per_second": 16.114,
+      "eval_steps_per_second": 2.02,
+      "step": 3100
+    },
+    {
+      "entropy": 0.11322665181942285,
+      "epoch": 7.51901025950513,
+      "grad_norm": 0.6846067905426025,
+      "learning_rate": 6.642097297592908e-05,
+      "loss": 0.08772538304328918,
+      "mean_token_accuracy": 0.9691362135112286,
+      "num_tokens": 7225302.0,
+      "step": 3120
+    },
+    {
+      "epoch": 7.51901025950513,
+      "eval_entropy": 0.2670605613776807,
+      "eval_loss": 0.9130566120147705,
+      "eval_mean_token_accuracy": 0.826366903741708,
+      "eval_num_tokens": 7225302.0,
+      "eval_runtime": 88.6035,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3120
+    },
+    {
+      "entropy": 0.10985926580615342,
+      "epoch": 7.567290283645142,
+      "grad_norm": 0.4799748361110687,
+      "learning_rate": 6.402491332045683e-05,
+      "loss": 0.08672296404838561,
+      "mean_token_accuracy": 0.9698824048042297,
+      "num_tokens": 7274881.0,
+      "step": 3140
+    },
+    {
+      "epoch": 7.567290283645142,
+      "eval_entropy": 0.26657010866015146,
+      "eval_loss": 0.9056380987167358,
+      "eval_mean_token_accuracy": 0.8272326461384806,
+      "eval_num_tokens": 7274881.0,
+      "eval_runtime": 88.2967,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3140
+    },
+    {
+      "entropy": 0.11890473761595785,
+      "epoch": 7.615570307785154,
+      "grad_norm": 0.4081912934780121,
+      "learning_rate": 6.166397771942987e-05,
+      "loss": 0.09396827220916748,
+      "mean_token_accuracy": 0.967512384802103,
+      "num_tokens": 7321163.0,
+      "step": 3160
+    },
+    {
+      "epoch": 7.615570307785154,
+      "eval_entropy": 0.26364776575833226,
+      "eval_loss": 0.9204733967781067,
+      "eval_mean_token_accuracy": 0.827419729714983,
+      "eval_num_tokens": 7321163.0,
+      "eval_runtime": 88.3441,
+      "eval_samples_per_second": 16.074,
+      "eval_steps_per_second": 2.015,
+      "step": 3160
+    },
+    {
+      "entropy": 0.12209134344011545,
+      "epoch": 7.663850331925166,
+      "grad_norm": 0.5159774422645569,
+      "learning_rate": 5.93388342896962e-05,
+      "loss": 0.09292978644371033,
+      "mean_token_accuracy": 0.9658421002328396,
+      "num_tokens": 7366217.0,
+      "step": 3180
+    },
+    {
+      "epoch": 7.663850331925166,
+      "eval_entropy": 0.2714110068725736,
+      "eval_loss": 0.9143092036247253,
+      "eval_mean_token_accuracy": 0.8262251099843657,
+      "eval_num_tokens": 7366217.0,
+      "eval_runtime": 87.9635,
+      "eval_samples_per_second": 16.143,
+      "eval_steps_per_second": 2.024,
+      "step": 3180
+    },
+    {
+      "entropy": 0.11955609126016498,
+      "epoch": 7.712130356065178,
+      "grad_norm": 0.7005976438522339,
+      "learning_rate": 5.705014101934252e-05,
+      "loss": 0.09319776892662049,
+      "mean_token_accuracy": 0.9674513377249241,
+      "num_tokens": 7410887.0,
+      "step": 3200
+    },
+    {
+      "epoch": 7.712130356065178,
+      "eval_entropy": 0.2626291513442993,
+      "eval_loss": 0.91884446144104,
+      "eval_mean_token_accuracy": 0.827563955207889,
+      "eval_num_tokens": 7410887.0,
+      "eval_runtime": 87.8849,
+      "eval_samples_per_second": 16.157,
+      "eval_steps_per_second": 2.025,
+      "step": 3200
+    },
+    {
+      "entropy": 0.11767532899975777,
+      "epoch": 7.76041038020519,
+      "grad_norm": 0.4675362706184387,
+      "learning_rate": 5.479854558149187e-05,
+      "loss": 0.09038087725639343,
+      "mean_token_accuracy": 0.9669680051505566,
+      "num_tokens": 7457578.0,
+      "step": 3220
+    },
+    {
+      "epoch": 7.76041038020519,
+      "eval_entropy": 0.26581037203582486,
+      "eval_loss": 0.9146883487701416,
+      "eval_mean_token_accuracy": 0.8270345812433222,
+      "eval_num_tokens": 7457578.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 3220
+    },
+    {
+      "entropy": 0.11383061073720455,
+      "epoch": 7.808690404345202,
+      "grad_norm": 0.541247546672821,
+      "learning_rate": 5.258468515101918e-05,
+      "loss": 0.08942405581474304,
+      "mean_token_accuracy": 0.968398705124855,
+      "num_tokens": 7505761.0,
+      "step": 3240
+    },
+    {
+      "epoch": 7.808690404345202,
+      "eval_entropy": 0.26687998960861997,
+      "eval_loss": 0.9037085175514221,
+      "eval_mean_token_accuracy": 0.8274977863504646,
+      "eval_num_tokens": 7505761.0,
+      "eval_runtime": 88.5163,
+      "eval_samples_per_second": 16.042,
+      "eval_steps_per_second": 2.011,
+      "step": 3240
+    },
+    {
+      "entropy": 0.12004058873280883,
+      "epoch": 7.856970428485214,
+      "grad_norm": 0.5289552211761475,
+      "learning_rate": 5.040918622423899e-05,
+      "loss": 0.09064805507659912,
+      "mean_token_accuracy": 0.9673330299556255,
+      "num_tokens": 7552557.0,
+      "step": 3260
+    },
+    {
+      "epoch": 7.856970428485214,
+      "eval_entropy": 0.26288379251622085,
+      "eval_loss": 0.9146178364753723,
+      "eval_mean_token_accuracy": 0.8276662448149049,
+      "eval_num_tokens": 7552557.0,
+      "eval_runtime": 88.0723,
+      "eval_samples_per_second": 16.123,
+      "eval_steps_per_second": 2.021,
+      "step": 3260
+    },
+    {
+      "entropy": 0.1180689456872642,
+      "epoch": 7.905250452625227,
+      "grad_norm": 0.4754861891269684,
+      "learning_rate": 4.827266444161377e-05,
+      "loss": 0.09402655363082886,
+      "mean_token_accuracy": 0.9663361787796021,
+      "num_tokens": 7596998.0,
+      "step": 3280
+    },
+    {
+      "epoch": 7.905250452625227,
+      "eval_entropy": 0.261690996922134,
+      "eval_loss": 0.9120186567306519,
+      "eval_mean_token_accuracy": 0.8278995015647974,
+      "eval_num_tokens": 7596998.0,
+      "eval_runtime": 88.6573,
+      "eval_samples_per_second": 16.017,
+      "eval_steps_per_second": 2.008,
+      "step": 3280
+    },
+    {
+      "entropy": 0.1162164019420743,
+      "epoch": 7.953530476765239,
+      "grad_norm": 0.6203398108482361,
+      "learning_rate": 4.61757244135355e-05,
+      "loss": 0.08832768201828003,
+      "mean_token_accuracy": 0.9694220684468746,
+      "num_tokens": 7643288.0,
+      "step": 3300
+    },
+    {
+      "epoch": 7.953530476765239,
+      "eval_entropy": 0.26158678129817664,
+      "eval_loss": 0.9301361441612244,
+      "eval_mean_token_accuracy": 0.8269603885291668,
+      "eval_num_tokens": 7643288.0,
+      "eval_runtime": 88.7064,
+      "eval_samples_per_second": 16.008,
+      "eval_steps_per_second": 2.007,
+      "step": 3300
+    },
+    {
+      "entropy": 0.10877202204489089,
+      "epoch": 8.0,
+      "grad_norm": 3.058673143386841,
+      "learning_rate": 4.411895954922763e-05,
+      "loss": 0.09159615635871887,
+      "mean_token_accuracy": 0.9686878834451947,
+      "num_tokens": 7688608.0,
+      "step": 3320
+    },
+    {
+      "epoch": 8.0,
+      "eval_entropy": 0.2590438057197614,
+      "eval_loss": 0.9262065887451172,
+      "eval_mean_token_accuracy": 0.8277404546067956,
+      "eval_num_tokens": 7688608.0,
+      "eval_runtime": 87.7557,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 3320
+    },
+    {
+      "entropy": 0.09969092034734786,
+      "epoch": 8.048280024140013,
+      "grad_norm": 0.2561911642551422,
+      "learning_rate": 4.2102951888817644e-05,
+      "loss": 0.07228735685348511,
+      "mean_token_accuracy": 0.9743556551635265,
+      "num_tokens": 7732856.0,
+      "step": 3340
+    },
+    {
+      "epoch": 8.048280024140013,
+      "eval_entropy": 0.2462534454933713,
+      "eval_loss": 0.9755983352661133,
+      "eval_mean_token_accuracy": 0.827510209565752,
+      "eval_num_tokens": 7732856.0,
+      "eval_runtime": 88.0297,
+      "eval_samples_per_second": 16.131,
+      "eval_steps_per_second": 2.022,
+      "step": 3340
+    },
+    {
+      "entropy": 0.09477624730207027,
+      "epoch": 8.096560048280024,
+      "grad_norm": 0.3670356571674347,
+      "learning_rate": 4.012827193862633e-05,
+      "loss": 0.0686036467552185,
+      "mean_token_accuracy": 0.9743670664727688,
+      "num_tokens": 7778585.0,
+      "step": 3360
+    },
+    {
+      "epoch": 8.096560048280024,
+      "eval_entropy": 0.2440882951356052,
+      "eval_loss": 1.0118423700332642,
+      "eval_mean_token_accuracy": 0.8271241070849172,
+      "eval_num_tokens": 7778585.0,
+      "eval_runtime": 88.1364,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 3360
+    },
+    {
+      "entropy": 0.09417980071157217,
+      "epoch": 8.144840072420036,
+      "grad_norm": 0.5785111784934998,
+      "learning_rate": 3.819547850972133e-05,
+      "loss": 0.07100698947906495,
+      "mean_token_accuracy": 0.9737804099917412,
+      "num_tokens": 7824287.0,
+      "step": 3380
+    },
+    {
+      "epoch": 8.144840072420036,
+      "eval_entropy": 0.24603172860453637,
+      "eval_loss": 1.0021991729736328,
+      "eval_mean_token_accuracy": 0.827162122793412,
+      "eval_num_tokens": 7824287.0,
+      "eval_runtime": 88.4534,
+      "eval_samples_per_second": 16.054,
+      "eval_steps_per_second": 2.012,
+      "step": 3380
+    },
+    {
+      "entropy": 0.087222247896716,
+      "epoch": 8.193120096560047,
+      "grad_norm": 0.44848504662513733,
+      "learning_rate": 3.6305118559780393e-05,
+      "loss": 0.06325234174728393,
+      "mean_token_accuracy": 0.9769335232675076,
+      "num_tokens": 7874784.0,
+      "step": 3400
+    },
+    {
+      "epoch": 8.193120096560047,
+      "eval_entropy": 0.24544863905129807,
+      "eval_loss": 1.005008339881897,
+      "eval_mean_token_accuracy": 0.8273035240976998,
+      "eval_num_tokens": 7874784.0,
+      "eval_runtime": 88.6737,
+      "eval_samples_per_second": 16.014,
+      "eval_steps_per_second": 2.007,
+      "step": 3400
+    },
+    {
+      "entropy": 0.09037275062873959,
+      "epoch": 8.24140012070006,
+      "grad_norm": 0.23529155552387238,
+      "learning_rate": 3.4457727038308435e-05,
+      "loss": 0.06589569449424744,
+      "mean_token_accuracy": 0.9751470290124417,
+      "num_tokens": 7925240.0,
+      "step": 3420
+    },
+    {
+      "epoch": 8.24140012070006,
+      "eval_entropy": 0.24203423494368456,
+      "eval_loss": 1.0052824020385742,
+      "eval_mean_token_accuracy": 0.8275575503874361,
+      "eval_num_tokens": 7925240.0,
+      "eval_runtime": 88.3486,
+      "eval_samples_per_second": 16.073,
+      "eval_steps_per_second": 2.015,
+      "step": 3420
+    },
+    {
+      "entropy": 0.10373208015225828,
+      "epoch": 8.289680144840073,
+      "grad_norm": 0.24796408414840698,
+      "learning_rate": 3.2653826735253594e-05,
+      "loss": 0.07714771628379821,
+      "mean_token_accuracy": 0.9722891598939896,
+      "num_tokens": 7966685.0,
+      "step": 3440
+    },
+    {
+      "epoch": 8.289680144840073,
+      "eval_entropy": 0.23461157050025597,
+      "eval_loss": 1.012108325958252,
+      "eval_mean_token_accuracy": 0.82825900564033,
+      "eval_num_tokens": 7966685.0,
+      "eval_runtime": 88.1018,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 3440
+    },
+    {
+      "entropy": 0.08967954274266958,
+      "epoch": 8.337960168980084,
+      "grad_norm": 0.5314674973487854,
+      "learning_rate": 3.08939281330635e-05,
+      "loss": 0.06746266484260559,
+      "mean_token_accuracy": 0.9753133699297905,
+      "num_tokens": 8013656.0,
+      "step": 3460
+    },
+    {
+      "epoch": 8.337960168980084,
+      "eval_entropy": 0.23781998589467468,
+      "eval_loss": 1.0119913816452026,
+      "eval_mean_token_accuracy": 0.8273287157664139,
+      "eval_num_tokens": 8013656.0,
+      "eval_runtime": 88.6038,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3460
+    },
+    {
+      "entropy": 0.09280927330255509,
+      "epoch": 8.386240193120097,
+      "grad_norm": 0.30658936500549316,
+      "learning_rate": 2.9178529262224946e-05,
+      "loss": 0.06936421990394592,
+      "mean_token_accuracy": 0.9742105580866337,
+      "num_tokens": 8061108.0,
+      "step": 3480
+    },
+    {
+      "epoch": 8.386240193120097,
+      "eval_entropy": 0.23672426734747512,
+      "eval_loss": 1.0162041187286377,
+      "eval_mean_token_accuracy": 0.8278394281194451,
+      "eval_num_tokens": 8061108.0,
+      "eval_runtime": 88.6015,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 3480
+    },
+    {
+      "entropy": 0.09195122267119586,
+      "epoch": 8.43452021726011,
+      "grad_norm": 0.24512454867362976,
+      "learning_rate": 2.7508115560327256e-05,
+      "loss": 0.06996064782142639,
+      "mean_token_accuracy": 0.974104854464531,
+      "num_tokens": 8106924.0,
+      "step": 3500
+    },
+    {
+      "epoch": 8.43452021726011,
+      "eval_entropy": 0.23758245828781235,
+      "eval_loss": 1.0157747268676758,
+      "eval_mean_token_accuracy": 0.8278501579600773,
+      "eval_num_tokens": 8106924.0,
+      "eval_runtime": 88.5678,
+      "eval_samples_per_second": 16.033,
+      "eval_steps_per_second": 2.01,
+      "step": 3500
+    },
+    {
+      "entropy": 0.09518477590754629,
+      "epoch": 8.48280024140012,
+      "grad_norm": 0.2373228371143341,
+      "learning_rate": 2.5883159734688868e-05,
+      "loss": 0.06832797527313232,
+      "mean_token_accuracy": 0.9743806682527065,
+      "num_tokens": 8153070.0,
+      "step": 3520
+    },
+    {
+      "epoch": 8.48280024140012,
+      "eval_entropy": 0.2379308074210467,
+      "eval_loss": 1.0180530548095703,
+      "eval_mean_token_accuracy": 0.8282477052024241,
+      "eval_num_tokens": 8153070.0,
+      "eval_runtime": 88.4651,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 3520
+    },
+    {
+      "entropy": 0.0906269280705601,
+      "epoch": 8.531080265540133,
+      "grad_norm": 0.37602657079696655,
+      "learning_rate": 2.4304121628587173e-05,
+      "loss": 0.06921588778495788,
+      "mean_token_accuracy": 0.9745571479201317,
+      "num_tokens": 8200066.0,
+      "step": 3540
+    },
+    {
+      "epoch": 8.531080265540133,
+      "eval_entropy": 0.2357490769263064,
+      "eval_loss": 1.017139196395874,
+      "eval_mean_token_accuracy": 0.828713178299786,
+      "eval_num_tokens": 8200066.0,
+      "eval_runtime": 88.6424,
+      "eval_samples_per_second": 16.019,
+      "eval_steps_per_second": 2.008,
+      "step": 3540
+    },
+    {
+      "entropy": 0.09273334196768701,
+      "epoch": 8.579360289680144,
+      "grad_norm": 0.2978838086128235,
+      "learning_rate": 2.277144809112794e-05,
+      "loss": 0.06970887780189514,
+      "mean_token_accuracy": 0.9751374177634716,
+      "num_tokens": 8246157.0,
+      "step": 3560
+    },
+    {
+      "epoch": 8.579360289680144,
+      "eval_entropy": 0.23590933138065123,
+      "eval_loss": 1.019231915473938,
+      "eval_mean_token_accuracy": 0.8279163412163767,
+      "eval_num_tokens": 8246157.0,
+      "eval_runtime": 88.2691,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 3560
+    },
+    {
+      "entropy": 0.09854201921261847,
+      "epoch": 8.627640313820157,
+      "grad_norm": 0.5381243228912354,
+      "learning_rate": 2.1285572850792413e-05,
+      "loss": 0.07123501300811767,
+      "mean_token_accuracy": 0.9723988175392151,
+      "num_tokens": 8291752.0,
+      "step": 3580
+    },
+    {
+      "epoch": 8.627640313820157,
+      "eval_entropy": 0.23592024949494372,
+      "eval_loss": 1.020250678062439,
+      "eval_mean_token_accuracy": 0.8282466226079491,
+      "eval_num_tokens": 8291752.0,
+      "eval_runtime": 87.8939,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 3580
+    },
+    {
+      "entropy": 0.09092000788077712,
+      "epoch": 8.67592033796017,
+      "grad_norm": 0.29089561104774475,
+      "learning_rate": 1.9846916392697493e-05,
+      "loss": 0.06982720494270325,
+      "mean_token_accuracy": 0.9749995484948158,
+      "num_tokens": 8337651.0,
+      "step": 3600
+    },
+    {
+      "epoch": 8.67592033796017,
+      "eval_entropy": 0.23658913382318583,
+      "eval_loss": 1.019862413406372,
+      "eval_mean_token_accuracy": 0.8280127842104836,
+      "eval_num_tokens": 8337651.0,
+      "eval_runtime": 88.2855,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 3600
+    },
+    {
+      "entropy": 0.08708424470387399,
+      "epoch": 8.72420036210018,
+      "grad_norm": 0.29904839396476746,
+      "learning_rate": 1.8455885839603106e-05,
+      "loss": 0.06801370978355407,
+      "mean_token_accuracy": 0.9755801670253277,
+      "num_tokens": 8385330.0,
+      "step": 3620
+    },
+    {
+      "epoch": 8.72420036210018,
+      "eval_entropy": 0.23627035771862845,
+      "eval_loss": 1.0204569101333618,
+      "eval_mean_token_accuracy": 0.8279967894045155,
+      "eval_num_tokens": 8385330.0,
+      "eval_runtime": 88.1753,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 3620
+    },
+    {
+      "entropy": 0.0916255566291511,
+      "epoch": 8.772480386240193,
+      "grad_norm": 0.20571033656597137,
+      "learning_rate": 1.7112874836701502e-05,
+      "loss": 0.0692249894142151,
+      "mean_token_accuracy": 0.9745148062705994,
+      "num_tokens": 8432134.0,
+      "step": 3640
+    },
+    {
+      "epoch": 8.772480386240193,
+      "eval_entropy": 0.2340821772813797,
+      "eval_loss": 1.0212821960449219,
+      "eval_mean_token_accuracy": 0.828373622358515,
+      "eval_num_tokens": 8432134.0,
+      "eval_runtime": 87.2518,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 3640
+    },
+    {
+      "entropy": 0.0926868049427867,
+      "epoch": 8.820760410380204,
+      "grad_norm": 0.2618265450000763,
+      "learning_rate": 1.5818263440220203e-05,
+      "loss": 0.07338647246360779,
+      "mean_token_accuracy": 0.9724708944559097,
+      "num_tokens": 8476758.0,
+      "step": 3660
+    },
+    {
+      "epoch": 8.820760410380204,
+      "eval_entropy": 0.23637074575330433,
+      "eval_loss": 1.0186725854873657,
+      "eval_mean_token_accuracy": 0.8277707605549459,
+      "eval_num_tokens": 8476758.0,
+      "eval_runtime": 88.4695,
+      "eval_samples_per_second": 16.051,
+      "eval_steps_per_second": 2.012,
+      "step": 3660
+    },
+    {
+      "entropy": 0.09175782897509635,
+      "epoch": 8.869040434520217,
+      "grad_norm": 0.2889852821826935,
+      "learning_rate": 1.4572418009870604e-05,
+      "loss": 0.07066041231155396,
+      "mean_token_accuracy": 0.9725673109292984,
+      "num_tokens": 8523294.0,
+      "step": 3680
+    },
+    {
+      "epoch": 8.869040434520217,
+      "eval_entropy": 0.23945164404223473,
+      "eval_loss": 1.0145231485366821,
+      "eval_mean_token_accuracy": 0.8281202523895864,
+      "eval_num_tokens": 8523294.0,
+      "eval_runtime": 88.368,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 3680
+    },
+    {
+      "entropy": 0.09265611860901117,
+      "epoch": 8.91732045866023,
+      "grad_norm": 0.41092950105667114,
+      "learning_rate": 1.3375691105172625e-05,
+      "loss": 0.06975027322769164,
+      "mean_token_accuracy": 0.974636297672987,
+      "num_tokens": 8570358.0,
+      "step": 3700
+    },
+    {
+      "epoch": 8.91732045866023,
+      "eval_entropy": 0.239588919901446,
+      "eval_loss": 1.014059066772461,
+      "eval_mean_token_accuracy": 0.828028041660116,
+      "eval_num_tokens": 8570358.0,
+      "eval_runtime": 88.1039,
+      "eval_samples_per_second": 16.117,
+      "eval_steps_per_second": 2.02,
+      "step": 3700
+    },
+    {
+      "entropy": 0.0896465003490448,
+      "epoch": 8.965600482800241,
+      "grad_norm": 0.24000254273414612,
+      "learning_rate": 1.2228421385684305e-05,
+      "loss": 0.06743953824043274,
+      "mean_token_accuracy": 0.974006575345993,
+      "num_tokens": 8618246.0,
+      "step": 3720
+    },
+    {
+      "epoch": 8.965600482800241,
+      "eval_entropy": 0.2390461369846644,
+      "eval_loss": 1.011997938156128,
+      "eval_mean_token_accuracy": 0.8284533211354459,
+      "eval_num_tokens": 8618246.0,
+      "eval_runtime": 88.6202,
+      "eval_samples_per_second": 16.023,
+      "eval_steps_per_second": 2.009,
+      "step": 3720
+    },
+    {
+      "entropy": 0.09575109251520851,
+      "epoch": 9.012070006035003,
+      "grad_norm": 0.18088576197624207,
+      "learning_rate": 1.1130933515165429e-05,
+      "loss": 0.07295305132865906,
+      "mean_token_accuracy": 0.9739722857227573,
+      "num_tokens": 8661177.0,
+      "step": 3740
+    },
+    {
+      "epoch": 9.012070006035003,
+      "eval_entropy": 0.23728882387447892,
+      "eval_loss": 1.0134038925170898,
+      "eval_mean_token_accuracy": 0.8286508951294288,
+      "eval_num_tokens": 8661177.0,
+      "eval_runtime": 88.6434,
+      "eval_samples_per_second": 16.019,
+      "eval_steps_per_second": 2.008,
+      "step": 3740
+    },
+    {
+      "entropy": 0.08883201805874705,
+      "epoch": 9.060350030175016,
+      "grad_norm": 0.23120898008346558,
+      "learning_rate": 1.008353806970132e-05,
+      "loss": 0.06326142549514771,
+      "mean_token_accuracy": 0.9780020996928215,
+      "num_tokens": 8707602.0,
+      "step": 3760
+    },
+    {
+      "epoch": 9.060350030175016,
+      "eval_entropy": 0.23416905331142832,
+      "eval_loss": 1.0284100770950317,
+      "eval_mean_token_accuracy": 0.8284083213029283,
+      "eval_num_tokens": 8707602.0,
+      "eval_runtime": 88.2933,
+      "eval_samples_per_second": 16.083,
+      "eval_steps_per_second": 2.016,
+      "step": 3760
+    },
+    {
+      "entropy": 0.08861468276008963,
+      "epoch": 9.108630054315027,
+      "grad_norm": 0.17277656495571136,
+      "learning_rate": 9.086531449813489e-06,
+      "loss": 0.06727416515350342,
+      "mean_token_accuracy": 0.9760390736162663,
+      "num_tokens": 8751287.0,
+      "step": 3780
+    },
+    {
+      "epoch": 9.108630054315027,
+      "eval_entropy": 0.23370317386442355,
+      "eval_loss": 1.0381091833114624,
+      "eval_mean_token_accuracy": 0.8282668178670862,
+      "eval_num_tokens": 8751287.0,
+      "eval_runtime": 88.5738,
+      "eval_samples_per_second": 16.032,
+      "eval_steps_per_second": 2.01,
+      "step": 3780
+    },
+    {
+      "entropy": 0.09032954541034996,
+      "epoch": 9.15691007845504,
+      "grad_norm": 0.19481085240840912,
+      "learning_rate": 8.140195796581981e-06,
+      "loss": 0.06117786169052124,
+      "mean_token_accuracy": 0.9764454193413258,
+      "num_tokens": 8799982.0,
+      "step": 3800
+    },
+    {
+      "epoch": 9.15691007845504,
+      "eval_entropy": 0.23215936609868254,
+      "eval_loss": 1.0467767715454102,
+      "eval_mean_token_accuracy": 0.828405868806196,
+      "eval_num_tokens": 8799982.0,
+      "eval_runtime": 87.9551,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 3800
+    },
+    {
+      "entropy": 0.08824449330568314,
+      "epoch": 9.20519010259505,
+      "grad_norm": 0.20382793247699738,
+      "learning_rate": 7.2447989118025455e-06,
+      "loss": 0.06422553658485412,
+      "mean_token_accuracy": 0.9763121999800205,
+      "num_tokens": 8845536.0,
+      "step": 3820
+    },
+    {
+      "epoch": 9.20519010259505,
+      "eval_entropy": 0.23169661472352704,
+      "eval_loss": 1.0529043674468994,
+      "eval_mean_token_accuracy": 0.8282025610463003,
+      "eval_num_tokens": 8845536.0,
+      "eval_runtime": 88.5548,
+      "eval_samples_per_second": 16.035,
+      "eval_steps_per_second": 2.01,
+      "step": 3820
+    },
+    {
+      "entropy": 0.07768499720841646,
+      "epoch": 9.253470126735063,
+      "grad_norm": 0.11727919429540634,
+      "learning_rate": 6.400594182202084e-06,
+      "loss": 0.057279336452484134,
+      "mean_token_accuracy": 0.9795166745781898,
+      "num_tokens": 8896053.0,
+      "step": 3840
+    },
+    {
+      "epoch": 9.253470126735063,
+      "eval_entropy": 0.2301936373114586,
+      "eval_loss": 1.0576049089431763,
+      "eval_mean_token_accuracy": 0.8283417489421502,
+      "eval_num_tokens": 8896053.0,
+      "eval_runtime": 88.7854,
+      "eval_samples_per_second": 15.994,
+      "eval_steps_per_second": 2.005,
+      "step": 3840
+    },
+    {
+      "entropy": 0.09015200780704617,
+      "epoch": 9.301750150875076,
+      "grad_norm": 0.2089058756828308,
+      "learning_rate": 5.6078205077330705e-06,
+      "loss": 0.06626194715499878,
+      "mean_token_accuracy": 0.9747506938874722,
+      "num_tokens": 8940591.0,
+      "step": 3860
+    },
+    {
+      "epoch": 9.301750150875076,
+      "eval_entropy": 0.23024900676159377,
+      "eval_loss": 1.0599786043167114,
+      "eval_mean_token_accuracy": 0.8283051915383071,
+      "eval_num_tokens": 8940591.0,
+      "eval_runtime": 88.2963,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3860
+    },
+    {
+      "entropy": 0.08484094133600592,
+      "epoch": 9.350030175015087,
+      "grad_norm": 0.21646778285503387,
+      "learning_rate": 4.866702233967785e-06,
+      "loss": 0.05998039245605469,
+      "mean_token_accuracy": 0.9771868668496608,
+      "num_tokens": 8989035.0,
+      "step": 3880
+    },
+    {
+      "epoch": 9.350030175015087,
+      "eval_entropy": 0.2300723305578982,
+      "eval_loss": 1.0633587837219238,
+      "eval_mean_token_accuracy": 0.828371500701047,
+      "eval_num_tokens": 8989035.0,
+      "eval_runtime": 88.0233,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 3880
+    },
+    {
+      "entropy": 0.08905019680969417,
+      "epoch": 9.3983101991551,
+      "grad_norm": 0.21842406690120697,
+      "learning_rate": 4.177449088611095e-06,
+      "loss": 0.06666865348815917,
+      "mean_token_accuracy": 0.9753445640206337,
+      "num_tokens": 9033383.0,
+      "step": 3900
+    },
+    {
+      "epoch": 9.3983101991551,
+      "eval_entropy": 0.2295504841911659,
+      "eval_loss": 1.0662461519241333,
+      "eval_mean_token_accuracy": 0.8284122297603093,
+      "eval_num_tokens": 9033383.0,
+      "eval_runtime": 88.5525,
+      "eval_samples_per_second": 16.036,
+      "eval_steps_per_second": 2.01,
+      "step": 3900
+    },
+    {
+      "entropy": 0.08582521872594953,
+      "epoch": 9.446590223295111,
+      "grad_norm": 0.2000240683555603,
+      "learning_rate": 3.540256122150088e-06,
+      "loss": 0.06327944993972778,
+      "mean_token_accuracy": 0.9765432856976985,
+      "num_tokens": 9079672.0,
+      "step": 3920
+    },
+    {
+      "epoch": 9.446590223295111,
+      "eval_entropy": 0.2297678625818049,
+      "eval_loss": 1.0676169395446777,
+      "eval_mean_token_accuracy": 0.8283703879693921,
+      "eval_num_tokens": 9079672.0,
+      "eval_runtime": 87.6897,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 3920
+    },
+    {
+      "entropy": 0.09261773955076932,
+      "epoch": 9.494870247435124,
+      "grad_norm": 0.19104281067848206,
+      "learning_rate": 2.955303652656991e-06,
+      "loss": 0.06111966967582703,
+      "mean_token_accuracy": 0.9738784030079841,
+      "num_tokens": 9127647.0,
+      "step": 3940
+    },
+    {
+      "epoch": 9.494870247435124,
+      "eval_entropy": 0.22986975881490815,
+      "eval_loss": 1.0687925815582275,
+      "eval_mean_token_accuracy": 0.8282996858103892,
+      "eval_num_tokens": 9127647.0,
+      "eval_runtime": 88.2224,
+      "eval_samples_per_second": 16.096,
+      "eval_steps_per_second": 2.018,
+      "step": 3940
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.859804607209636e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3960/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3960/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3960/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3960/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3960/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3960/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3960/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3960/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3960/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3960/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3960/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3960/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3960/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3960/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..14c38f67f8dfc98059ebd429f4295856a62d195e
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3960/trainer_state.json
@@ -0,0 +1,4192 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 9.543150271575136,
+  "eval_steps": 20,
+  "global_step": 3960,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    },
+    {
+      "entropy": 0.1326121488586068,
+      "epoch": 7.181050090525045,
+      "grad_norm": 0.4843117296695709,
+      "learning_rate": 8.411870386507724e-05,
+      "loss": 0.09537842869758606,
+      "mean_token_accuracy": 0.9645151920616627,
+      "num_tokens": 6898444.0,
+      "step": 2980
+    },
+    {
+      "epoch": 7.181050090525045,
+      "eval_entropy": 0.27999477783280813,
+      "eval_loss": 0.9031037092208862,
+      "eval_mean_token_accuracy": 0.8261882204018282,
+      "eval_num_tokens": 6898444.0,
+      "eval_runtime": 88.5602,
+      "eval_samples_per_second": 16.034,
+      "eval_steps_per_second": 2.01,
+      "step": 2980
+    },
+    {
+      "entropy": 0.1156751093454659,
+      "epoch": 7.229330114665057,
+      "grad_norm": 0.5297922492027283,
+      "learning_rate": 8.149629351430326e-05,
+      "loss": 0.09203402996063233,
+      "mean_token_accuracy": 0.9689733698964119,
+      "num_tokens": 6947106.0,
+      "step": 3000
+    },
+    {
+      "epoch": 7.229330114665057,
+      "eval_entropy": 0.2740225845508361,
+      "eval_loss": 0.8876440525054932,
+      "eval_mean_token_accuracy": 0.8265116857678703,
+      "eval_num_tokens": 6947106.0,
+      "eval_runtime": 87.7971,
+      "eval_samples_per_second": 16.174,
+      "eval_steps_per_second": 2.027,
+      "step": 3000
+    },
+    {
+      "entropy": 0.12116114404052496,
+      "epoch": 7.27761013880507,
+      "grad_norm": 0.5467398166656494,
+      "learning_rate": 7.890406302404618e-05,
+      "loss": 0.0933056652545929,
+      "mean_token_accuracy": 0.9676455594599247,
+      "num_tokens": 6992784.0,
+      "step": 3020
+    },
+    {
+      "epoch": 7.27761013880507,
+      "eval_entropy": 0.27238342793804876,
+      "eval_loss": 0.9166117310523987,
+      "eval_mean_token_accuracy": 0.825499015242866,
+      "eval_num_tokens": 6992784.0,
+      "eval_runtime": 88.3276,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 3020
+    },
+    {
+      "entropy": 0.11715119928121567,
+      "epoch": 7.325890162945082,
+      "grad_norm": 0.4976702332496643,
+      "learning_rate": 7.634274596487176e-05,
+      "loss": 0.09449659585952759,
+      "mean_token_accuracy": 0.966506478190422,
+      "num_tokens": 7040482.0,
+      "step": 3040
+    },
+    {
+      "epoch": 7.325890162945082,
+      "eval_entropy": 0.2788975556077582,
+      "eval_loss": 0.8833065629005432,
+      "eval_mean_token_accuracy": 0.8269920596915684,
+      "eval_num_tokens": 7040482.0,
+      "eval_runtime": 88.9016,
+      "eval_samples_per_second": 15.973,
+      "eval_steps_per_second": 2.002,
+      "step": 3040
+    },
+    {
+      "entropy": 0.12565911412239075,
+      "epoch": 7.3741701870850935,
+      "grad_norm": 0.5728679895401001,
+      "learning_rate": 7.381306715921025e-05,
+      "loss": 0.09626244902610778,
+      "mean_token_accuracy": 0.9661444149911403,
+      "num_tokens": 7086261.0,
+      "step": 3060
+    },
+    {
+      "epoch": 7.3741701870850935,
+      "eval_entropy": 0.2762492011939542,
+      "eval_loss": 0.8961160778999329,
+      "eval_mean_token_accuracy": 0.826369910092836,
+      "eval_num_tokens": 7086261.0,
+      "eval_runtime": 88.5204,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 3060
+    },
+    {
+      "entropy": 0.12258567921817302,
+      "epoch": 7.422450211225105,
+      "grad_norm": 0.468985915184021,
+      "learning_rate": 7.131574247624031e-05,
+      "loss": 0.09120266437530518,
+      "mean_token_accuracy": 0.9674938447773457,
+      "num_tokens": 7133247.0,
+      "step": 3080
+    },
+    {
+      "epoch": 7.422450211225105,
+      "eval_entropy": 0.2688512086366,
+      "eval_loss": 0.9048654437065125,
+      "eval_mean_token_accuracy": 0.8265582302982888,
+      "eval_num_tokens": 7133247.0,
+      "eval_runtime": 88.164,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 3080
+    },
+    {
+      "entropy": 0.12070536343380808,
+      "epoch": 7.470730235365117,
+      "grad_norm": 0.5310841798782349,
+      "learning_rate": 6.885147862930688e-05,
+      "loss": 0.09386247992515565,
+      "mean_token_accuracy": 0.96664222702384,
+      "num_tokens": 7177810.0,
+      "step": 3100
+    },
+    {
+      "epoch": 7.470730235365117,
+      "eval_entropy": 0.26289632874593305,
+      "eval_loss": 0.902837336063385,
+      "eval_mean_token_accuracy": 0.828052184554968,
+      "eval_num_tokens": 7177810.0,
+      "eval_runtime": 88.1201,
+      "eval_samples_per_second": 16.114,
+      "eval_steps_per_second": 2.02,
+      "step": 3100
+    },
+    {
+      "entropy": 0.11322665181942285,
+      "epoch": 7.51901025950513,
+      "grad_norm": 0.6846067905426025,
+      "learning_rate": 6.642097297592908e-05,
+      "loss": 0.08772538304328918,
+      "mean_token_accuracy": 0.9691362135112286,
+      "num_tokens": 7225302.0,
+      "step": 3120
+    },
+    {
+      "epoch": 7.51901025950513,
+      "eval_entropy": 0.2670605613776807,
+      "eval_loss": 0.9130566120147705,
+      "eval_mean_token_accuracy": 0.826366903741708,
+      "eval_num_tokens": 7225302.0,
+      "eval_runtime": 88.6035,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3120
+    },
+    {
+      "entropy": 0.10985926580615342,
+      "epoch": 7.567290283645142,
+      "grad_norm": 0.4799748361110687,
+      "learning_rate": 6.402491332045683e-05,
+      "loss": 0.08672296404838561,
+      "mean_token_accuracy": 0.9698824048042297,
+      "num_tokens": 7274881.0,
+      "step": 3140
+    },
+    {
+      "epoch": 7.567290283645142,
+      "eval_entropy": 0.26657010866015146,
+      "eval_loss": 0.9056380987167358,
+      "eval_mean_token_accuracy": 0.8272326461384806,
+      "eval_num_tokens": 7274881.0,
+      "eval_runtime": 88.2967,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3140
+    },
+    {
+      "entropy": 0.11890473761595785,
+      "epoch": 7.615570307785154,
+      "grad_norm": 0.4081912934780121,
+      "learning_rate": 6.166397771942987e-05,
+      "loss": 0.09396827220916748,
+      "mean_token_accuracy": 0.967512384802103,
+      "num_tokens": 7321163.0,
+      "step": 3160
+    },
+    {
+      "epoch": 7.615570307785154,
+      "eval_entropy": 0.26364776575833226,
+      "eval_loss": 0.9204733967781067,
+      "eval_mean_token_accuracy": 0.827419729714983,
+      "eval_num_tokens": 7321163.0,
+      "eval_runtime": 88.3441,
+      "eval_samples_per_second": 16.074,
+      "eval_steps_per_second": 2.015,
+      "step": 3160
+    },
+    {
+      "entropy": 0.12209134344011545,
+      "epoch": 7.663850331925166,
+      "grad_norm": 0.5159774422645569,
+      "learning_rate": 5.93388342896962e-05,
+      "loss": 0.09292978644371033,
+      "mean_token_accuracy": 0.9658421002328396,
+      "num_tokens": 7366217.0,
+      "step": 3180
+    },
+    {
+      "epoch": 7.663850331925166,
+      "eval_entropy": 0.2714110068725736,
+      "eval_loss": 0.9143092036247253,
+      "eval_mean_token_accuracy": 0.8262251099843657,
+      "eval_num_tokens": 7366217.0,
+      "eval_runtime": 87.9635,
+      "eval_samples_per_second": 16.143,
+      "eval_steps_per_second": 2.024,
+      "step": 3180
+    },
+    {
+      "entropy": 0.11955609126016498,
+      "epoch": 7.712130356065178,
+      "grad_norm": 0.7005976438522339,
+      "learning_rate": 5.705014101934252e-05,
+      "loss": 0.09319776892662049,
+      "mean_token_accuracy": 0.9674513377249241,
+      "num_tokens": 7410887.0,
+      "step": 3200
+    },
+    {
+      "epoch": 7.712130356065178,
+      "eval_entropy": 0.2626291513442993,
+      "eval_loss": 0.91884446144104,
+      "eval_mean_token_accuracy": 0.827563955207889,
+      "eval_num_tokens": 7410887.0,
+      "eval_runtime": 87.8849,
+      "eval_samples_per_second": 16.157,
+      "eval_steps_per_second": 2.025,
+      "step": 3200
+    },
+    {
+      "entropy": 0.11767532899975777,
+      "epoch": 7.76041038020519,
+      "grad_norm": 0.4675362706184387,
+      "learning_rate": 5.479854558149187e-05,
+      "loss": 0.09038087725639343,
+      "mean_token_accuracy": 0.9669680051505566,
+      "num_tokens": 7457578.0,
+      "step": 3220
+    },
+    {
+      "epoch": 7.76041038020519,
+      "eval_entropy": 0.26581037203582486,
+      "eval_loss": 0.9146883487701416,
+      "eval_mean_token_accuracy": 0.8270345812433222,
+      "eval_num_tokens": 7457578.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 3220
+    },
+    {
+      "entropy": 0.11383061073720455,
+      "epoch": 7.808690404345202,
+      "grad_norm": 0.541247546672821,
+      "learning_rate": 5.258468515101918e-05,
+      "loss": 0.08942405581474304,
+      "mean_token_accuracy": 0.968398705124855,
+      "num_tokens": 7505761.0,
+      "step": 3240
+    },
+    {
+      "epoch": 7.808690404345202,
+      "eval_entropy": 0.26687998960861997,
+      "eval_loss": 0.9037085175514221,
+      "eval_mean_token_accuracy": 0.8274977863504646,
+      "eval_num_tokens": 7505761.0,
+      "eval_runtime": 88.5163,
+      "eval_samples_per_second": 16.042,
+      "eval_steps_per_second": 2.011,
+      "step": 3240
+    },
+    {
+      "entropy": 0.12004058873280883,
+      "epoch": 7.856970428485214,
+      "grad_norm": 0.5289552211761475,
+      "learning_rate": 5.040918622423899e-05,
+      "loss": 0.09064805507659912,
+      "mean_token_accuracy": 0.9673330299556255,
+      "num_tokens": 7552557.0,
+      "step": 3260
+    },
+    {
+      "epoch": 7.856970428485214,
+      "eval_entropy": 0.26288379251622085,
+      "eval_loss": 0.9146178364753723,
+      "eval_mean_token_accuracy": 0.8276662448149049,
+      "eval_num_tokens": 7552557.0,
+      "eval_runtime": 88.0723,
+      "eval_samples_per_second": 16.123,
+      "eval_steps_per_second": 2.021,
+      "step": 3260
+    },
+    {
+      "entropy": 0.1180689456872642,
+      "epoch": 7.905250452625227,
+      "grad_norm": 0.4754861891269684,
+      "learning_rate": 4.827266444161377e-05,
+      "loss": 0.09402655363082886,
+      "mean_token_accuracy": 0.9663361787796021,
+      "num_tokens": 7596998.0,
+      "step": 3280
+    },
+    {
+      "epoch": 7.905250452625227,
+      "eval_entropy": 0.261690996922134,
+      "eval_loss": 0.9120186567306519,
+      "eval_mean_token_accuracy": 0.8278995015647974,
+      "eval_num_tokens": 7596998.0,
+      "eval_runtime": 88.6573,
+      "eval_samples_per_second": 16.017,
+      "eval_steps_per_second": 2.008,
+      "step": 3280
+    },
+    {
+      "entropy": 0.1162164019420743,
+      "epoch": 7.953530476765239,
+      "grad_norm": 0.6203398108482361,
+      "learning_rate": 4.61757244135355e-05,
+      "loss": 0.08832768201828003,
+      "mean_token_accuracy": 0.9694220684468746,
+      "num_tokens": 7643288.0,
+      "step": 3300
+    },
+    {
+      "epoch": 7.953530476765239,
+      "eval_entropy": 0.26158678129817664,
+      "eval_loss": 0.9301361441612244,
+      "eval_mean_token_accuracy": 0.8269603885291668,
+      "eval_num_tokens": 7643288.0,
+      "eval_runtime": 88.7064,
+      "eval_samples_per_second": 16.008,
+      "eval_steps_per_second": 2.007,
+      "step": 3300
+    },
+    {
+      "entropy": 0.10877202204489089,
+      "epoch": 8.0,
+      "grad_norm": 3.058673143386841,
+      "learning_rate": 4.411895954922763e-05,
+      "loss": 0.09159615635871887,
+      "mean_token_accuracy": 0.9686878834451947,
+      "num_tokens": 7688608.0,
+      "step": 3320
+    },
+    {
+      "epoch": 8.0,
+      "eval_entropy": 0.2590438057197614,
+      "eval_loss": 0.9262065887451172,
+      "eval_mean_token_accuracy": 0.8277404546067956,
+      "eval_num_tokens": 7688608.0,
+      "eval_runtime": 87.7557,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 3320
+    },
+    {
+      "entropy": 0.09969092034734786,
+      "epoch": 8.048280024140013,
+      "grad_norm": 0.2561911642551422,
+      "learning_rate": 4.2102951888817644e-05,
+      "loss": 0.07228735685348511,
+      "mean_token_accuracy": 0.9743556551635265,
+      "num_tokens": 7732856.0,
+      "step": 3340
+    },
+    {
+      "epoch": 8.048280024140013,
+      "eval_entropy": 0.2462534454933713,
+      "eval_loss": 0.9755983352661133,
+      "eval_mean_token_accuracy": 0.827510209565752,
+      "eval_num_tokens": 7732856.0,
+      "eval_runtime": 88.0297,
+      "eval_samples_per_second": 16.131,
+      "eval_steps_per_second": 2.022,
+      "step": 3340
+    },
+    {
+      "entropy": 0.09477624730207027,
+      "epoch": 8.096560048280024,
+      "grad_norm": 0.3670356571674347,
+      "learning_rate": 4.012827193862633e-05,
+      "loss": 0.0686036467552185,
+      "mean_token_accuracy": 0.9743670664727688,
+      "num_tokens": 7778585.0,
+      "step": 3360
+    },
+    {
+      "epoch": 8.096560048280024,
+      "eval_entropy": 0.2440882951356052,
+      "eval_loss": 1.0118423700332642,
+      "eval_mean_token_accuracy": 0.8271241070849172,
+      "eval_num_tokens": 7778585.0,
+      "eval_runtime": 88.1364,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 3360
+    },
+    {
+      "entropy": 0.09417980071157217,
+      "epoch": 8.144840072420036,
+      "grad_norm": 0.5785111784934998,
+      "learning_rate": 3.819547850972133e-05,
+      "loss": 0.07100698947906495,
+      "mean_token_accuracy": 0.9737804099917412,
+      "num_tokens": 7824287.0,
+      "step": 3380
+    },
+    {
+      "epoch": 8.144840072420036,
+      "eval_entropy": 0.24603172860453637,
+      "eval_loss": 1.0021991729736328,
+      "eval_mean_token_accuracy": 0.827162122793412,
+      "eval_num_tokens": 7824287.0,
+      "eval_runtime": 88.4534,
+      "eval_samples_per_second": 16.054,
+      "eval_steps_per_second": 2.012,
+      "step": 3380
+    },
+    {
+      "entropy": 0.087222247896716,
+      "epoch": 8.193120096560047,
+      "grad_norm": 0.44848504662513733,
+      "learning_rate": 3.6305118559780393e-05,
+      "loss": 0.06325234174728393,
+      "mean_token_accuracy": 0.9769335232675076,
+      "num_tokens": 7874784.0,
+      "step": 3400
+    },
+    {
+      "epoch": 8.193120096560047,
+      "eval_entropy": 0.24544863905129807,
+      "eval_loss": 1.005008339881897,
+      "eval_mean_token_accuracy": 0.8273035240976998,
+      "eval_num_tokens": 7874784.0,
+      "eval_runtime": 88.6737,
+      "eval_samples_per_second": 16.014,
+      "eval_steps_per_second": 2.007,
+      "step": 3400
+    },
+    {
+      "entropy": 0.09037275062873959,
+      "epoch": 8.24140012070006,
+      "grad_norm": 0.23529155552387238,
+      "learning_rate": 3.4457727038308435e-05,
+      "loss": 0.06589569449424744,
+      "mean_token_accuracy": 0.9751470290124417,
+      "num_tokens": 7925240.0,
+      "step": 3420
+    },
+    {
+      "epoch": 8.24140012070006,
+      "eval_entropy": 0.24203423494368456,
+      "eval_loss": 1.0052824020385742,
+      "eval_mean_token_accuracy": 0.8275575503874361,
+      "eval_num_tokens": 7925240.0,
+      "eval_runtime": 88.3486,
+      "eval_samples_per_second": 16.073,
+      "eval_steps_per_second": 2.015,
+      "step": 3420
+    },
+    {
+      "entropy": 0.10373208015225828,
+      "epoch": 8.289680144840073,
+      "grad_norm": 0.24796408414840698,
+      "learning_rate": 3.2653826735253594e-05,
+      "loss": 0.07714771628379821,
+      "mean_token_accuracy": 0.9722891598939896,
+      "num_tokens": 7966685.0,
+      "step": 3440
+    },
+    {
+      "epoch": 8.289680144840073,
+      "eval_entropy": 0.23461157050025597,
+      "eval_loss": 1.012108325958252,
+      "eval_mean_token_accuracy": 0.82825900564033,
+      "eval_num_tokens": 7966685.0,
+      "eval_runtime": 88.1018,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 3440
+    },
+    {
+      "entropy": 0.08967954274266958,
+      "epoch": 8.337960168980084,
+      "grad_norm": 0.5314674973487854,
+      "learning_rate": 3.08939281330635e-05,
+      "loss": 0.06746266484260559,
+      "mean_token_accuracy": 0.9753133699297905,
+      "num_tokens": 8013656.0,
+      "step": 3460
+    },
+    {
+      "epoch": 8.337960168980084,
+      "eval_entropy": 0.23781998589467468,
+      "eval_loss": 1.0119913816452026,
+      "eval_mean_token_accuracy": 0.8273287157664139,
+      "eval_num_tokens": 8013656.0,
+      "eval_runtime": 88.6038,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3460
+    },
+    {
+      "entropy": 0.09280927330255509,
+      "epoch": 8.386240193120097,
+      "grad_norm": 0.30658936500549316,
+      "learning_rate": 2.9178529262224946e-05,
+      "loss": 0.06936421990394592,
+      "mean_token_accuracy": 0.9742105580866337,
+      "num_tokens": 8061108.0,
+      "step": 3480
+    },
+    {
+      "epoch": 8.386240193120097,
+      "eval_entropy": 0.23672426734747512,
+      "eval_loss": 1.0162041187286377,
+      "eval_mean_token_accuracy": 0.8278394281194451,
+      "eval_num_tokens": 8061108.0,
+      "eval_runtime": 88.6015,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 3480
+    },
+    {
+      "entropy": 0.09195122267119586,
+      "epoch": 8.43452021726011,
+      "grad_norm": 0.24512454867362976,
+      "learning_rate": 2.7508115560327256e-05,
+      "loss": 0.06996064782142639,
+      "mean_token_accuracy": 0.974104854464531,
+      "num_tokens": 8106924.0,
+      "step": 3500
+    },
+    {
+      "epoch": 8.43452021726011,
+      "eval_entropy": 0.23758245828781235,
+      "eval_loss": 1.0157747268676758,
+      "eval_mean_token_accuracy": 0.8278501579600773,
+      "eval_num_tokens": 8106924.0,
+      "eval_runtime": 88.5678,
+      "eval_samples_per_second": 16.033,
+      "eval_steps_per_second": 2.01,
+      "step": 3500
+    },
+    {
+      "entropy": 0.09518477590754629,
+      "epoch": 8.48280024140012,
+      "grad_norm": 0.2373228371143341,
+      "learning_rate": 2.5883159734688868e-05,
+      "loss": 0.06832797527313232,
+      "mean_token_accuracy": 0.9743806682527065,
+      "num_tokens": 8153070.0,
+      "step": 3520
+    },
+    {
+      "epoch": 8.48280024140012,
+      "eval_entropy": 0.2379308074210467,
+      "eval_loss": 1.0180530548095703,
+      "eval_mean_token_accuracy": 0.8282477052024241,
+      "eval_num_tokens": 8153070.0,
+      "eval_runtime": 88.4651,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 3520
+    },
+    {
+      "entropy": 0.0906269280705601,
+      "epoch": 8.531080265540133,
+      "grad_norm": 0.37602657079696655,
+      "learning_rate": 2.4304121628587173e-05,
+      "loss": 0.06921588778495788,
+      "mean_token_accuracy": 0.9745571479201317,
+      "num_tokens": 8200066.0,
+      "step": 3540
+    },
+    {
+      "epoch": 8.531080265540133,
+      "eval_entropy": 0.2357490769263064,
+      "eval_loss": 1.017139196395874,
+      "eval_mean_token_accuracy": 0.828713178299786,
+      "eval_num_tokens": 8200066.0,
+      "eval_runtime": 88.6424,
+      "eval_samples_per_second": 16.019,
+      "eval_steps_per_second": 2.008,
+      "step": 3540
+    },
+    {
+      "entropy": 0.09273334196768701,
+      "epoch": 8.579360289680144,
+      "grad_norm": 0.2978838086128235,
+      "learning_rate": 2.277144809112794e-05,
+      "loss": 0.06970887780189514,
+      "mean_token_accuracy": 0.9751374177634716,
+      "num_tokens": 8246157.0,
+      "step": 3560
+    },
+    {
+      "epoch": 8.579360289680144,
+      "eval_entropy": 0.23590933138065123,
+      "eval_loss": 1.019231915473938,
+      "eval_mean_token_accuracy": 0.8279163412163767,
+      "eval_num_tokens": 8246157.0,
+      "eval_runtime": 88.2691,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 3560
+    },
+    {
+      "entropy": 0.09854201921261847,
+      "epoch": 8.627640313820157,
+      "grad_norm": 0.5381243228912354,
+      "learning_rate": 2.1285572850792413e-05,
+      "loss": 0.07123501300811767,
+      "mean_token_accuracy": 0.9723988175392151,
+      "num_tokens": 8291752.0,
+      "step": 3580
+    },
+    {
+      "epoch": 8.627640313820157,
+      "eval_entropy": 0.23592024949494372,
+      "eval_loss": 1.020250678062439,
+      "eval_mean_token_accuracy": 0.8282466226079491,
+      "eval_num_tokens": 8291752.0,
+      "eval_runtime": 87.8939,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 3580
+    },
+    {
+      "entropy": 0.09092000788077712,
+      "epoch": 8.67592033796017,
+      "grad_norm": 0.29089561104774475,
+      "learning_rate": 1.9846916392697493e-05,
+      "loss": 0.06982720494270325,
+      "mean_token_accuracy": 0.9749995484948158,
+      "num_tokens": 8337651.0,
+      "step": 3600
+    },
+    {
+      "epoch": 8.67592033796017,
+      "eval_entropy": 0.23658913382318583,
+      "eval_loss": 1.019862413406372,
+      "eval_mean_token_accuracy": 0.8280127842104836,
+      "eval_num_tokens": 8337651.0,
+      "eval_runtime": 88.2855,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 3600
+    },
+    {
+      "entropy": 0.08708424470387399,
+      "epoch": 8.72420036210018,
+      "grad_norm": 0.29904839396476746,
+      "learning_rate": 1.8455885839603106e-05,
+      "loss": 0.06801370978355407,
+      "mean_token_accuracy": 0.9755801670253277,
+      "num_tokens": 8385330.0,
+      "step": 3620
+    },
+    {
+      "epoch": 8.72420036210018,
+      "eval_entropy": 0.23627035771862845,
+      "eval_loss": 1.0204569101333618,
+      "eval_mean_token_accuracy": 0.8279967894045155,
+      "eval_num_tokens": 8385330.0,
+      "eval_runtime": 88.1753,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 3620
+    },
+    {
+      "entropy": 0.0916255566291511,
+      "epoch": 8.772480386240193,
+      "grad_norm": 0.20571033656597137,
+      "learning_rate": 1.7112874836701502e-05,
+      "loss": 0.0692249894142151,
+      "mean_token_accuracy": 0.9745148062705994,
+      "num_tokens": 8432134.0,
+      "step": 3640
+    },
+    {
+      "epoch": 8.772480386240193,
+      "eval_entropy": 0.2340821772813797,
+      "eval_loss": 1.0212821960449219,
+      "eval_mean_token_accuracy": 0.828373622358515,
+      "eval_num_tokens": 8432134.0,
+      "eval_runtime": 87.2518,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 3640
+    },
+    {
+      "entropy": 0.0926868049427867,
+      "epoch": 8.820760410380204,
+      "grad_norm": 0.2618265450000763,
+      "learning_rate": 1.5818263440220203e-05,
+      "loss": 0.07338647246360779,
+      "mean_token_accuracy": 0.9724708944559097,
+      "num_tokens": 8476758.0,
+      "step": 3660
+    },
+    {
+      "epoch": 8.820760410380204,
+      "eval_entropy": 0.23637074575330433,
+      "eval_loss": 1.0186725854873657,
+      "eval_mean_token_accuracy": 0.8277707605549459,
+      "eval_num_tokens": 8476758.0,
+      "eval_runtime": 88.4695,
+      "eval_samples_per_second": 16.051,
+      "eval_steps_per_second": 2.012,
+      "step": 3660
+    },
+    {
+      "entropy": 0.09175782897509635,
+      "epoch": 8.869040434520217,
+      "grad_norm": 0.2889852821826935,
+      "learning_rate": 1.4572418009870604e-05,
+      "loss": 0.07066041231155396,
+      "mean_token_accuracy": 0.9725673109292984,
+      "num_tokens": 8523294.0,
+      "step": 3680
+    },
+    {
+      "epoch": 8.869040434520217,
+      "eval_entropy": 0.23945164404223473,
+      "eval_loss": 1.0145231485366821,
+      "eval_mean_token_accuracy": 0.8281202523895864,
+      "eval_num_tokens": 8523294.0,
+      "eval_runtime": 88.368,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 3680
+    },
+    {
+      "entropy": 0.09265611860901117,
+      "epoch": 8.91732045866023,
+      "grad_norm": 0.41092950105667114,
+      "learning_rate": 1.3375691105172625e-05,
+      "loss": 0.06975027322769164,
+      "mean_token_accuracy": 0.974636297672987,
+      "num_tokens": 8570358.0,
+      "step": 3700
+    },
+    {
+      "epoch": 8.91732045866023,
+      "eval_entropy": 0.239588919901446,
+      "eval_loss": 1.014059066772461,
+      "eval_mean_token_accuracy": 0.828028041660116,
+      "eval_num_tokens": 8570358.0,
+      "eval_runtime": 88.1039,
+      "eval_samples_per_second": 16.117,
+      "eval_steps_per_second": 2.02,
+      "step": 3700
+    },
+    {
+      "entropy": 0.0896465003490448,
+      "epoch": 8.965600482800241,
+      "grad_norm": 0.24000254273414612,
+      "learning_rate": 1.2228421385684305e-05,
+      "loss": 0.06743953824043274,
+      "mean_token_accuracy": 0.974006575345993,
+      "num_tokens": 8618246.0,
+      "step": 3720
+    },
+    {
+      "epoch": 8.965600482800241,
+      "eval_entropy": 0.2390461369846644,
+      "eval_loss": 1.011997938156128,
+      "eval_mean_token_accuracy": 0.8284533211354459,
+      "eval_num_tokens": 8618246.0,
+      "eval_runtime": 88.6202,
+      "eval_samples_per_second": 16.023,
+      "eval_steps_per_second": 2.009,
+      "step": 3720
+    },
+    {
+      "entropy": 0.09575109251520851,
+      "epoch": 9.012070006035003,
+      "grad_norm": 0.18088576197624207,
+      "learning_rate": 1.1130933515165429e-05,
+      "loss": 0.07295305132865906,
+      "mean_token_accuracy": 0.9739722857227573,
+      "num_tokens": 8661177.0,
+      "step": 3740
+    },
+    {
+      "epoch": 9.012070006035003,
+      "eval_entropy": 0.23728882387447892,
+      "eval_loss": 1.0134038925170898,
+      "eval_mean_token_accuracy": 0.8286508951294288,
+      "eval_num_tokens": 8661177.0,
+      "eval_runtime": 88.6434,
+      "eval_samples_per_second": 16.019,
+      "eval_steps_per_second": 2.008,
+      "step": 3740
+    },
+    {
+      "entropy": 0.08883201805874705,
+      "epoch": 9.060350030175016,
+      "grad_norm": 0.23120898008346558,
+      "learning_rate": 1.008353806970132e-05,
+      "loss": 0.06326142549514771,
+      "mean_token_accuracy": 0.9780020996928215,
+      "num_tokens": 8707602.0,
+      "step": 3760
+    },
+    {
+      "epoch": 9.060350030175016,
+      "eval_entropy": 0.23416905331142832,
+      "eval_loss": 1.0284100770950317,
+      "eval_mean_token_accuracy": 0.8284083213029283,
+      "eval_num_tokens": 8707602.0,
+      "eval_runtime": 88.2933,
+      "eval_samples_per_second": 16.083,
+      "eval_steps_per_second": 2.016,
+      "step": 3760
+    },
+    {
+      "entropy": 0.08861468276008963,
+      "epoch": 9.108630054315027,
+      "grad_norm": 0.17277656495571136,
+      "learning_rate": 9.086531449813489e-06,
+      "loss": 0.06727416515350342,
+      "mean_token_accuracy": 0.9760390736162663,
+      "num_tokens": 8751287.0,
+      "step": 3780
+    },
+    {
+      "epoch": 9.108630054315027,
+      "eval_entropy": 0.23370317386442355,
+      "eval_loss": 1.0381091833114624,
+      "eval_mean_token_accuracy": 0.8282668178670862,
+      "eval_num_tokens": 8751287.0,
+      "eval_runtime": 88.5738,
+      "eval_samples_per_second": 16.032,
+      "eval_steps_per_second": 2.01,
+      "step": 3780
+    },
+    {
+      "entropy": 0.09032954541034996,
+      "epoch": 9.15691007845504,
+      "grad_norm": 0.19481085240840912,
+      "learning_rate": 8.140195796581981e-06,
+      "loss": 0.06117786169052124,
+      "mean_token_accuracy": 0.9764454193413258,
+      "num_tokens": 8799982.0,
+      "step": 3800
+    },
+    {
+      "epoch": 9.15691007845504,
+      "eval_entropy": 0.23215936609868254,
+      "eval_loss": 1.0467767715454102,
+      "eval_mean_token_accuracy": 0.828405868806196,
+      "eval_num_tokens": 8799982.0,
+      "eval_runtime": 87.9551,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 3800
+    },
+    {
+      "entropy": 0.08824449330568314,
+      "epoch": 9.20519010259505,
+      "grad_norm": 0.20382793247699738,
+      "learning_rate": 7.2447989118025455e-06,
+      "loss": 0.06422553658485412,
+      "mean_token_accuracy": 0.9763121999800205,
+      "num_tokens": 8845536.0,
+      "step": 3820
+    },
+    {
+      "epoch": 9.20519010259505,
+      "eval_entropy": 0.23169661472352704,
+      "eval_loss": 1.0529043674468994,
+      "eval_mean_token_accuracy": 0.8282025610463003,
+      "eval_num_tokens": 8845536.0,
+      "eval_runtime": 88.5548,
+      "eval_samples_per_second": 16.035,
+      "eval_steps_per_second": 2.01,
+      "step": 3820
+    },
+    {
+      "entropy": 0.07768499720841646,
+      "epoch": 9.253470126735063,
+      "grad_norm": 0.11727919429540634,
+      "learning_rate": 6.400594182202084e-06,
+      "loss": 0.057279336452484134,
+      "mean_token_accuracy": 0.9795166745781898,
+      "num_tokens": 8896053.0,
+      "step": 3840
+    },
+    {
+      "epoch": 9.253470126735063,
+      "eval_entropy": 0.2301936373114586,
+      "eval_loss": 1.0576049089431763,
+      "eval_mean_token_accuracy": 0.8283417489421502,
+      "eval_num_tokens": 8896053.0,
+      "eval_runtime": 88.7854,
+      "eval_samples_per_second": 15.994,
+      "eval_steps_per_second": 2.005,
+      "step": 3840
+    },
+    {
+      "entropy": 0.09015200780704617,
+      "epoch": 9.301750150875076,
+      "grad_norm": 0.2089058756828308,
+      "learning_rate": 5.6078205077330705e-06,
+      "loss": 0.06626194715499878,
+      "mean_token_accuracy": 0.9747506938874722,
+      "num_tokens": 8940591.0,
+      "step": 3860
+    },
+    {
+      "epoch": 9.301750150875076,
+      "eval_entropy": 0.23024900676159377,
+      "eval_loss": 1.0599786043167114,
+      "eval_mean_token_accuracy": 0.8283051915383071,
+      "eval_num_tokens": 8940591.0,
+      "eval_runtime": 88.2963,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3860
+    },
+    {
+      "entropy": 0.08484094133600592,
+      "epoch": 9.350030175015087,
+      "grad_norm": 0.21646778285503387,
+      "learning_rate": 4.866702233967785e-06,
+      "loss": 0.05998039245605469,
+      "mean_token_accuracy": 0.9771868668496608,
+      "num_tokens": 8989035.0,
+      "step": 3880
+    },
+    {
+      "epoch": 9.350030175015087,
+      "eval_entropy": 0.2300723305578982,
+      "eval_loss": 1.0633587837219238,
+      "eval_mean_token_accuracy": 0.828371500701047,
+      "eval_num_tokens": 8989035.0,
+      "eval_runtime": 88.0233,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 3880
+    },
+    {
+      "entropy": 0.08905019680969417,
+      "epoch": 9.3983101991551,
+      "grad_norm": 0.21842406690120697,
+      "learning_rate": 4.177449088611095e-06,
+      "loss": 0.06666865348815917,
+      "mean_token_accuracy": 0.9753445640206337,
+      "num_tokens": 9033383.0,
+      "step": 3900
+    },
+    {
+      "epoch": 9.3983101991551,
+      "eval_entropy": 0.2295504841911659,
+      "eval_loss": 1.0662461519241333,
+      "eval_mean_token_accuracy": 0.8284122297603093,
+      "eval_num_tokens": 9033383.0,
+      "eval_runtime": 88.5525,
+      "eval_samples_per_second": 16.036,
+      "eval_steps_per_second": 2.01,
+      "step": 3900
+    },
+    {
+      "entropy": 0.08582521872594953,
+      "epoch": 9.446590223295111,
+      "grad_norm": 0.2000240683555603,
+      "learning_rate": 3.540256122150088e-06,
+      "loss": 0.06327944993972778,
+      "mean_token_accuracy": 0.9765432856976985,
+      "num_tokens": 9079672.0,
+      "step": 3920
+    },
+    {
+      "epoch": 9.446590223295111,
+      "eval_entropy": 0.2297678625818049,
+      "eval_loss": 1.0676169395446777,
+      "eval_mean_token_accuracy": 0.8283703879693921,
+      "eval_num_tokens": 9079672.0,
+      "eval_runtime": 87.6897,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 3920
+    },
+    {
+      "entropy": 0.09261773955076932,
+      "epoch": 9.494870247435124,
+      "grad_norm": 0.19104281067848206,
+      "learning_rate": 2.955303652656991e-06,
+      "loss": 0.06111966967582703,
+      "mean_token_accuracy": 0.9738784030079841,
+      "num_tokens": 9127647.0,
+      "step": 3940
+    },
+    {
+      "epoch": 9.494870247435124,
+      "eval_entropy": 0.22986975881490815,
+      "eval_loss": 1.0687925815582275,
+      "eval_mean_token_accuracy": 0.8282996858103892,
+      "eval_num_tokens": 9127647.0,
+      "eval_runtime": 88.2224,
+      "eval_samples_per_second": 16.096,
+      "eval_steps_per_second": 2.018,
+      "step": 3940
+    },
+    {
+      "entropy": 0.08347254800610245,
+      "epoch": 9.543150271575136,
+      "grad_norm": 0.18671369552612305,
+      "learning_rate": 2.422757214761356e-06,
+      "loss": 0.06384263038635254,
+      "mean_token_accuracy": 0.9770128890872002,
+      "num_tokens": 9172587.0,
+      "step": 3960
+    },
+    {
+      "epoch": 9.543150271575136,
+      "eval_entropy": 0.22975517851248217,
+      "eval_loss": 1.0701013803482056,
+      "eval_mean_token_accuracy": 0.8282244285840666,
+      "eval_num_tokens": 9172587.0,
+      "eval_runtime": 88.5794,
+      "eval_samples_per_second": 16.031,
+      "eval_steps_per_second": 2.009,
+      "step": 3960
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.87856170682196e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3980/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3980/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3980/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3980/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3980/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3980/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3980/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3980/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3980/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3980/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3980/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3980/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3980/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3980/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..3391e518c009fb23e7fe7e06eecdd8f2635c2635
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-3980/trainer_state.json
@@ -0,0 +1,4213 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 9.591430295715147,
+  "eval_steps": 20,
+  "global_step": 3980,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    },
+    {
+      "entropy": 0.1326121488586068,
+      "epoch": 7.181050090525045,
+      "grad_norm": 0.4843117296695709,
+      "learning_rate": 8.411870386507724e-05,
+      "loss": 0.09537842869758606,
+      "mean_token_accuracy": 0.9645151920616627,
+      "num_tokens": 6898444.0,
+      "step": 2980
+    },
+    {
+      "epoch": 7.181050090525045,
+      "eval_entropy": 0.27999477783280813,
+      "eval_loss": 0.9031037092208862,
+      "eval_mean_token_accuracy": 0.8261882204018282,
+      "eval_num_tokens": 6898444.0,
+      "eval_runtime": 88.5602,
+      "eval_samples_per_second": 16.034,
+      "eval_steps_per_second": 2.01,
+      "step": 2980
+    },
+    {
+      "entropy": 0.1156751093454659,
+      "epoch": 7.229330114665057,
+      "grad_norm": 0.5297922492027283,
+      "learning_rate": 8.149629351430326e-05,
+      "loss": 0.09203402996063233,
+      "mean_token_accuracy": 0.9689733698964119,
+      "num_tokens": 6947106.0,
+      "step": 3000
+    },
+    {
+      "epoch": 7.229330114665057,
+      "eval_entropy": 0.2740225845508361,
+      "eval_loss": 0.8876440525054932,
+      "eval_mean_token_accuracy": 0.8265116857678703,
+      "eval_num_tokens": 6947106.0,
+      "eval_runtime": 87.7971,
+      "eval_samples_per_second": 16.174,
+      "eval_steps_per_second": 2.027,
+      "step": 3000
+    },
+    {
+      "entropy": 0.12116114404052496,
+      "epoch": 7.27761013880507,
+      "grad_norm": 0.5467398166656494,
+      "learning_rate": 7.890406302404618e-05,
+      "loss": 0.0933056652545929,
+      "mean_token_accuracy": 0.9676455594599247,
+      "num_tokens": 6992784.0,
+      "step": 3020
+    },
+    {
+      "epoch": 7.27761013880507,
+      "eval_entropy": 0.27238342793804876,
+      "eval_loss": 0.9166117310523987,
+      "eval_mean_token_accuracy": 0.825499015242866,
+      "eval_num_tokens": 6992784.0,
+      "eval_runtime": 88.3276,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 3020
+    },
+    {
+      "entropy": 0.11715119928121567,
+      "epoch": 7.325890162945082,
+      "grad_norm": 0.4976702332496643,
+      "learning_rate": 7.634274596487176e-05,
+      "loss": 0.09449659585952759,
+      "mean_token_accuracy": 0.966506478190422,
+      "num_tokens": 7040482.0,
+      "step": 3040
+    },
+    {
+      "epoch": 7.325890162945082,
+      "eval_entropy": 0.2788975556077582,
+      "eval_loss": 0.8833065629005432,
+      "eval_mean_token_accuracy": 0.8269920596915684,
+      "eval_num_tokens": 7040482.0,
+      "eval_runtime": 88.9016,
+      "eval_samples_per_second": 15.973,
+      "eval_steps_per_second": 2.002,
+      "step": 3040
+    },
+    {
+      "entropy": 0.12565911412239075,
+      "epoch": 7.3741701870850935,
+      "grad_norm": 0.5728679895401001,
+      "learning_rate": 7.381306715921025e-05,
+      "loss": 0.09626244902610778,
+      "mean_token_accuracy": 0.9661444149911403,
+      "num_tokens": 7086261.0,
+      "step": 3060
+    },
+    {
+      "epoch": 7.3741701870850935,
+      "eval_entropy": 0.2762492011939542,
+      "eval_loss": 0.8961160778999329,
+      "eval_mean_token_accuracy": 0.826369910092836,
+      "eval_num_tokens": 7086261.0,
+      "eval_runtime": 88.5204,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 3060
+    },
+    {
+      "entropy": 0.12258567921817302,
+      "epoch": 7.422450211225105,
+      "grad_norm": 0.468985915184021,
+      "learning_rate": 7.131574247624031e-05,
+      "loss": 0.09120266437530518,
+      "mean_token_accuracy": 0.9674938447773457,
+      "num_tokens": 7133247.0,
+      "step": 3080
+    },
+    {
+      "epoch": 7.422450211225105,
+      "eval_entropy": 0.2688512086366,
+      "eval_loss": 0.9048654437065125,
+      "eval_mean_token_accuracy": 0.8265582302982888,
+      "eval_num_tokens": 7133247.0,
+      "eval_runtime": 88.164,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 3080
+    },
+    {
+      "entropy": 0.12070536343380808,
+      "epoch": 7.470730235365117,
+      "grad_norm": 0.5310841798782349,
+      "learning_rate": 6.885147862930688e-05,
+      "loss": 0.09386247992515565,
+      "mean_token_accuracy": 0.96664222702384,
+      "num_tokens": 7177810.0,
+      "step": 3100
+    },
+    {
+      "epoch": 7.470730235365117,
+      "eval_entropy": 0.26289632874593305,
+      "eval_loss": 0.902837336063385,
+      "eval_mean_token_accuracy": 0.828052184554968,
+      "eval_num_tokens": 7177810.0,
+      "eval_runtime": 88.1201,
+      "eval_samples_per_second": 16.114,
+      "eval_steps_per_second": 2.02,
+      "step": 3100
+    },
+    {
+      "entropy": 0.11322665181942285,
+      "epoch": 7.51901025950513,
+      "grad_norm": 0.6846067905426025,
+      "learning_rate": 6.642097297592908e-05,
+      "loss": 0.08772538304328918,
+      "mean_token_accuracy": 0.9691362135112286,
+      "num_tokens": 7225302.0,
+      "step": 3120
+    },
+    {
+      "epoch": 7.51901025950513,
+      "eval_entropy": 0.2670605613776807,
+      "eval_loss": 0.9130566120147705,
+      "eval_mean_token_accuracy": 0.826366903741708,
+      "eval_num_tokens": 7225302.0,
+      "eval_runtime": 88.6035,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3120
+    },
+    {
+      "entropy": 0.10985926580615342,
+      "epoch": 7.567290283645142,
+      "grad_norm": 0.4799748361110687,
+      "learning_rate": 6.402491332045683e-05,
+      "loss": 0.08672296404838561,
+      "mean_token_accuracy": 0.9698824048042297,
+      "num_tokens": 7274881.0,
+      "step": 3140
+    },
+    {
+      "epoch": 7.567290283645142,
+      "eval_entropy": 0.26657010866015146,
+      "eval_loss": 0.9056380987167358,
+      "eval_mean_token_accuracy": 0.8272326461384806,
+      "eval_num_tokens": 7274881.0,
+      "eval_runtime": 88.2967,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3140
+    },
+    {
+      "entropy": 0.11890473761595785,
+      "epoch": 7.615570307785154,
+      "grad_norm": 0.4081912934780121,
+      "learning_rate": 6.166397771942987e-05,
+      "loss": 0.09396827220916748,
+      "mean_token_accuracy": 0.967512384802103,
+      "num_tokens": 7321163.0,
+      "step": 3160
+    },
+    {
+      "epoch": 7.615570307785154,
+      "eval_entropy": 0.26364776575833226,
+      "eval_loss": 0.9204733967781067,
+      "eval_mean_token_accuracy": 0.827419729714983,
+      "eval_num_tokens": 7321163.0,
+      "eval_runtime": 88.3441,
+      "eval_samples_per_second": 16.074,
+      "eval_steps_per_second": 2.015,
+      "step": 3160
+    },
+    {
+      "entropy": 0.12209134344011545,
+      "epoch": 7.663850331925166,
+      "grad_norm": 0.5159774422645569,
+      "learning_rate": 5.93388342896962e-05,
+      "loss": 0.09292978644371033,
+      "mean_token_accuracy": 0.9658421002328396,
+      "num_tokens": 7366217.0,
+      "step": 3180
+    },
+    {
+      "epoch": 7.663850331925166,
+      "eval_entropy": 0.2714110068725736,
+      "eval_loss": 0.9143092036247253,
+      "eval_mean_token_accuracy": 0.8262251099843657,
+      "eval_num_tokens": 7366217.0,
+      "eval_runtime": 87.9635,
+      "eval_samples_per_second": 16.143,
+      "eval_steps_per_second": 2.024,
+      "step": 3180
+    },
+    {
+      "entropy": 0.11955609126016498,
+      "epoch": 7.712130356065178,
+      "grad_norm": 0.7005976438522339,
+      "learning_rate": 5.705014101934252e-05,
+      "loss": 0.09319776892662049,
+      "mean_token_accuracy": 0.9674513377249241,
+      "num_tokens": 7410887.0,
+      "step": 3200
+    },
+    {
+      "epoch": 7.712130356065178,
+      "eval_entropy": 0.2626291513442993,
+      "eval_loss": 0.91884446144104,
+      "eval_mean_token_accuracy": 0.827563955207889,
+      "eval_num_tokens": 7410887.0,
+      "eval_runtime": 87.8849,
+      "eval_samples_per_second": 16.157,
+      "eval_steps_per_second": 2.025,
+      "step": 3200
+    },
+    {
+      "entropy": 0.11767532899975777,
+      "epoch": 7.76041038020519,
+      "grad_norm": 0.4675362706184387,
+      "learning_rate": 5.479854558149187e-05,
+      "loss": 0.09038087725639343,
+      "mean_token_accuracy": 0.9669680051505566,
+      "num_tokens": 7457578.0,
+      "step": 3220
+    },
+    {
+      "epoch": 7.76041038020519,
+      "eval_entropy": 0.26581037203582486,
+      "eval_loss": 0.9146883487701416,
+      "eval_mean_token_accuracy": 0.8270345812433222,
+      "eval_num_tokens": 7457578.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 3220
+    },
+    {
+      "entropy": 0.11383061073720455,
+      "epoch": 7.808690404345202,
+      "grad_norm": 0.541247546672821,
+      "learning_rate": 5.258468515101918e-05,
+      "loss": 0.08942405581474304,
+      "mean_token_accuracy": 0.968398705124855,
+      "num_tokens": 7505761.0,
+      "step": 3240
+    },
+    {
+      "epoch": 7.808690404345202,
+      "eval_entropy": 0.26687998960861997,
+      "eval_loss": 0.9037085175514221,
+      "eval_mean_token_accuracy": 0.8274977863504646,
+      "eval_num_tokens": 7505761.0,
+      "eval_runtime": 88.5163,
+      "eval_samples_per_second": 16.042,
+      "eval_steps_per_second": 2.011,
+      "step": 3240
+    },
+    {
+      "entropy": 0.12004058873280883,
+      "epoch": 7.856970428485214,
+      "grad_norm": 0.5289552211761475,
+      "learning_rate": 5.040918622423899e-05,
+      "loss": 0.09064805507659912,
+      "mean_token_accuracy": 0.9673330299556255,
+      "num_tokens": 7552557.0,
+      "step": 3260
+    },
+    {
+      "epoch": 7.856970428485214,
+      "eval_entropy": 0.26288379251622085,
+      "eval_loss": 0.9146178364753723,
+      "eval_mean_token_accuracy": 0.8276662448149049,
+      "eval_num_tokens": 7552557.0,
+      "eval_runtime": 88.0723,
+      "eval_samples_per_second": 16.123,
+      "eval_steps_per_second": 2.021,
+      "step": 3260
+    },
+    {
+      "entropy": 0.1180689456872642,
+      "epoch": 7.905250452625227,
+      "grad_norm": 0.4754861891269684,
+      "learning_rate": 4.827266444161377e-05,
+      "loss": 0.09402655363082886,
+      "mean_token_accuracy": 0.9663361787796021,
+      "num_tokens": 7596998.0,
+      "step": 3280
+    },
+    {
+      "epoch": 7.905250452625227,
+      "eval_entropy": 0.261690996922134,
+      "eval_loss": 0.9120186567306519,
+      "eval_mean_token_accuracy": 0.8278995015647974,
+      "eval_num_tokens": 7596998.0,
+      "eval_runtime": 88.6573,
+      "eval_samples_per_second": 16.017,
+      "eval_steps_per_second": 2.008,
+      "step": 3280
+    },
+    {
+      "entropy": 0.1162164019420743,
+      "epoch": 7.953530476765239,
+      "grad_norm": 0.6203398108482361,
+      "learning_rate": 4.61757244135355e-05,
+      "loss": 0.08832768201828003,
+      "mean_token_accuracy": 0.9694220684468746,
+      "num_tokens": 7643288.0,
+      "step": 3300
+    },
+    {
+      "epoch": 7.953530476765239,
+      "eval_entropy": 0.26158678129817664,
+      "eval_loss": 0.9301361441612244,
+      "eval_mean_token_accuracy": 0.8269603885291668,
+      "eval_num_tokens": 7643288.0,
+      "eval_runtime": 88.7064,
+      "eval_samples_per_second": 16.008,
+      "eval_steps_per_second": 2.007,
+      "step": 3300
+    },
+    {
+      "entropy": 0.10877202204489089,
+      "epoch": 8.0,
+      "grad_norm": 3.058673143386841,
+      "learning_rate": 4.411895954922763e-05,
+      "loss": 0.09159615635871887,
+      "mean_token_accuracy": 0.9686878834451947,
+      "num_tokens": 7688608.0,
+      "step": 3320
+    },
+    {
+      "epoch": 8.0,
+      "eval_entropy": 0.2590438057197614,
+      "eval_loss": 0.9262065887451172,
+      "eval_mean_token_accuracy": 0.8277404546067956,
+      "eval_num_tokens": 7688608.0,
+      "eval_runtime": 87.7557,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 3320
+    },
+    {
+      "entropy": 0.09969092034734786,
+      "epoch": 8.048280024140013,
+      "grad_norm": 0.2561911642551422,
+      "learning_rate": 4.2102951888817644e-05,
+      "loss": 0.07228735685348511,
+      "mean_token_accuracy": 0.9743556551635265,
+      "num_tokens": 7732856.0,
+      "step": 3340
+    },
+    {
+      "epoch": 8.048280024140013,
+      "eval_entropy": 0.2462534454933713,
+      "eval_loss": 0.9755983352661133,
+      "eval_mean_token_accuracy": 0.827510209565752,
+      "eval_num_tokens": 7732856.0,
+      "eval_runtime": 88.0297,
+      "eval_samples_per_second": 16.131,
+      "eval_steps_per_second": 2.022,
+      "step": 3340
+    },
+    {
+      "entropy": 0.09477624730207027,
+      "epoch": 8.096560048280024,
+      "grad_norm": 0.3670356571674347,
+      "learning_rate": 4.012827193862633e-05,
+      "loss": 0.0686036467552185,
+      "mean_token_accuracy": 0.9743670664727688,
+      "num_tokens": 7778585.0,
+      "step": 3360
+    },
+    {
+      "epoch": 8.096560048280024,
+      "eval_entropy": 0.2440882951356052,
+      "eval_loss": 1.0118423700332642,
+      "eval_mean_token_accuracy": 0.8271241070849172,
+      "eval_num_tokens": 7778585.0,
+      "eval_runtime": 88.1364,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 3360
+    },
+    {
+      "entropy": 0.09417980071157217,
+      "epoch": 8.144840072420036,
+      "grad_norm": 0.5785111784934998,
+      "learning_rate": 3.819547850972133e-05,
+      "loss": 0.07100698947906495,
+      "mean_token_accuracy": 0.9737804099917412,
+      "num_tokens": 7824287.0,
+      "step": 3380
+    },
+    {
+      "epoch": 8.144840072420036,
+      "eval_entropy": 0.24603172860453637,
+      "eval_loss": 1.0021991729736328,
+      "eval_mean_token_accuracy": 0.827162122793412,
+      "eval_num_tokens": 7824287.0,
+      "eval_runtime": 88.4534,
+      "eval_samples_per_second": 16.054,
+      "eval_steps_per_second": 2.012,
+      "step": 3380
+    },
+    {
+      "entropy": 0.087222247896716,
+      "epoch": 8.193120096560047,
+      "grad_norm": 0.44848504662513733,
+      "learning_rate": 3.6305118559780393e-05,
+      "loss": 0.06325234174728393,
+      "mean_token_accuracy": 0.9769335232675076,
+      "num_tokens": 7874784.0,
+      "step": 3400
+    },
+    {
+      "epoch": 8.193120096560047,
+      "eval_entropy": 0.24544863905129807,
+      "eval_loss": 1.005008339881897,
+      "eval_mean_token_accuracy": 0.8273035240976998,
+      "eval_num_tokens": 7874784.0,
+      "eval_runtime": 88.6737,
+      "eval_samples_per_second": 16.014,
+      "eval_steps_per_second": 2.007,
+      "step": 3400
+    },
+    {
+      "entropy": 0.09037275062873959,
+      "epoch": 8.24140012070006,
+      "grad_norm": 0.23529155552387238,
+      "learning_rate": 3.4457727038308435e-05,
+      "loss": 0.06589569449424744,
+      "mean_token_accuracy": 0.9751470290124417,
+      "num_tokens": 7925240.0,
+      "step": 3420
+    },
+    {
+      "epoch": 8.24140012070006,
+      "eval_entropy": 0.24203423494368456,
+      "eval_loss": 1.0052824020385742,
+      "eval_mean_token_accuracy": 0.8275575503874361,
+      "eval_num_tokens": 7925240.0,
+      "eval_runtime": 88.3486,
+      "eval_samples_per_second": 16.073,
+      "eval_steps_per_second": 2.015,
+      "step": 3420
+    },
+    {
+      "entropy": 0.10373208015225828,
+      "epoch": 8.289680144840073,
+      "grad_norm": 0.24796408414840698,
+      "learning_rate": 3.2653826735253594e-05,
+      "loss": 0.07714771628379821,
+      "mean_token_accuracy": 0.9722891598939896,
+      "num_tokens": 7966685.0,
+      "step": 3440
+    },
+    {
+      "epoch": 8.289680144840073,
+      "eval_entropy": 0.23461157050025597,
+      "eval_loss": 1.012108325958252,
+      "eval_mean_token_accuracy": 0.82825900564033,
+      "eval_num_tokens": 7966685.0,
+      "eval_runtime": 88.1018,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 3440
+    },
+    {
+      "entropy": 0.08967954274266958,
+      "epoch": 8.337960168980084,
+      "grad_norm": 0.5314674973487854,
+      "learning_rate": 3.08939281330635e-05,
+      "loss": 0.06746266484260559,
+      "mean_token_accuracy": 0.9753133699297905,
+      "num_tokens": 8013656.0,
+      "step": 3460
+    },
+    {
+      "epoch": 8.337960168980084,
+      "eval_entropy": 0.23781998589467468,
+      "eval_loss": 1.0119913816452026,
+      "eval_mean_token_accuracy": 0.8273287157664139,
+      "eval_num_tokens": 8013656.0,
+      "eval_runtime": 88.6038,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3460
+    },
+    {
+      "entropy": 0.09280927330255509,
+      "epoch": 8.386240193120097,
+      "grad_norm": 0.30658936500549316,
+      "learning_rate": 2.9178529262224946e-05,
+      "loss": 0.06936421990394592,
+      "mean_token_accuracy": 0.9742105580866337,
+      "num_tokens": 8061108.0,
+      "step": 3480
+    },
+    {
+      "epoch": 8.386240193120097,
+      "eval_entropy": 0.23672426734747512,
+      "eval_loss": 1.0162041187286377,
+      "eval_mean_token_accuracy": 0.8278394281194451,
+      "eval_num_tokens": 8061108.0,
+      "eval_runtime": 88.6015,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 3480
+    },
+    {
+      "entropy": 0.09195122267119586,
+      "epoch": 8.43452021726011,
+      "grad_norm": 0.24512454867362976,
+      "learning_rate": 2.7508115560327256e-05,
+      "loss": 0.06996064782142639,
+      "mean_token_accuracy": 0.974104854464531,
+      "num_tokens": 8106924.0,
+      "step": 3500
+    },
+    {
+      "epoch": 8.43452021726011,
+      "eval_entropy": 0.23758245828781235,
+      "eval_loss": 1.0157747268676758,
+      "eval_mean_token_accuracy": 0.8278501579600773,
+      "eval_num_tokens": 8106924.0,
+      "eval_runtime": 88.5678,
+      "eval_samples_per_second": 16.033,
+      "eval_steps_per_second": 2.01,
+      "step": 3500
+    },
+    {
+      "entropy": 0.09518477590754629,
+      "epoch": 8.48280024140012,
+      "grad_norm": 0.2373228371143341,
+      "learning_rate": 2.5883159734688868e-05,
+      "loss": 0.06832797527313232,
+      "mean_token_accuracy": 0.9743806682527065,
+      "num_tokens": 8153070.0,
+      "step": 3520
+    },
+    {
+      "epoch": 8.48280024140012,
+      "eval_entropy": 0.2379308074210467,
+      "eval_loss": 1.0180530548095703,
+      "eval_mean_token_accuracy": 0.8282477052024241,
+      "eval_num_tokens": 8153070.0,
+      "eval_runtime": 88.4651,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 3520
+    },
+    {
+      "entropy": 0.0906269280705601,
+      "epoch": 8.531080265540133,
+      "grad_norm": 0.37602657079696655,
+      "learning_rate": 2.4304121628587173e-05,
+      "loss": 0.06921588778495788,
+      "mean_token_accuracy": 0.9745571479201317,
+      "num_tokens": 8200066.0,
+      "step": 3540
+    },
+    {
+      "epoch": 8.531080265540133,
+      "eval_entropy": 0.2357490769263064,
+      "eval_loss": 1.017139196395874,
+      "eval_mean_token_accuracy": 0.828713178299786,
+      "eval_num_tokens": 8200066.0,
+      "eval_runtime": 88.6424,
+      "eval_samples_per_second": 16.019,
+      "eval_steps_per_second": 2.008,
+      "step": 3540
+    },
+    {
+      "entropy": 0.09273334196768701,
+      "epoch": 8.579360289680144,
+      "grad_norm": 0.2978838086128235,
+      "learning_rate": 2.277144809112794e-05,
+      "loss": 0.06970887780189514,
+      "mean_token_accuracy": 0.9751374177634716,
+      "num_tokens": 8246157.0,
+      "step": 3560
+    },
+    {
+      "epoch": 8.579360289680144,
+      "eval_entropy": 0.23590933138065123,
+      "eval_loss": 1.019231915473938,
+      "eval_mean_token_accuracy": 0.8279163412163767,
+      "eval_num_tokens": 8246157.0,
+      "eval_runtime": 88.2691,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 3560
+    },
+    {
+      "entropy": 0.09854201921261847,
+      "epoch": 8.627640313820157,
+      "grad_norm": 0.5381243228912354,
+      "learning_rate": 2.1285572850792413e-05,
+      "loss": 0.07123501300811767,
+      "mean_token_accuracy": 0.9723988175392151,
+      "num_tokens": 8291752.0,
+      "step": 3580
+    },
+    {
+      "epoch": 8.627640313820157,
+      "eval_entropy": 0.23592024949494372,
+      "eval_loss": 1.020250678062439,
+      "eval_mean_token_accuracy": 0.8282466226079491,
+      "eval_num_tokens": 8291752.0,
+      "eval_runtime": 87.8939,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 3580
+    },
+    {
+      "entropy": 0.09092000788077712,
+      "epoch": 8.67592033796017,
+      "grad_norm": 0.29089561104774475,
+      "learning_rate": 1.9846916392697493e-05,
+      "loss": 0.06982720494270325,
+      "mean_token_accuracy": 0.9749995484948158,
+      "num_tokens": 8337651.0,
+      "step": 3600
+    },
+    {
+      "epoch": 8.67592033796017,
+      "eval_entropy": 0.23658913382318583,
+      "eval_loss": 1.019862413406372,
+      "eval_mean_token_accuracy": 0.8280127842104836,
+      "eval_num_tokens": 8337651.0,
+      "eval_runtime": 88.2855,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 3600
+    },
+    {
+      "entropy": 0.08708424470387399,
+      "epoch": 8.72420036210018,
+      "grad_norm": 0.29904839396476746,
+      "learning_rate": 1.8455885839603106e-05,
+      "loss": 0.06801370978355407,
+      "mean_token_accuracy": 0.9755801670253277,
+      "num_tokens": 8385330.0,
+      "step": 3620
+    },
+    {
+      "epoch": 8.72420036210018,
+      "eval_entropy": 0.23627035771862845,
+      "eval_loss": 1.0204569101333618,
+      "eval_mean_token_accuracy": 0.8279967894045155,
+      "eval_num_tokens": 8385330.0,
+      "eval_runtime": 88.1753,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 3620
+    },
+    {
+      "entropy": 0.0916255566291511,
+      "epoch": 8.772480386240193,
+      "grad_norm": 0.20571033656597137,
+      "learning_rate": 1.7112874836701502e-05,
+      "loss": 0.0692249894142151,
+      "mean_token_accuracy": 0.9745148062705994,
+      "num_tokens": 8432134.0,
+      "step": 3640
+    },
+    {
+      "epoch": 8.772480386240193,
+      "eval_entropy": 0.2340821772813797,
+      "eval_loss": 1.0212821960449219,
+      "eval_mean_token_accuracy": 0.828373622358515,
+      "eval_num_tokens": 8432134.0,
+      "eval_runtime": 87.2518,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 3640
+    },
+    {
+      "entropy": 0.0926868049427867,
+      "epoch": 8.820760410380204,
+      "grad_norm": 0.2618265450000763,
+      "learning_rate": 1.5818263440220203e-05,
+      "loss": 0.07338647246360779,
+      "mean_token_accuracy": 0.9724708944559097,
+      "num_tokens": 8476758.0,
+      "step": 3660
+    },
+    {
+      "epoch": 8.820760410380204,
+      "eval_entropy": 0.23637074575330433,
+      "eval_loss": 1.0186725854873657,
+      "eval_mean_token_accuracy": 0.8277707605549459,
+      "eval_num_tokens": 8476758.0,
+      "eval_runtime": 88.4695,
+      "eval_samples_per_second": 16.051,
+      "eval_steps_per_second": 2.012,
+      "step": 3660
+    },
+    {
+      "entropy": 0.09175782897509635,
+      "epoch": 8.869040434520217,
+      "grad_norm": 0.2889852821826935,
+      "learning_rate": 1.4572418009870604e-05,
+      "loss": 0.07066041231155396,
+      "mean_token_accuracy": 0.9725673109292984,
+      "num_tokens": 8523294.0,
+      "step": 3680
+    },
+    {
+      "epoch": 8.869040434520217,
+      "eval_entropy": 0.23945164404223473,
+      "eval_loss": 1.0145231485366821,
+      "eval_mean_token_accuracy": 0.8281202523895864,
+      "eval_num_tokens": 8523294.0,
+      "eval_runtime": 88.368,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 3680
+    },
+    {
+      "entropy": 0.09265611860901117,
+      "epoch": 8.91732045866023,
+      "grad_norm": 0.41092950105667114,
+      "learning_rate": 1.3375691105172625e-05,
+      "loss": 0.06975027322769164,
+      "mean_token_accuracy": 0.974636297672987,
+      "num_tokens": 8570358.0,
+      "step": 3700
+    },
+    {
+      "epoch": 8.91732045866023,
+      "eval_entropy": 0.239588919901446,
+      "eval_loss": 1.014059066772461,
+      "eval_mean_token_accuracy": 0.828028041660116,
+      "eval_num_tokens": 8570358.0,
+      "eval_runtime": 88.1039,
+      "eval_samples_per_second": 16.117,
+      "eval_steps_per_second": 2.02,
+      "step": 3700
+    },
+    {
+      "entropy": 0.0896465003490448,
+      "epoch": 8.965600482800241,
+      "grad_norm": 0.24000254273414612,
+      "learning_rate": 1.2228421385684305e-05,
+      "loss": 0.06743953824043274,
+      "mean_token_accuracy": 0.974006575345993,
+      "num_tokens": 8618246.0,
+      "step": 3720
+    },
+    {
+      "epoch": 8.965600482800241,
+      "eval_entropy": 0.2390461369846644,
+      "eval_loss": 1.011997938156128,
+      "eval_mean_token_accuracy": 0.8284533211354459,
+      "eval_num_tokens": 8618246.0,
+      "eval_runtime": 88.6202,
+      "eval_samples_per_second": 16.023,
+      "eval_steps_per_second": 2.009,
+      "step": 3720
+    },
+    {
+      "entropy": 0.09575109251520851,
+      "epoch": 9.012070006035003,
+      "grad_norm": 0.18088576197624207,
+      "learning_rate": 1.1130933515165429e-05,
+      "loss": 0.07295305132865906,
+      "mean_token_accuracy": 0.9739722857227573,
+      "num_tokens": 8661177.0,
+      "step": 3740
+    },
+    {
+      "epoch": 9.012070006035003,
+      "eval_entropy": 0.23728882387447892,
+      "eval_loss": 1.0134038925170898,
+      "eval_mean_token_accuracy": 0.8286508951294288,
+      "eval_num_tokens": 8661177.0,
+      "eval_runtime": 88.6434,
+      "eval_samples_per_second": 16.019,
+      "eval_steps_per_second": 2.008,
+      "step": 3740
+    },
+    {
+      "entropy": 0.08883201805874705,
+      "epoch": 9.060350030175016,
+      "grad_norm": 0.23120898008346558,
+      "learning_rate": 1.008353806970132e-05,
+      "loss": 0.06326142549514771,
+      "mean_token_accuracy": 0.9780020996928215,
+      "num_tokens": 8707602.0,
+      "step": 3760
+    },
+    {
+      "epoch": 9.060350030175016,
+      "eval_entropy": 0.23416905331142832,
+      "eval_loss": 1.0284100770950317,
+      "eval_mean_token_accuracy": 0.8284083213029283,
+      "eval_num_tokens": 8707602.0,
+      "eval_runtime": 88.2933,
+      "eval_samples_per_second": 16.083,
+      "eval_steps_per_second": 2.016,
+      "step": 3760
+    },
+    {
+      "entropy": 0.08861468276008963,
+      "epoch": 9.108630054315027,
+      "grad_norm": 0.17277656495571136,
+      "learning_rate": 9.086531449813489e-06,
+      "loss": 0.06727416515350342,
+      "mean_token_accuracy": 0.9760390736162663,
+      "num_tokens": 8751287.0,
+      "step": 3780
+    },
+    {
+      "epoch": 9.108630054315027,
+      "eval_entropy": 0.23370317386442355,
+      "eval_loss": 1.0381091833114624,
+      "eval_mean_token_accuracy": 0.8282668178670862,
+      "eval_num_tokens": 8751287.0,
+      "eval_runtime": 88.5738,
+      "eval_samples_per_second": 16.032,
+      "eval_steps_per_second": 2.01,
+      "step": 3780
+    },
+    {
+      "entropy": 0.09032954541034996,
+      "epoch": 9.15691007845504,
+      "grad_norm": 0.19481085240840912,
+      "learning_rate": 8.140195796581981e-06,
+      "loss": 0.06117786169052124,
+      "mean_token_accuracy": 0.9764454193413258,
+      "num_tokens": 8799982.0,
+      "step": 3800
+    },
+    {
+      "epoch": 9.15691007845504,
+      "eval_entropy": 0.23215936609868254,
+      "eval_loss": 1.0467767715454102,
+      "eval_mean_token_accuracy": 0.828405868806196,
+      "eval_num_tokens": 8799982.0,
+      "eval_runtime": 87.9551,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 3800
+    },
+    {
+      "entropy": 0.08824449330568314,
+      "epoch": 9.20519010259505,
+      "grad_norm": 0.20382793247699738,
+      "learning_rate": 7.2447989118025455e-06,
+      "loss": 0.06422553658485412,
+      "mean_token_accuracy": 0.9763121999800205,
+      "num_tokens": 8845536.0,
+      "step": 3820
+    },
+    {
+      "epoch": 9.20519010259505,
+      "eval_entropy": 0.23169661472352704,
+      "eval_loss": 1.0529043674468994,
+      "eval_mean_token_accuracy": 0.8282025610463003,
+      "eval_num_tokens": 8845536.0,
+      "eval_runtime": 88.5548,
+      "eval_samples_per_second": 16.035,
+      "eval_steps_per_second": 2.01,
+      "step": 3820
+    },
+    {
+      "entropy": 0.07768499720841646,
+      "epoch": 9.253470126735063,
+      "grad_norm": 0.11727919429540634,
+      "learning_rate": 6.400594182202084e-06,
+      "loss": 0.057279336452484134,
+      "mean_token_accuracy": 0.9795166745781898,
+      "num_tokens": 8896053.0,
+      "step": 3840
+    },
+    {
+      "epoch": 9.253470126735063,
+      "eval_entropy": 0.2301936373114586,
+      "eval_loss": 1.0576049089431763,
+      "eval_mean_token_accuracy": 0.8283417489421502,
+      "eval_num_tokens": 8896053.0,
+      "eval_runtime": 88.7854,
+      "eval_samples_per_second": 15.994,
+      "eval_steps_per_second": 2.005,
+      "step": 3840
+    },
+    {
+      "entropy": 0.09015200780704617,
+      "epoch": 9.301750150875076,
+      "grad_norm": 0.2089058756828308,
+      "learning_rate": 5.6078205077330705e-06,
+      "loss": 0.06626194715499878,
+      "mean_token_accuracy": 0.9747506938874722,
+      "num_tokens": 8940591.0,
+      "step": 3860
+    },
+    {
+      "epoch": 9.301750150875076,
+      "eval_entropy": 0.23024900676159377,
+      "eval_loss": 1.0599786043167114,
+      "eval_mean_token_accuracy": 0.8283051915383071,
+      "eval_num_tokens": 8940591.0,
+      "eval_runtime": 88.2963,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3860
+    },
+    {
+      "entropy": 0.08484094133600592,
+      "epoch": 9.350030175015087,
+      "grad_norm": 0.21646778285503387,
+      "learning_rate": 4.866702233967785e-06,
+      "loss": 0.05998039245605469,
+      "mean_token_accuracy": 0.9771868668496608,
+      "num_tokens": 8989035.0,
+      "step": 3880
+    },
+    {
+      "epoch": 9.350030175015087,
+      "eval_entropy": 0.2300723305578982,
+      "eval_loss": 1.0633587837219238,
+      "eval_mean_token_accuracy": 0.828371500701047,
+      "eval_num_tokens": 8989035.0,
+      "eval_runtime": 88.0233,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 3880
+    },
+    {
+      "entropy": 0.08905019680969417,
+      "epoch": 9.3983101991551,
+      "grad_norm": 0.21842406690120697,
+      "learning_rate": 4.177449088611095e-06,
+      "loss": 0.06666865348815917,
+      "mean_token_accuracy": 0.9753445640206337,
+      "num_tokens": 9033383.0,
+      "step": 3900
+    },
+    {
+      "epoch": 9.3983101991551,
+      "eval_entropy": 0.2295504841911659,
+      "eval_loss": 1.0662461519241333,
+      "eval_mean_token_accuracy": 0.8284122297603093,
+      "eval_num_tokens": 9033383.0,
+      "eval_runtime": 88.5525,
+      "eval_samples_per_second": 16.036,
+      "eval_steps_per_second": 2.01,
+      "step": 3900
+    },
+    {
+      "entropy": 0.08582521872594953,
+      "epoch": 9.446590223295111,
+      "grad_norm": 0.2000240683555603,
+      "learning_rate": 3.540256122150088e-06,
+      "loss": 0.06327944993972778,
+      "mean_token_accuracy": 0.9765432856976985,
+      "num_tokens": 9079672.0,
+      "step": 3920
+    },
+    {
+      "epoch": 9.446590223295111,
+      "eval_entropy": 0.2297678625818049,
+      "eval_loss": 1.0676169395446777,
+      "eval_mean_token_accuracy": 0.8283703879693921,
+      "eval_num_tokens": 9079672.0,
+      "eval_runtime": 87.6897,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 3920
+    },
+    {
+      "entropy": 0.09261773955076932,
+      "epoch": 9.494870247435124,
+      "grad_norm": 0.19104281067848206,
+      "learning_rate": 2.955303652656991e-06,
+      "loss": 0.06111966967582703,
+      "mean_token_accuracy": 0.9738784030079841,
+      "num_tokens": 9127647.0,
+      "step": 3940
+    },
+    {
+      "epoch": 9.494870247435124,
+      "eval_entropy": 0.22986975881490815,
+      "eval_loss": 1.0687925815582275,
+      "eval_mean_token_accuracy": 0.8282996858103892,
+      "eval_num_tokens": 9127647.0,
+      "eval_runtime": 88.2224,
+      "eval_samples_per_second": 16.096,
+      "eval_steps_per_second": 2.018,
+      "step": 3940
+    },
+    {
+      "entropy": 0.08347254800610245,
+      "epoch": 9.543150271575136,
+      "grad_norm": 0.18671369552612305,
+      "learning_rate": 2.422757214761356e-06,
+      "loss": 0.06384263038635254,
+      "mean_token_accuracy": 0.9770128890872002,
+      "num_tokens": 9172587.0,
+      "step": 3960
+    },
+    {
+      "epoch": 9.543150271575136,
+      "eval_entropy": 0.22975517851248217,
+      "eval_loss": 1.0701013803482056,
+      "eval_mean_token_accuracy": 0.8282244285840666,
+      "eval_num_tokens": 9172587.0,
+      "eval_runtime": 88.5794,
+      "eval_samples_per_second": 16.031,
+      "eval_steps_per_second": 2.009,
+      "step": 3960
+    },
+    {
+      "entropy": 0.08136763009242713,
+      "epoch": 9.591430295715147,
+      "grad_norm": 0.24811454117298126,
+      "learning_rate": 1.942767512805673e-06,
+      "loss": 0.06044899821281433,
+      "mean_token_accuracy": 0.9776685133576393,
+      "num_tokens": 9219278.0,
+      "step": 3980
+    },
+    {
+      "epoch": 9.591430295715147,
+      "eval_entropy": 0.22938149347064202,
+      "eval_loss": 1.0711463689804077,
+      "eval_mean_token_accuracy": 0.8282721410976367,
+      "eval_num_tokens": 9219278.0,
+      "eval_runtime": 88.4978,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 3980
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.8971635425970586e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-40/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-40/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-40/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-40/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-40/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-40/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-40/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-40/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-40/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-40/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-40/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-40/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-40/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-40/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..a8b1dc24f245625563d9ff26f7666b6b0b7952c9
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-40/trainer_state.json
@@ -0,0 +1,76 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.09656004828002414,
+  "eval_steps": 20,
+  "global_step": 40,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3980456012292096.0,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-400/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-400/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-400/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-400/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-400/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-400/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-400/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-400/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-400/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-400/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-400/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-400/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-400/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-400/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..5572511bea1989f5cb896b94b70fabb6a0ae3fa0
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-400/trainer_state.json
@@ -0,0 +1,454 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.9656004828002414,
+  "eval_steps": 20,
+  "global_step": 400,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.928464556742246e+16,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4000/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4000/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4000/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4000/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4000/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4000/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4000/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4000/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4000/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4000/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4000/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4000/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4000/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4000/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..825df727d6aab3debeb0074036a1c5191c68132c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4000/trainer_state.json
@@ -0,0 +1,4234 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 9.63971031985516,
+  "eval_steps": 20,
+  "global_step": 4000,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    },
+    {
+      "entropy": 0.1326121488586068,
+      "epoch": 7.181050090525045,
+      "grad_norm": 0.4843117296695709,
+      "learning_rate": 8.411870386507724e-05,
+      "loss": 0.09537842869758606,
+      "mean_token_accuracy": 0.9645151920616627,
+      "num_tokens": 6898444.0,
+      "step": 2980
+    },
+    {
+      "epoch": 7.181050090525045,
+      "eval_entropy": 0.27999477783280813,
+      "eval_loss": 0.9031037092208862,
+      "eval_mean_token_accuracy": 0.8261882204018282,
+      "eval_num_tokens": 6898444.0,
+      "eval_runtime": 88.5602,
+      "eval_samples_per_second": 16.034,
+      "eval_steps_per_second": 2.01,
+      "step": 2980
+    },
+    {
+      "entropy": 0.1156751093454659,
+      "epoch": 7.229330114665057,
+      "grad_norm": 0.5297922492027283,
+      "learning_rate": 8.149629351430326e-05,
+      "loss": 0.09203402996063233,
+      "mean_token_accuracy": 0.9689733698964119,
+      "num_tokens": 6947106.0,
+      "step": 3000
+    },
+    {
+      "epoch": 7.229330114665057,
+      "eval_entropy": 0.2740225845508361,
+      "eval_loss": 0.8876440525054932,
+      "eval_mean_token_accuracy": 0.8265116857678703,
+      "eval_num_tokens": 6947106.0,
+      "eval_runtime": 87.7971,
+      "eval_samples_per_second": 16.174,
+      "eval_steps_per_second": 2.027,
+      "step": 3000
+    },
+    {
+      "entropy": 0.12116114404052496,
+      "epoch": 7.27761013880507,
+      "grad_norm": 0.5467398166656494,
+      "learning_rate": 7.890406302404618e-05,
+      "loss": 0.0933056652545929,
+      "mean_token_accuracy": 0.9676455594599247,
+      "num_tokens": 6992784.0,
+      "step": 3020
+    },
+    {
+      "epoch": 7.27761013880507,
+      "eval_entropy": 0.27238342793804876,
+      "eval_loss": 0.9166117310523987,
+      "eval_mean_token_accuracy": 0.825499015242866,
+      "eval_num_tokens": 6992784.0,
+      "eval_runtime": 88.3276,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 3020
+    },
+    {
+      "entropy": 0.11715119928121567,
+      "epoch": 7.325890162945082,
+      "grad_norm": 0.4976702332496643,
+      "learning_rate": 7.634274596487176e-05,
+      "loss": 0.09449659585952759,
+      "mean_token_accuracy": 0.966506478190422,
+      "num_tokens": 7040482.0,
+      "step": 3040
+    },
+    {
+      "epoch": 7.325890162945082,
+      "eval_entropy": 0.2788975556077582,
+      "eval_loss": 0.8833065629005432,
+      "eval_mean_token_accuracy": 0.8269920596915684,
+      "eval_num_tokens": 7040482.0,
+      "eval_runtime": 88.9016,
+      "eval_samples_per_second": 15.973,
+      "eval_steps_per_second": 2.002,
+      "step": 3040
+    },
+    {
+      "entropy": 0.12565911412239075,
+      "epoch": 7.3741701870850935,
+      "grad_norm": 0.5728679895401001,
+      "learning_rate": 7.381306715921025e-05,
+      "loss": 0.09626244902610778,
+      "mean_token_accuracy": 0.9661444149911403,
+      "num_tokens": 7086261.0,
+      "step": 3060
+    },
+    {
+      "epoch": 7.3741701870850935,
+      "eval_entropy": 0.2762492011939542,
+      "eval_loss": 0.8961160778999329,
+      "eval_mean_token_accuracy": 0.826369910092836,
+      "eval_num_tokens": 7086261.0,
+      "eval_runtime": 88.5204,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 3060
+    },
+    {
+      "entropy": 0.12258567921817302,
+      "epoch": 7.422450211225105,
+      "grad_norm": 0.468985915184021,
+      "learning_rate": 7.131574247624031e-05,
+      "loss": 0.09120266437530518,
+      "mean_token_accuracy": 0.9674938447773457,
+      "num_tokens": 7133247.0,
+      "step": 3080
+    },
+    {
+      "epoch": 7.422450211225105,
+      "eval_entropy": 0.2688512086366,
+      "eval_loss": 0.9048654437065125,
+      "eval_mean_token_accuracy": 0.8265582302982888,
+      "eval_num_tokens": 7133247.0,
+      "eval_runtime": 88.164,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 3080
+    },
+    {
+      "entropy": 0.12070536343380808,
+      "epoch": 7.470730235365117,
+      "grad_norm": 0.5310841798782349,
+      "learning_rate": 6.885147862930688e-05,
+      "loss": 0.09386247992515565,
+      "mean_token_accuracy": 0.96664222702384,
+      "num_tokens": 7177810.0,
+      "step": 3100
+    },
+    {
+      "epoch": 7.470730235365117,
+      "eval_entropy": 0.26289632874593305,
+      "eval_loss": 0.902837336063385,
+      "eval_mean_token_accuracy": 0.828052184554968,
+      "eval_num_tokens": 7177810.0,
+      "eval_runtime": 88.1201,
+      "eval_samples_per_second": 16.114,
+      "eval_steps_per_second": 2.02,
+      "step": 3100
+    },
+    {
+      "entropy": 0.11322665181942285,
+      "epoch": 7.51901025950513,
+      "grad_norm": 0.6846067905426025,
+      "learning_rate": 6.642097297592908e-05,
+      "loss": 0.08772538304328918,
+      "mean_token_accuracy": 0.9691362135112286,
+      "num_tokens": 7225302.0,
+      "step": 3120
+    },
+    {
+      "epoch": 7.51901025950513,
+      "eval_entropy": 0.2670605613776807,
+      "eval_loss": 0.9130566120147705,
+      "eval_mean_token_accuracy": 0.826366903741708,
+      "eval_num_tokens": 7225302.0,
+      "eval_runtime": 88.6035,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3120
+    },
+    {
+      "entropy": 0.10985926580615342,
+      "epoch": 7.567290283645142,
+      "grad_norm": 0.4799748361110687,
+      "learning_rate": 6.402491332045683e-05,
+      "loss": 0.08672296404838561,
+      "mean_token_accuracy": 0.9698824048042297,
+      "num_tokens": 7274881.0,
+      "step": 3140
+    },
+    {
+      "epoch": 7.567290283645142,
+      "eval_entropy": 0.26657010866015146,
+      "eval_loss": 0.9056380987167358,
+      "eval_mean_token_accuracy": 0.8272326461384806,
+      "eval_num_tokens": 7274881.0,
+      "eval_runtime": 88.2967,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3140
+    },
+    {
+      "entropy": 0.11890473761595785,
+      "epoch": 7.615570307785154,
+      "grad_norm": 0.4081912934780121,
+      "learning_rate": 6.166397771942987e-05,
+      "loss": 0.09396827220916748,
+      "mean_token_accuracy": 0.967512384802103,
+      "num_tokens": 7321163.0,
+      "step": 3160
+    },
+    {
+      "epoch": 7.615570307785154,
+      "eval_entropy": 0.26364776575833226,
+      "eval_loss": 0.9204733967781067,
+      "eval_mean_token_accuracy": 0.827419729714983,
+      "eval_num_tokens": 7321163.0,
+      "eval_runtime": 88.3441,
+      "eval_samples_per_second": 16.074,
+      "eval_steps_per_second": 2.015,
+      "step": 3160
+    },
+    {
+      "entropy": 0.12209134344011545,
+      "epoch": 7.663850331925166,
+      "grad_norm": 0.5159774422645569,
+      "learning_rate": 5.93388342896962e-05,
+      "loss": 0.09292978644371033,
+      "mean_token_accuracy": 0.9658421002328396,
+      "num_tokens": 7366217.0,
+      "step": 3180
+    },
+    {
+      "epoch": 7.663850331925166,
+      "eval_entropy": 0.2714110068725736,
+      "eval_loss": 0.9143092036247253,
+      "eval_mean_token_accuracy": 0.8262251099843657,
+      "eval_num_tokens": 7366217.0,
+      "eval_runtime": 87.9635,
+      "eval_samples_per_second": 16.143,
+      "eval_steps_per_second": 2.024,
+      "step": 3180
+    },
+    {
+      "entropy": 0.11955609126016498,
+      "epoch": 7.712130356065178,
+      "grad_norm": 0.7005976438522339,
+      "learning_rate": 5.705014101934252e-05,
+      "loss": 0.09319776892662049,
+      "mean_token_accuracy": 0.9674513377249241,
+      "num_tokens": 7410887.0,
+      "step": 3200
+    },
+    {
+      "epoch": 7.712130356065178,
+      "eval_entropy": 0.2626291513442993,
+      "eval_loss": 0.91884446144104,
+      "eval_mean_token_accuracy": 0.827563955207889,
+      "eval_num_tokens": 7410887.0,
+      "eval_runtime": 87.8849,
+      "eval_samples_per_second": 16.157,
+      "eval_steps_per_second": 2.025,
+      "step": 3200
+    },
+    {
+      "entropy": 0.11767532899975777,
+      "epoch": 7.76041038020519,
+      "grad_norm": 0.4675362706184387,
+      "learning_rate": 5.479854558149187e-05,
+      "loss": 0.09038087725639343,
+      "mean_token_accuracy": 0.9669680051505566,
+      "num_tokens": 7457578.0,
+      "step": 3220
+    },
+    {
+      "epoch": 7.76041038020519,
+      "eval_entropy": 0.26581037203582486,
+      "eval_loss": 0.9146883487701416,
+      "eval_mean_token_accuracy": 0.8270345812433222,
+      "eval_num_tokens": 7457578.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 3220
+    },
+    {
+      "entropy": 0.11383061073720455,
+      "epoch": 7.808690404345202,
+      "grad_norm": 0.541247546672821,
+      "learning_rate": 5.258468515101918e-05,
+      "loss": 0.08942405581474304,
+      "mean_token_accuracy": 0.968398705124855,
+      "num_tokens": 7505761.0,
+      "step": 3240
+    },
+    {
+      "epoch": 7.808690404345202,
+      "eval_entropy": 0.26687998960861997,
+      "eval_loss": 0.9037085175514221,
+      "eval_mean_token_accuracy": 0.8274977863504646,
+      "eval_num_tokens": 7505761.0,
+      "eval_runtime": 88.5163,
+      "eval_samples_per_second": 16.042,
+      "eval_steps_per_second": 2.011,
+      "step": 3240
+    },
+    {
+      "entropy": 0.12004058873280883,
+      "epoch": 7.856970428485214,
+      "grad_norm": 0.5289552211761475,
+      "learning_rate": 5.040918622423899e-05,
+      "loss": 0.09064805507659912,
+      "mean_token_accuracy": 0.9673330299556255,
+      "num_tokens": 7552557.0,
+      "step": 3260
+    },
+    {
+      "epoch": 7.856970428485214,
+      "eval_entropy": 0.26288379251622085,
+      "eval_loss": 0.9146178364753723,
+      "eval_mean_token_accuracy": 0.8276662448149049,
+      "eval_num_tokens": 7552557.0,
+      "eval_runtime": 88.0723,
+      "eval_samples_per_second": 16.123,
+      "eval_steps_per_second": 2.021,
+      "step": 3260
+    },
+    {
+      "entropy": 0.1180689456872642,
+      "epoch": 7.905250452625227,
+      "grad_norm": 0.4754861891269684,
+      "learning_rate": 4.827266444161377e-05,
+      "loss": 0.09402655363082886,
+      "mean_token_accuracy": 0.9663361787796021,
+      "num_tokens": 7596998.0,
+      "step": 3280
+    },
+    {
+      "epoch": 7.905250452625227,
+      "eval_entropy": 0.261690996922134,
+      "eval_loss": 0.9120186567306519,
+      "eval_mean_token_accuracy": 0.8278995015647974,
+      "eval_num_tokens": 7596998.0,
+      "eval_runtime": 88.6573,
+      "eval_samples_per_second": 16.017,
+      "eval_steps_per_second": 2.008,
+      "step": 3280
+    },
+    {
+      "entropy": 0.1162164019420743,
+      "epoch": 7.953530476765239,
+      "grad_norm": 0.6203398108482361,
+      "learning_rate": 4.61757244135355e-05,
+      "loss": 0.08832768201828003,
+      "mean_token_accuracy": 0.9694220684468746,
+      "num_tokens": 7643288.0,
+      "step": 3300
+    },
+    {
+      "epoch": 7.953530476765239,
+      "eval_entropy": 0.26158678129817664,
+      "eval_loss": 0.9301361441612244,
+      "eval_mean_token_accuracy": 0.8269603885291668,
+      "eval_num_tokens": 7643288.0,
+      "eval_runtime": 88.7064,
+      "eval_samples_per_second": 16.008,
+      "eval_steps_per_second": 2.007,
+      "step": 3300
+    },
+    {
+      "entropy": 0.10877202204489089,
+      "epoch": 8.0,
+      "grad_norm": 3.058673143386841,
+      "learning_rate": 4.411895954922763e-05,
+      "loss": 0.09159615635871887,
+      "mean_token_accuracy": 0.9686878834451947,
+      "num_tokens": 7688608.0,
+      "step": 3320
+    },
+    {
+      "epoch": 8.0,
+      "eval_entropy": 0.2590438057197614,
+      "eval_loss": 0.9262065887451172,
+      "eval_mean_token_accuracy": 0.8277404546067956,
+      "eval_num_tokens": 7688608.0,
+      "eval_runtime": 87.7557,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 3320
+    },
+    {
+      "entropy": 0.09969092034734786,
+      "epoch": 8.048280024140013,
+      "grad_norm": 0.2561911642551422,
+      "learning_rate": 4.2102951888817644e-05,
+      "loss": 0.07228735685348511,
+      "mean_token_accuracy": 0.9743556551635265,
+      "num_tokens": 7732856.0,
+      "step": 3340
+    },
+    {
+      "epoch": 8.048280024140013,
+      "eval_entropy": 0.2462534454933713,
+      "eval_loss": 0.9755983352661133,
+      "eval_mean_token_accuracy": 0.827510209565752,
+      "eval_num_tokens": 7732856.0,
+      "eval_runtime": 88.0297,
+      "eval_samples_per_second": 16.131,
+      "eval_steps_per_second": 2.022,
+      "step": 3340
+    },
+    {
+      "entropy": 0.09477624730207027,
+      "epoch": 8.096560048280024,
+      "grad_norm": 0.3670356571674347,
+      "learning_rate": 4.012827193862633e-05,
+      "loss": 0.0686036467552185,
+      "mean_token_accuracy": 0.9743670664727688,
+      "num_tokens": 7778585.0,
+      "step": 3360
+    },
+    {
+      "epoch": 8.096560048280024,
+      "eval_entropy": 0.2440882951356052,
+      "eval_loss": 1.0118423700332642,
+      "eval_mean_token_accuracy": 0.8271241070849172,
+      "eval_num_tokens": 7778585.0,
+      "eval_runtime": 88.1364,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 3360
+    },
+    {
+      "entropy": 0.09417980071157217,
+      "epoch": 8.144840072420036,
+      "grad_norm": 0.5785111784934998,
+      "learning_rate": 3.819547850972133e-05,
+      "loss": 0.07100698947906495,
+      "mean_token_accuracy": 0.9737804099917412,
+      "num_tokens": 7824287.0,
+      "step": 3380
+    },
+    {
+      "epoch": 8.144840072420036,
+      "eval_entropy": 0.24603172860453637,
+      "eval_loss": 1.0021991729736328,
+      "eval_mean_token_accuracy": 0.827162122793412,
+      "eval_num_tokens": 7824287.0,
+      "eval_runtime": 88.4534,
+      "eval_samples_per_second": 16.054,
+      "eval_steps_per_second": 2.012,
+      "step": 3380
+    },
+    {
+      "entropy": 0.087222247896716,
+      "epoch": 8.193120096560047,
+      "grad_norm": 0.44848504662513733,
+      "learning_rate": 3.6305118559780393e-05,
+      "loss": 0.06325234174728393,
+      "mean_token_accuracy": 0.9769335232675076,
+      "num_tokens": 7874784.0,
+      "step": 3400
+    },
+    {
+      "epoch": 8.193120096560047,
+      "eval_entropy": 0.24544863905129807,
+      "eval_loss": 1.005008339881897,
+      "eval_mean_token_accuracy": 0.8273035240976998,
+      "eval_num_tokens": 7874784.0,
+      "eval_runtime": 88.6737,
+      "eval_samples_per_second": 16.014,
+      "eval_steps_per_second": 2.007,
+      "step": 3400
+    },
+    {
+      "entropy": 0.09037275062873959,
+      "epoch": 8.24140012070006,
+      "grad_norm": 0.23529155552387238,
+      "learning_rate": 3.4457727038308435e-05,
+      "loss": 0.06589569449424744,
+      "mean_token_accuracy": 0.9751470290124417,
+      "num_tokens": 7925240.0,
+      "step": 3420
+    },
+    {
+      "epoch": 8.24140012070006,
+      "eval_entropy": 0.24203423494368456,
+      "eval_loss": 1.0052824020385742,
+      "eval_mean_token_accuracy": 0.8275575503874361,
+      "eval_num_tokens": 7925240.0,
+      "eval_runtime": 88.3486,
+      "eval_samples_per_second": 16.073,
+      "eval_steps_per_second": 2.015,
+      "step": 3420
+    },
+    {
+      "entropy": 0.10373208015225828,
+      "epoch": 8.289680144840073,
+      "grad_norm": 0.24796408414840698,
+      "learning_rate": 3.2653826735253594e-05,
+      "loss": 0.07714771628379821,
+      "mean_token_accuracy": 0.9722891598939896,
+      "num_tokens": 7966685.0,
+      "step": 3440
+    },
+    {
+      "epoch": 8.289680144840073,
+      "eval_entropy": 0.23461157050025597,
+      "eval_loss": 1.012108325958252,
+      "eval_mean_token_accuracy": 0.82825900564033,
+      "eval_num_tokens": 7966685.0,
+      "eval_runtime": 88.1018,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 3440
+    },
+    {
+      "entropy": 0.08967954274266958,
+      "epoch": 8.337960168980084,
+      "grad_norm": 0.5314674973487854,
+      "learning_rate": 3.08939281330635e-05,
+      "loss": 0.06746266484260559,
+      "mean_token_accuracy": 0.9753133699297905,
+      "num_tokens": 8013656.0,
+      "step": 3460
+    },
+    {
+      "epoch": 8.337960168980084,
+      "eval_entropy": 0.23781998589467468,
+      "eval_loss": 1.0119913816452026,
+      "eval_mean_token_accuracy": 0.8273287157664139,
+      "eval_num_tokens": 8013656.0,
+      "eval_runtime": 88.6038,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3460
+    },
+    {
+      "entropy": 0.09280927330255509,
+      "epoch": 8.386240193120097,
+      "grad_norm": 0.30658936500549316,
+      "learning_rate": 2.9178529262224946e-05,
+      "loss": 0.06936421990394592,
+      "mean_token_accuracy": 0.9742105580866337,
+      "num_tokens": 8061108.0,
+      "step": 3480
+    },
+    {
+      "epoch": 8.386240193120097,
+      "eval_entropy": 0.23672426734747512,
+      "eval_loss": 1.0162041187286377,
+      "eval_mean_token_accuracy": 0.8278394281194451,
+      "eval_num_tokens": 8061108.0,
+      "eval_runtime": 88.6015,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 3480
+    },
+    {
+      "entropy": 0.09195122267119586,
+      "epoch": 8.43452021726011,
+      "grad_norm": 0.24512454867362976,
+      "learning_rate": 2.7508115560327256e-05,
+      "loss": 0.06996064782142639,
+      "mean_token_accuracy": 0.974104854464531,
+      "num_tokens": 8106924.0,
+      "step": 3500
+    },
+    {
+      "epoch": 8.43452021726011,
+      "eval_entropy": 0.23758245828781235,
+      "eval_loss": 1.0157747268676758,
+      "eval_mean_token_accuracy": 0.8278501579600773,
+      "eval_num_tokens": 8106924.0,
+      "eval_runtime": 88.5678,
+      "eval_samples_per_second": 16.033,
+      "eval_steps_per_second": 2.01,
+      "step": 3500
+    },
+    {
+      "entropy": 0.09518477590754629,
+      "epoch": 8.48280024140012,
+      "grad_norm": 0.2373228371143341,
+      "learning_rate": 2.5883159734688868e-05,
+      "loss": 0.06832797527313232,
+      "mean_token_accuracy": 0.9743806682527065,
+      "num_tokens": 8153070.0,
+      "step": 3520
+    },
+    {
+      "epoch": 8.48280024140012,
+      "eval_entropy": 0.2379308074210467,
+      "eval_loss": 1.0180530548095703,
+      "eval_mean_token_accuracy": 0.8282477052024241,
+      "eval_num_tokens": 8153070.0,
+      "eval_runtime": 88.4651,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 3520
+    },
+    {
+      "entropy": 0.0906269280705601,
+      "epoch": 8.531080265540133,
+      "grad_norm": 0.37602657079696655,
+      "learning_rate": 2.4304121628587173e-05,
+      "loss": 0.06921588778495788,
+      "mean_token_accuracy": 0.9745571479201317,
+      "num_tokens": 8200066.0,
+      "step": 3540
+    },
+    {
+      "epoch": 8.531080265540133,
+      "eval_entropy": 0.2357490769263064,
+      "eval_loss": 1.017139196395874,
+      "eval_mean_token_accuracy": 0.828713178299786,
+      "eval_num_tokens": 8200066.0,
+      "eval_runtime": 88.6424,
+      "eval_samples_per_second": 16.019,
+      "eval_steps_per_second": 2.008,
+      "step": 3540
+    },
+    {
+      "entropy": 0.09273334196768701,
+      "epoch": 8.579360289680144,
+      "grad_norm": 0.2978838086128235,
+      "learning_rate": 2.277144809112794e-05,
+      "loss": 0.06970887780189514,
+      "mean_token_accuracy": 0.9751374177634716,
+      "num_tokens": 8246157.0,
+      "step": 3560
+    },
+    {
+      "epoch": 8.579360289680144,
+      "eval_entropy": 0.23590933138065123,
+      "eval_loss": 1.019231915473938,
+      "eval_mean_token_accuracy": 0.8279163412163767,
+      "eval_num_tokens": 8246157.0,
+      "eval_runtime": 88.2691,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 3560
+    },
+    {
+      "entropy": 0.09854201921261847,
+      "epoch": 8.627640313820157,
+      "grad_norm": 0.5381243228912354,
+      "learning_rate": 2.1285572850792413e-05,
+      "loss": 0.07123501300811767,
+      "mean_token_accuracy": 0.9723988175392151,
+      "num_tokens": 8291752.0,
+      "step": 3580
+    },
+    {
+      "epoch": 8.627640313820157,
+      "eval_entropy": 0.23592024949494372,
+      "eval_loss": 1.020250678062439,
+      "eval_mean_token_accuracy": 0.8282466226079491,
+      "eval_num_tokens": 8291752.0,
+      "eval_runtime": 87.8939,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 3580
+    },
+    {
+      "entropy": 0.09092000788077712,
+      "epoch": 8.67592033796017,
+      "grad_norm": 0.29089561104774475,
+      "learning_rate": 1.9846916392697493e-05,
+      "loss": 0.06982720494270325,
+      "mean_token_accuracy": 0.9749995484948158,
+      "num_tokens": 8337651.0,
+      "step": 3600
+    },
+    {
+      "epoch": 8.67592033796017,
+      "eval_entropy": 0.23658913382318583,
+      "eval_loss": 1.019862413406372,
+      "eval_mean_token_accuracy": 0.8280127842104836,
+      "eval_num_tokens": 8337651.0,
+      "eval_runtime": 88.2855,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 3600
+    },
+    {
+      "entropy": 0.08708424470387399,
+      "epoch": 8.72420036210018,
+      "grad_norm": 0.29904839396476746,
+      "learning_rate": 1.8455885839603106e-05,
+      "loss": 0.06801370978355407,
+      "mean_token_accuracy": 0.9755801670253277,
+      "num_tokens": 8385330.0,
+      "step": 3620
+    },
+    {
+      "epoch": 8.72420036210018,
+      "eval_entropy": 0.23627035771862845,
+      "eval_loss": 1.0204569101333618,
+      "eval_mean_token_accuracy": 0.8279967894045155,
+      "eval_num_tokens": 8385330.0,
+      "eval_runtime": 88.1753,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 3620
+    },
+    {
+      "entropy": 0.0916255566291511,
+      "epoch": 8.772480386240193,
+      "grad_norm": 0.20571033656597137,
+      "learning_rate": 1.7112874836701502e-05,
+      "loss": 0.0692249894142151,
+      "mean_token_accuracy": 0.9745148062705994,
+      "num_tokens": 8432134.0,
+      "step": 3640
+    },
+    {
+      "epoch": 8.772480386240193,
+      "eval_entropy": 0.2340821772813797,
+      "eval_loss": 1.0212821960449219,
+      "eval_mean_token_accuracy": 0.828373622358515,
+      "eval_num_tokens": 8432134.0,
+      "eval_runtime": 87.2518,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 3640
+    },
+    {
+      "entropy": 0.0926868049427867,
+      "epoch": 8.820760410380204,
+      "grad_norm": 0.2618265450000763,
+      "learning_rate": 1.5818263440220203e-05,
+      "loss": 0.07338647246360779,
+      "mean_token_accuracy": 0.9724708944559097,
+      "num_tokens": 8476758.0,
+      "step": 3660
+    },
+    {
+      "epoch": 8.820760410380204,
+      "eval_entropy": 0.23637074575330433,
+      "eval_loss": 1.0186725854873657,
+      "eval_mean_token_accuracy": 0.8277707605549459,
+      "eval_num_tokens": 8476758.0,
+      "eval_runtime": 88.4695,
+      "eval_samples_per_second": 16.051,
+      "eval_steps_per_second": 2.012,
+      "step": 3660
+    },
+    {
+      "entropy": 0.09175782897509635,
+      "epoch": 8.869040434520217,
+      "grad_norm": 0.2889852821826935,
+      "learning_rate": 1.4572418009870604e-05,
+      "loss": 0.07066041231155396,
+      "mean_token_accuracy": 0.9725673109292984,
+      "num_tokens": 8523294.0,
+      "step": 3680
+    },
+    {
+      "epoch": 8.869040434520217,
+      "eval_entropy": 0.23945164404223473,
+      "eval_loss": 1.0145231485366821,
+      "eval_mean_token_accuracy": 0.8281202523895864,
+      "eval_num_tokens": 8523294.0,
+      "eval_runtime": 88.368,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 3680
+    },
+    {
+      "entropy": 0.09265611860901117,
+      "epoch": 8.91732045866023,
+      "grad_norm": 0.41092950105667114,
+      "learning_rate": 1.3375691105172625e-05,
+      "loss": 0.06975027322769164,
+      "mean_token_accuracy": 0.974636297672987,
+      "num_tokens": 8570358.0,
+      "step": 3700
+    },
+    {
+      "epoch": 8.91732045866023,
+      "eval_entropy": 0.239588919901446,
+      "eval_loss": 1.014059066772461,
+      "eval_mean_token_accuracy": 0.828028041660116,
+      "eval_num_tokens": 8570358.0,
+      "eval_runtime": 88.1039,
+      "eval_samples_per_second": 16.117,
+      "eval_steps_per_second": 2.02,
+      "step": 3700
+    },
+    {
+      "entropy": 0.0896465003490448,
+      "epoch": 8.965600482800241,
+      "grad_norm": 0.24000254273414612,
+      "learning_rate": 1.2228421385684305e-05,
+      "loss": 0.06743953824043274,
+      "mean_token_accuracy": 0.974006575345993,
+      "num_tokens": 8618246.0,
+      "step": 3720
+    },
+    {
+      "epoch": 8.965600482800241,
+      "eval_entropy": 0.2390461369846644,
+      "eval_loss": 1.011997938156128,
+      "eval_mean_token_accuracy": 0.8284533211354459,
+      "eval_num_tokens": 8618246.0,
+      "eval_runtime": 88.6202,
+      "eval_samples_per_second": 16.023,
+      "eval_steps_per_second": 2.009,
+      "step": 3720
+    },
+    {
+      "entropy": 0.09575109251520851,
+      "epoch": 9.012070006035003,
+      "grad_norm": 0.18088576197624207,
+      "learning_rate": 1.1130933515165429e-05,
+      "loss": 0.07295305132865906,
+      "mean_token_accuracy": 0.9739722857227573,
+      "num_tokens": 8661177.0,
+      "step": 3740
+    },
+    {
+      "epoch": 9.012070006035003,
+      "eval_entropy": 0.23728882387447892,
+      "eval_loss": 1.0134038925170898,
+      "eval_mean_token_accuracy": 0.8286508951294288,
+      "eval_num_tokens": 8661177.0,
+      "eval_runtime": 88.6434,
+      "eval_samples_per_second": 16.019,
+      "eval_steps_per_second": 2.008,
+      "step": 3740
+    },
+    {
+      "entropy": 0.08883201805874705,
+      "epoch": 9.060350030175016,
+      "grad_norm": 0.23120898008346558,
+      "learning_rate": 1.008353806970132e-05,
+      "loss": 0.06326142549514771,
+      "mean_token_accuracy": 0.9780020996928215,
+      "num_tokens": 8707602.0,
+      "step": 3760
+    },
+    {
+      "epoch": 9.060350030175016,
+      "eval_entropy": 0.23416905331142832,
+      "eval_loss": 1.0284100770950317,
+      "eval_mean_token_accuracy": 0.8284083213029283,
+      "eval_num_tokens": 8707602.0,
+      "eval_runtime": 88.2933,
+      "eval_samples_per_second": 16.083,
+      "eval_steps_per_second": 2.016,
+      "step": 3760
+    },
+    {
+      "entropy": 0.08861468276008963,
+      "epoch": 9.108630054315027,
+      "grad_norm": 0.17277656495571136,
+      "learning_rate": 9.086531449813489e-06,
+      "loss": 0.06727416515350342,
+      "mean_token_accuracy": 0.9760390736162663,
+      "num_tokens": 8751287.0,
+      "step": 3780
+    },
+    {
+      "epoch": 9.108630054315027,
+      "eval_entropy": 0.23370317386442355,
+      "eval_loss": 1.0381091833114624,
+      "eval_mean_token_accuracy": 0.8282668178670862,
+      "eval_num_tokens": 8751287.0,
+      "eval_runtime": 88.5738,
+      "eval_samples_per_second": 16.032,
+      "eval_steps_per_second": 2.01,
+      "step": 3780
+    },
+    {
+      "entropy": 0.09032954541034996,
+      "epoch": 9.15691007845504,
+      "grad_norm": 0.19481085240840912,
+      "learning_rate": 8.140195796581981e-06,
+      "loss": 0.06117786169052124,
+      "mean_token_accuracy": 0.9764454193413258,
+      "num_tokens": 8799982.0,
+      "step": 3800
+    },
+    {
+      "epoch": 9.15691007845504,
+      "eval_entropy": 0.23215936609868254,
+      "eval_loss": 1.0467767715454102,
+      "eval_mean_token_accuracy": 0.828405868806196,
+      "eval_num_tokens": 8799982.0,
+      "eval_runtime": 87.9551,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 3800
+    },
+    {
+      "entropy": 0.08824449330568314,
+      "epoch": 9.20519010259505,
+      "grad_norm": 0.20382793247699738,
+      "learning_rate": 7.2447989118025455e-06,
+      "loss": 0.06422553658485412,
+      "mean_token_accuracy": 0.9763121999800205,
+      "num_tokens": 8845536.0,
+      "step": 3820
+    },
+    {
+      "epoch": 9.20519010259505,
+      "eval_entropy": 0.23169661472352704,
+      "eval_loss": 1.0529043674468994,
+      "eval_mean_token_accuracy": 0.8282025610463003,
+      "eval_num_tokens": 8845536.0,
+      "eval_runtime": 88.5548,
+      "eval_samples_per_second": 16.035,
+      "eval_steps_per_second": 2.01,
+      "step": 3820
+    },
+    {
+      "entropy": 0.07768499720841646,
+      "epoch": 9.253470126735063,
+      "grad_norm": 0.11727919429540634,
+      "learning_rate": 6.400594182202084e-06,
+      "loss": 0.057279336452484134,
+      "mean_token_accuracy": 0.9795166745781898,
+      "num_tokens": 8896053.0,
+      "step": 3840
+    },
+    {
+      "epoch": 9.253470126735063,
+      "eval_entropy": 0.2301936373114586,
+      "eval_loss": 1.0576049089431763,
+      "eval_mean_token_accuracy": 0.8283417489421502,
+      "eval_num_tokens": 8896053.0,
+      "eval_runtime": 88.7854,
+      "eval_samples_per_second": 15.994,
+      "eval_steps_per_second": 2.005,
+      "step": 3840
+    },
+    {
+      "entropy": 0.09015200780704617,
+      "epoch": 9.301750150875076,
+      "grad_norm": 0.2089058756828308,
+      "learning_rate": 5.6078205077330705e-06,
+      "loss": 0.06626194715499878,
+      "mean_token_accuracy": 0.9747506938874722,
+      "num_tokens": 8940591.0,
+      "step": 3860
+    },
+    {
+      "epoch": 9.301750150875076,
+      "eval_entropy": 0.23024900676159377,
+      "eval_loss": 1.0599786043167114,
+      "eval_mean_token_accuracy": 0.8283051915383071,
+      "eval_num_tokens": 8940591.0,
+      "eval_runtime": 88.2963,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3860
+    },
+    {
+      "entropy": 0.08484094133600592,
+      "epoch": 9.350030175015087,
+      "grad_norm": 0.21646778285503387,
+      "learning_rate": 4.866702233967785e-06,
+      "loss": 0.05998039245605469,
+      "mean_token_accuracy": 0.9771868668496608,
+      "num_tokens": 8989035.0,
+      "step": 3880
+    },
+    {
+      "epoch": 9.350030175015087,
+      "eval_entropy": 0.2300723305578982,
+      "eval_loss": 1.0633587837219238,
+      "eval_mean_token_accuracy": 0.828371500701047,
+      "eval_num_tokens": 8989035.0,
+      "eval_runtime": 88.0233,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 3880
+    },
+    {
+      "entropy": 0.08905019680969417,
+      "epoch": 9.3983101991551,
+      "grad_norm": 0.21842406690120697,
+      "learning_rate": 4.177449088611095e-06,
+      "loss": 0.06666865348815917,
+      "mean_token_accuracy": 0.9753445640206337,
+      "num_tokens": 9033383.0,
+      "step": 3900
+    },
+    {
+      "epoch": 9.3983101991551,
+      "eval_entropy": 0.2295504841911659,
+      "eval_loss": 1.0662461519241333,
+      "eval_mean_token_accuracy": 0.8284122297603093,
+      "eval_num_tokens": 9033383.0,
+      "eval_runtime": 88.5525,
+      "eval_samples_per_second": 16.036,
+      "eval_steps_per_second": 2.01,
+      "step": 3900
+    },
+    {
+      "entropy": 0.08582521872594953,
+      "epoch": 9.446590223295111,
+      "grad_norm": 0.2000240683555603,
+      "learning_rate": 3.540256122150088e-06,
+      "loss": 0.06327944993972778,
+      "mean_token_accuracy": 0.9765432856976985,
+      "num_tokens": 9079672.0,
+      "step": 3920
+    },
+    {
+      "epoch": 9.446590223295111,
+      "eval_entropy": 0.2297678625818049,
+      "eval_loss": 1.0676169395446777,
+      "eval_mean_token_accuracy": 0.8283703879693921,
+      "eval_num_tokens": 9079672.0,
+      "eval_runtime": 87.6897,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 3920
+    },
+    {
+      "entropy": 0.09261773955076932,
+      "epoch": 9.494870247435124,
+      "grad_norm": 0.19104281067848206,
+      "learning_rate": 2.955303652656991e-06,
+      "loss": 0.06111966967582703,
+      "mean_token_accuracy": 0.9738784030079841,
+      "num_tokens": 9127647.0,
+      "step": 3940
+    },
+    {
+      "epoch": 9.494870247435124,
+      "eval_entropy": 0.22986975881490815,
+      "eval_loss": 1.0687925815582275,
+      "eval_mean_token_accuracy": 0.8282996858103892,
+      "eval_num_tokens": 9127647.0,
+      "eval_runtime": 88.2224,
+      "eval_samples_per_second": 16.096,
+      "eval_steps_per_second": 2.018,
+      "step": 3940
+    },
+    {
+      "entropy": 0.08347254800610245,
+      "epoch": 9.543150271575136,
+      "grad_norm": 0.18671369552612305,
+      "learning_rate": 2.422757214761356e-06,
+      "loss": 0.06384263038635254,
+      "mean_token_accuracy": 0.9770128890872002,
+      "num_tokens": 9172587.0,
+      "step": 3960
+    },
+    {
+      "epoch": 9.543150271575136,
+      "eval_entropy": 0.22975517851248217,
+      "eval_loss": 1.0701013803482056,
+      "eval_mean_token_accuracy": 0.8282244285840666,
+      "eval_num_tokens": 9172587.0,
+      "eval_runtime": 88.5794,
+      "eval_samples_per_second": 16.031,
+      "eval_steps_per_second": 2.009,
+      "step": 3960
+    },
+    {
+      "entropy": 0.08136763009242713,
+      "epoch": 9.591430295715147,
+      "grad_norm": 0.24811454117298126,
+      "learning_rate": 1.942767512805673e-06,
+      "loss": 0.06044899821281433,
+      "mean_token_accuracy": 0.9776685133576393,
+      "num_tokens": 9219278.0,
+      "step": 3980
+    },
+    {
+      "epoch": 9.591430295715147,
+      "eval_entropy": 0.22938149347064202,
+      "eval_loss": 1.0711463689804077,
+      "eval_mean_token_accuracy": 0.8282721410976367,
+      "eval_num_tokens": 9219278.0,
+      "eval_runtime": 88.4978,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 3980
+    },
+    {
+      "entropy": 0.09432941288687288,
+      "epoch": 9.63971031985516,
+      "grad_norm": 0.30733999609947205,
+      "learning_rate": 1.515470378197889e-06,
+      "loss": 0.0672213613986969,
+      "mean_token_accuracy": 0.9728612303733826,
+      "num_tokens": 9262346.0,
+      "step": 4000
+    },
+    {
+      "epoch": 9.63971031985516,
+      "eval_entropy": 0.22936245958121976,
+      "eval_loss": 1.071343183517456,
+      "eval_mean_token_accuracy": 0.8282827025049189,
+      "eval_num_tokens": 9262346.0,
+      "eval_runtime": 88.4428,
+      "eval_samples_per_second": 16.056,
+      "eval_steps_per_second": 2.013,
+      "step": 4000
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.915846957676462e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4020/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4020/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4020/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4020/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4020/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4020/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4020/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4020/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4020/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4020/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4020/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4020/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4020/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4020/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..a4f4b226a35650c6e7d300dc20df275ce95dc778
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4020/trainer_state.json
@@ -0,0 +1,4255 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 9.687990343995171,
+  "eval_steps": 20,
+  "global_step": 4020,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    },
+    {
+      "entropy": 0.1326121488586068,
+      "epoch": 7.181050090525045,
+      "grad_norm": 0.4843117296695709,
+      "learning_rate": 8.411870386507724e-05,
+      "loss": 0.09537842869758606,
+      "mean_token_accuracy": 0.9645151920616627,
+      "num_tokens": 6898444.0,
+      "step": 2980
+    },
+    {
+      "epoch": 7.181050090525045,
+      "eval_entropy": 0.27999477783280813,
+      "eval_loss": 0.9031037092208862,
+      "eval_mean_token_accuracy": 0.8261882204018282,
+      "eval_num_tokens": 6898444.0,
+      "eval_runtime": 88.5602,
+      "eval_samples_per_second": 16.034,
+      "eval_steps_per_second": 2.01,
+      "step": 2980
+    },
+    {
+      "entropy": 0.1156751093454659,
+      "epoch": 7.229330114665057,
+      "grad_norm": 0.5297922492027283,
+      "learning_rate": 8.149629351430326e-05,
+      "loss": 0.09203402996063233,
+      "mean_token_accuracy": 0.9689733698964119,
+      "num_tokens": 6947106.0,
+      "step": 3000
+    },
+    {
+      "epoch": 7.229330114665057,
+      "eval_entropy": 0.2740225845508361,
+      "eval_loss": 0.8876440525054932,
+      "eval_mean_token_accuracy": 0.8265116857678703,
+      "eval_num_tokens": 6947106.0,
+      "eval_runtime": 87.7971,
+      "eval_samples_per_second": 16.174,
+      "eval_steps_per_second": 2.027,
+      "step": 3000
+    },
+    {
+      "entropy": 0.12116114404052496,
+      "epoch": 7.27761013880507,
+      "grad_norm": 0.5467398166656494,
+      "learning_rate": 7.890406302404618e-05,
+      "loss": 0.0933056652545929,
+      "mean_token_accuracy": 0.9676455594599247,
+      "num_tokens": 6992784.0,
+      "step": 3020
+    },
+    {
+      "epoch": 7.27761013880507,
+      "eval_entropy": 0.27238342793804876,
+      "eval_loss": 0.9166117310523987,
+      "eval_mean_token_accuracy": 0.825499015242866,
+      "eval_num_tokens": 6992784.0,
+      "eval_runtime": 88.3276,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 3020
+    },
+    {
+      "entropy": 0.11715119928121567,
+      "epoch": 7.325890162945082,
+      "grad_norm": 0.4976702332496643,
+      "learning_rate": 7.634274596487176e-05,
+      "loss": 0.09449659585952759,
+      "mean_token_accuracy": 0.966506478190422,
+      "num_tokens": 7040482.0,
+      "step": 3040
+    },
+    {
+      "epoch": 7.325890162945082,
+      "eval_entropy": 0.2788975556077582,
+      "eval_loss": 0.8833065629005432,
+      "eval_mean_token_accuracy": 0.8269920596915684,
+      "eval_num_tokens": 7040482.0,
+      "eval_runtime": 88.9016,
+      "eval_samples_per_second": 15.973,
+      "eval_steps_per_second": 2.002,
+      "step": 3040
+    },
+    {
+      "entropy": 0.12565911412239075,
+      "epoch": 7.3741701870850935,
+      "grad_norm": 0.5728679895401001,
+      "learning_rate": 7.381306715921025e-05,
+      "loss": 0.09626244902610778,
+      "mean_token_accuracy": 0.9661444149911403,
+      "num_tokens": 7086261.0,
+      "step": 3060
+    },
+    {
+      "epoch": 7.3741701870850935,
+      "eval_entropy": 0.2762492011939542,
+      "eval_loss": 0.8961160778999329,
+      "eval_mean_token_accuracy": 0.826369910092836,
+      "eval_num_tokens": 7086261.0,
+      "eval_runtime": 88.5204,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 3060
+    },
+    {
+      "entropy": 0.12258567921817302,
+      "epoch": 7.422450211225105,
+      "grad_norm": 0.468985915184021,
+      "learning_rate": 7.131574247624031e-05,
+      "loss": 0.09120266437530518,
+      "mean_token_accuracy": 0.9674938447773457,
+      "num_tokens": 7133247.0,
+      "step": 3080
+    },
+    {
+      "epoch": 7.422450211225105,
+      "eval_entropy": 0.2688512086366,
+      "eval_loss": 0.9048654437065125,
+      "eval_mean_token_accuracy": 0.8265582302982888,
+      "eval_num_tokens": 7133247.0,
+      "eval_runtime": 88.164,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 3080
+    },
+    {
+      "entropy": 0.12070536343380808,
+      "epoch": 7.470730235365117,
+      "grad_norm": 0.5310841798782349,
+      "learning_rate": 6.885147862930688e-05,
+      "loss": 0.09386247992515565,
+      "mean_token_accuracy": 0.96664222702384,
+      "num_tokens": 7177810.0,
+      "step": 3100
+    },
+    {
+      "epoch": 7.470730235365117,
+      "eval_entropy": 0.26289632874593305,
+      "eval_loss": 0.902837336063385,
+      "eval_mean_token_accuracy": 0.828052184554968,
+      "eval_num_tokens": 7177810.0,
+      "eval_runtime": 88.1201,
+      "eval_samples_per_second": 16.114,
+      "eval_steps_per_second": 2.02,
+      "step": 3100
+    },
+    {
+      "entropy": 0.11322665181942285,
+      "epoch": 7.51901025950513,
+      "grad_norm": 0.6846067905426025,
+      "learning_rate": 6.642097297592908e-05,
+      "loss": 0.08772538304328918,
+      "mean_token_accuracy": 0.9691362135112286,
+      "num_tokens": 7225302.0,
+      "step": 3120
+    },
+    {
+      "epoch": 7.51901025950513,
+      "eval_entropy": 0.2670605613776807,
+      "eval_loss": 0.9130566120147705,
+      "eval_mean_token_accuracy": 0.826366903741708,
+      "eval_num_tokens": 7225302.0,
+      "eval_runtime": 88.6035,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3120
+    },
+    {
+      "entropy": 0.10985926580615342,
+      "epoch": 7.567290283645142,
+      "grad_norm": 0.4799748361110687,
+      "learning_rate": 6.402491332045683e-05,
+      "loss": 0.08672296404838561,
+      "mean_token_accuracy": 0.9698824048042297,
+      "num_tokens": 7274881.0,
+      "step": 3140
+    },
+    {
+      "epoch": 7.567290283645142,
+      "eval_entropy": 0.26657010866015146,
+      "eval_loss": 0.9056380987167358,
+      "eval_mean_token_accuracy": 0.8272326461384806,
+      "eval_num_tokens": 7274881.0,
+      "eval_runtime": 88.2967,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3140
+    },
+    {
+      "entropy": 0.11890473761595785,
+      "epoch": 7.615570307785154,
+      "grad_norm": 0.4081912934780121,
+      "learning_rate": 6.166397771942987e-05,
+      "loss": 0.09396827220916748,
+      "mean_token_accuracy": 0.967512384802103,
+      "num_tokens": 7321163.0,
+      "step": 3160
+    },
+    {
+      "epoch": 7.615570307785154,
+      "eval_entropy": 0.26364776575833226,
+      "eval_loss": 0.9204733967781067,
+      "eval_mean_token_accuracy": 0.827419729714983,
+      "eval_num_tokens": 7321163.0,
+      "eval_runtime": 88.3441,
+      "eval_samples_per_second": 16.074,
+      "eval_steps_per_second": 2.015,
+      "step": 3160
+    },
+    {
+      "entropy": 0.12209134344011545,
+      "epoch": 7.663850331925166,
+      "grad_norm": 0.5159774422645569,
+      "learning_rate": 5.93388342896962e-05,
+      "loss": 0.09292978644371033,
+      "mean_token_accuracy": 0.9658421002328396,
+      "num_tokens": 7366217.0,
+      "step": 3180
+    },
+    {
+      "epoch": 7.663850331925166,
+      "eval_entropy": 0.2714110068725736,
+      "eval_loss": 0.9143092036247253,
+      "eval_mean_token_accuracy": 0.8262251099843657,
+      "eval_num_tokens": 7366217.0,
+      "eval_runtime": 87.9635,
+      "eval_samples_per_second": 16.143,
+      "eval_steps_per_second": 2.024,
+      "step": 3180
+    },
+    {
+      "entropy": 0.11955609126016498,
+      "epoch": 7.712130356065178,
+      "grad_norm": 0.7005976438522339,
+      "learning_rate": 5.705014101934252e-05,
+      "loss": 0.09319776892662049,
+      "mean_token_accuracy": 0.9674513377249241,
+      "num_tokens": 7410887.0,
+      "step": 3200
+    },
+    {
+      "epoch": 7.712130356065178,
+      "eval_entropy": 0.2626291513442993,
+      "eval_loss": 0.91884446144104,
+      "eval_mean_token_accuracy": 0.827563955207889,
+      "eval_num_tokens": 7410887.0,
+      "eval_runtime": 87.8849,
+      "eval_samples_per_second": 16.157,
+      "eval_steps_per_second": 2.025,
+      "step": 3200
+    },
+    {
+      "entropy": 0.11767532899975777,
+      "epoch": 7.76041038020519,
+      "grad_norm": 0.4675362706184387,
+      "learning_rate": 5.479854558149187e-05,
+      "loss": 0.09038087725639343,
+      "mean_token_accuracy": 0.9669680051505566,
+      "num_tokens": 7457578.0,
+      "step": 3220
+    },
+    {
+      "epoch": 7.76041038020519,
+      "eval_entropy": 0.26581037203582486,
+      "eval_loss": 0.9146883487701416,
+      "eval_mean_token_accuracy": 0.8270345812433222,
+      "eval_num_tokens": 7457578.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 3220
+    },
+    {
+      "entropy": 0.11383061073720455,
+      "epoch": 7.808690404345202,
+      "grad_norm": 0.541247546672821,
+      "learning_rate": 5.258468515101918e-05,
+      "loss": 0.08942405581474304,
+      "mean_token_accuracy": 0.968398705124855,
+      "num_tokens": 7505761.0,
+      "step": 3240
+    },
+    {
+      "epoch": 7.808690404345202,
+      "eval_entropy": 0.26687998960861997,
+      "eval_loss": 0.9037085175514221,
+      "eval_mean_token_accuracy": 0.8274977863504646,
+      "eval_num_tokens": 7505761.0,
+      "eval_runtime": 88.5163,
+      "eval_samples_per_second": 16.042,
+      "eval_steps_per_second": 2.011,
+      "step": 3240
+    },
+    {
+      "entropy": 0.12004058873280883,
+      "epoch": 7.856970428485214,
+      "grad_norm": 0.5289552211761475,
+      "learning_rate": 5.040918622423899e-05,
+      "loss": 0.09064805507659912,
+      "mean_token_accuracy": 0.9673330299556255,
+      "num_tokens": 7552557.0,
+      "step": 3260
+    },
+    {
+      "epoch": 7.856970428485214,
+      "eval_entropy": 0.26288379251622085,
+      "eval_loss": 0.9146178364753723,
+      "eval_mean_token_accuracy": 0.8276662448149049,
+      "eval_num_tokens": 7552557.0,
+      "eval_runtime": 88.0723,
+      "eval_samples_per_second": 16.123,
+      "eval_steps_per_second": 2.021,
+      "step": 3260
+    },
+    {
+      "entropy": 0.1180689456872642,
+      "epoch": 7.905250452625227,
+      "grad_norm": 0.4754861891269684,
+      "learning_rate": 4.827266444161377e-05,
+      "loss": 0.09402655363082886,
+      "mean_token_accuracy": 0.9663361787796021,
+      "num_tokens": 7596998.0,
+      "step": 3280
+    },
+    {
+      "epoch": 7.905250452625227,
+      "eval_entropy": 0.261690996922134,
+      "eval_loss": 0.9120186567306519,
+      "eval_mean_token_accuracy": 0.8278995015647974,
+      "eval_num_tokens": 7596998.0,
+      "eval_runtime": 88.6573,
+      "eval_samples_per_second": 16.017,
+      "eval_steps_per_second": 2.008,
+      "step": 3280
+    },
+    {
+      "entropy": 0.1162164019420743,
+      "epoch": 7.953530476765239,
+      "grad_norm": 0.6203398108482361,
+      "learning_rate": 4.61757244135355e-05,
+      "loss": 0.08832768201828003,
+      "mean_token_accuracy": 0.9694220684468746,
+      "num_tokens": 7643288.0,
+      "step": 3300
+    },
+    {
+      "epoch": 7.953530476765239,
+      "eval_entropy": 0.26158678129817664,
+      "eval_loss": 0.9301361441612244,
+      "eval_mean_token_accuracy": 0.8269603885291668,
+      "eval_num_tokens": 7643288.0,
+      "eval_runtime": 88.7064,
+      "eval_samples_per_second": 16.008,
+      "eval_steps_per_second": 2.007,
+      "step": 3300
+    },
+    {
+      "entropy": 0.10877202204489089,
+      "epoch": 8.0,
+      "grad_norm": 3.058673143386841,
+      "learning_rate": 4.411895954922763e-05,
+      "loss": 0.09159615635871887,
+      "mean_token_accuracy": 0.9686878834451947,
+      "num_tokens": 7688608.0,
+      "step": 3320
+    },
+    {
+      "epoch": 8.0,
+      "eval_entropy": 0.2590438057197614,
+      "eval_loss": 0.9262065887451172,
+      "eval_mean_token_accuracy": 0.8277404546067956,
+      "eval_num_tokens": 7688608.0,
+      "eval_runtime": 87.7557,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 3320
+    },
+    {
+      "entropy": 0.09969092034734786,
+      "epoch": 8.048280024140013,
+      "grad_norm": 0.2561911642551422,
+      "learning_rate": 4.2102951888817644e-05,
+      "loss": 0.07228735685348511,
+      "mean_token_accuracy": 0.9743556551635265,
+      "num_tokens": 7732856.0,
+      "step": 3340
+    },
+    {
+      "epoch": 8.048280024140013,
+      "eval_entropy": 0.2462534454933713,
+      "eval_loss": 0.9755983352661133,
+      "eval_mean_token_accuracy": 0.827510209565752,
+      "eval_num_tokens": 7732856.0,
+      "eval_runtime": 88.0297,
+      "eval_samples_per_second": 16.131,
+      "eval_steps_per_second": 2.022,
+      "step": 3340
+    },
+    {
+      "entropy": 0.09477624730207027,
+      "epoch": 8.096560048280024,
+      "grad_norm": 0.3670356571674347,
+      "learning_rate": 4.012827193862633e-05,
+      "loss": 0.0686036467552185,
+      "mean_token_accuracy": 0.9743670664727688,
+      "num_tokens": 7778585.0,
+      "step": 3360
+    },
+    {
+      "epoch": 8.096560048280024,
+      "eval_entropy": 0.2440882951356052,
+      "eval_loss": 1.0118423700332642,
+      "eval_mean_token_accuracy": 0.8271241070849172,
+      "eval_num_tokens": 7778585.0,
+      "eval_runtime": 88.1364,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 3360
+    },
+    {
+      "entropy": 0.09417980071157217,
+      "epoch": 8.144840072420036,
+      "grad_norm": 0.5785111784934998,
+      "learning_rate": 3.819547850972133e-05,
+      "loss": 0.07100698947906495,
+      "mean_token_accuracy": 0.9737804099917412,
+      "num_tokens": 7824287.0,
+      "step": 3380
+    },
+    {
+      "epoch": 8.144840072420036,
+      "eval_entropy": 0.24603172860453637,
+      "eval_loss": 1.0021991729736328,
+      "eval_mean_token_accuracy": 0.827162122793412,
+      "eval_num_tokens": 7824287.0,
+      "eval_runtime": 88.4534,
+      "eval_samples_per_second": 16.054,
+      "eval_steps_per_second": 2.012,
+      "step": 3380
+    },
+    {
+      "entropy": 0.087222247896716,
+      "epoch": 8.193120096560047,
+      "grad_norm": 0.44848504662513733,
+      "learning_rate": 3.6305118559780393e-05,
+      "loss": 0.06325234174728393,
+      "mean_token_accuracy": 0.9769335232675076,
+      "num_tokens": 7874784.0,
+      "step": 3400
+    },
+    {
+      "epoch": 8.193120096560047,
+      "eval_entropy": 0.24544863905129807,
+      "eval_loss": 1.005008339881897,
+      "eval_mean_token_accuracy": 0.8273035240976998,
+      "eval_num_tokens": 7874784.0,
+      "eval_runtime": 88.6737,
+      "eval_samples_per_second": 16.014,
+      "eval_steps_per_second": 2.007,
+      "step": 3400
+    },
+    {
+      "entropy": 0.09037275062873959,
+      "epoch": 8.24140012070006,
+      "grad_norm": 0.23529155552387238,
+      "learning_rate": 3.4457727038308435e-05,
+      "loss": 0.06589569449424744,
+      "mean_token_accuracy": 0.9751470290124417,
+      "num_tokens": 7925240.0,
+      "step": 3420
+    },
+    {
+      "epoch": 8.24140012070006,
+      "eval_entropy": 0.24203423494368456,
+      "eval_loss": 1.0052824020385742,
+      "eval_mean_token_accuracy": 0.8275575503874361,
+      "eval_num_tokens": 7925240.0,
+      "eval_runtime": 88.3486,
+      "eval_samples_per_second": 16.073,
+      "eval_steps_per_second": 2.015,
+      "step": 3420
+    },
+    {
+      "entropy": 0.10373208015225828,
+      "epoch": 8.289680144840073,
+      "grad_norm": 0.24796408414840698,
+      "learning_rate": 3.2653826735253594e-05,
+      "loss": 0.07714771628379821,
+      "mean_token_accuracy": 0.9722891598939896,
+      "num_tokens": 7966685.0,
+      "step": 3440
+    },
+    {
+      "epoch": 8.289680144840073,
+      "eval_entropy": 0.23461157050025597,
+      "eval_loss": 1.012108325958252,
+      "eval_mean_token_accuracy": 0.82825900564033,
+      "eval_num_tokens": 7966685.0,
+      "eval_runtime": 88.1018,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 3440
+    },
+    {
+      "entropy": 0.08967954274266958,
+      "epoch": 8.337960168980084,
+      "grad_norm": 0.5314674973487854,
+      "learning_rate": 3.08939281330635e-05,
+      "loss": 0.06746266484260559,
+      "mean_token_accuracy": 0.9753133699297905,
+      "num_tokens": 8013656.0,
+      "step": 3460
+    },
+    {
+      "epoch": 8.337960168980084,
+      "eval_entropy": 0.23781998589467468,
+      "eval_loss": 1.0119913816452026,
+      "eval_mean_token_accuracy": 0.8273287157664139,
+      "eval_num_tokens": 8013656.0,
+      "eval_runtime": 88.6038,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3460
+    },
+    {
+      "entropy": 0.09280927330255509,
+      "epoch": 8.386240193120097,
+      "grad_norm": 0.30658936500549316,
+      "learning_rate": 2.9178529262224946e-05,
+      "loss": 0.06936421990394592,
+      "mean_token_accuracy": 0.9742105580866337,
+      "num_tokens": 8061108.0,
+      "step": 3480
+    },
+    {
+      "epoch": 8.386240193120097,
+      "eval_entropy": 0.23672426734747512,
+      "eval_loss": 1.0162041187286377,
+      "eval_mean_token_accuracy": 0.8278394281194451,
+      "eval_num_tokens": 8061108.0,
+      "eval_runtime": 88.6015,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 3480
+    },
+    {
+      "entropy": 0.09195122267119586,
+      "epoch": 8.43452021726011,
+      "grad_norm": 0.24512454867362976,
+      "learning_rate": 2.7508115560327256e-05,
+      "loss": 0.06996064782142639,
+      "mean_token_accuracy": 0.974104854464531,
+      "num_tokens": 8106924.0,
+      "step": 3500
+    },
+    {
+      "epoch": 8.43452021726011,
+      "eval_entropy": 0.23758245828781235,
+      "eval_loss": 1.0157747268676758,
+      "eval_mean_token_accuracy": 0.8278501579600773,
+      "eval_num_tokens": 8106924.0,
+      "eval_runtime": 88.5678,
+      "eval_samples_per_second": 16.033,
+      "eval_steps_per_second": 2.01,
+      "step": 3500
+    },
+    {
+      "entropy": 0.09518477590754629,
+      "epoch": 8.48280024140012,
+      "grad_norm": 0.2373228371143341,
+      "learning_rate": 2.5883159734688868e-05,
+      "loss": 0.06832797527313232,
+      "mean_token_accuracy": 0.9743806682527065,
+      "num_tokens": 8153070.0,
+      "step": 3520
+    },
+    {
+      "epoch": 8.48280024140012,
+      "eval_entropy": 0.2379308074210467,
+      "eval_loss": 1.0180530548095703,
+      "eval_mean_token_accuracy": 0.8282477052024241,
+      "eval_num_tokens": 8153070.0,
+      "eval_runtime": 88.4651,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 3520
+    },
+    {
+      "entropy": 0.0906269280705601,
+      "epoch": 8.531080265540133,
+      "grad_norm": 0.37602657079696655,
+      "learning_rate": 2.4304121628587173e-05,
+      "loss": 0.06921588778495788,
+      "mean_token_accuracy": 0.9745571479201317,
+      "num_tokens": 8200066.0,
+      "step": 3540
+    },
+    {
+      "epoch": 8.531080265540133,
+      "eval_entropy": 0.2357490769263064,
+      "eval_loss": 1.017139196395874,
+      "eval_mean_token_accuracy": 0.828713178299786,
+      "eval_num_tokens": 8200066.0,
+      "eval_runtime": 88.6424,
+      "eval_samples_per_second": 16.019,
+      "eval_steps_per_second": 2.008,
+      "step": 3540
+    },
+    {
+      "entropy": 0.09273334196768701,
+      "epoch": 8.579360289680144,
+      "grad_norm": 0.2978838086128235,
+      "learning_rate": 2.277144809112794e-05,
+      "loss": 0.06970887780189514,
+      "mean_token_accuracy": 0.9751374177634716,
+      "num_tokens": 8246157.0,
+      "step": 3560
+    },
+    {
+      "epoch": 8.579360289680144,
+      "eval_entropy": 0.23590933138065123,
+      "eval_loss": 1.019231915473938,
+      "eval_mean_token_accuracy": 0.8279163412163767,
+      "eval_num_tokens": 8246157.0,
+      "eval_runtime": 88.2691,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 3560
+    },
+    {
+      "entropy": 0.09854201921261847,
+      "epoch": 8.627640313820157,
+      "grad_norm": 0.5381243228912354,
+      "learning_rate": 2.1285572850792413e-05,
+      "loss": 0.07123501300811767,
+      "mean_token_accuracy": 0.9723988175392151,
+      "num_tokens": 8291752.0,
+      "step": 3580
+    },
+    {
+      "epoch": 8.627640313820157,
+      "eval_entropy": 0.23592024949494372,
+      "eval_loss": 1.020250678062439,
+      "eval_mean_token_accuracy": 0.8282466226079491,
+      "eval_num_tokens": 8291752.0,
+      "eval_runtime": 87.8939,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 3580
+    },
+    {
+      "entropy": 0.09092000788077712,
+      "epoch": 8.67592033796017,
+      "grad_norm": 0.29089561104774475,
+      "learning_rate": 1.9846916392697493e-05,
+      "loss": 0.06982720494270325,
+      "mean_token_accuracy": 0.9749995484948158,
+      "num_tokens": 8337651.0,
+      "step": 3600
+    },
+    {
+      "epoch": 8.67592033796017,
+      "eval_entropy": 0.23658913382318583,
+      "eval_loss": 1.019862413406372,
+      "eval_mean_token_accuracy": 0.8280127842104836,
+      "eval_num_tokens": 8337651.0,
+      "eval_runtime": 88.2855,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 3600
+    },
+    {
+      "entropy": 0.08708424470387399,
+      "epoch": 8.72420036210018,
+      "grad_norm": 0.29904839396476746,
+      "learning_rate": 1.8455885839603106e-05,
+      "loss": 0.06801370978355407,
+      "mean_token_accuracy": 0.9755801670253277,
+      "num_tokens": 8385330.0,
+      "step": 3620
+    },
+    {
+      "epoch": 8.72420036210018,
+      "eval_entropy": 0.23627035771862845,
+      "eval_loss": 1.0204569101333618,
+      "eval_mean_token_accuracy": 0.8279967894045155,
+      "eval_num_tokens": 8385330.0,
+      "eval_runtime": 88.1753,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 3620
+    },
+    {
+      "entropy": 0.0916255566291511,
+      "epoch": 8.772480386240193,
+      "grad_norm": 0.20571033656597137,
+      "learning_rate": 1.7112874836701502e-05,
+      "loss": 0.0692249894142151,
+      "mean_token_accuracy": 0.9745148062705994,
+      "num_tokens": 8432134.0,
+      "step": 3640
+    },
+    {
+      "epoch": 8.772480386240193,
+      "eval_entropy": 0.2340821772813797,
+      "eval_loss": 1.0212821960449219,
+      "eval_mean_token_accuracy": 0.828373622358515,
+      "eval_num_tokens": 8432134.0,
+      "eval_runtime": 87.2518,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 3640
+    },
+    {
+      "entropy": 0.0926868049427867,
+      "epoch": 8.820760410380204,
+      "grad_norm": 0.2618265450000763,
+      "learning_rate": 1.5818263440220203e-05,
+      "loss": 0.07338647246360779,
+      "mean_token_accuracy": 0.9724708944559097,
+      "num_tokens": 8476758.0,
+      "step": 3660
+    },
+    {
+      "epoch": 8.820760410380204,
+      "eval_entropy": 0.23637074575330433,
+      "eval_loss": 1.0186725854873657,
+      "eval_mean_token_accuracy": 0.8277707605549459,
+      "eval_num_tokens": 8476758.0,
+      "eval_runtime": 88.4695,
+      "eval_samples_per_second": 16.051,
+      "eval_steps_per_second": 2.012,
+      "step": 3660
+    },
+    {
+      "entropy": 0.09175782897509635,
+      "epoch": 8.869040434520217,
+      "grad_norm": 0.2889852821826935,
+      "learning_rate": 1.4572418009870604e-05,
+      "loss": 0.07066041231155396,
+      "mean_token_accuracy": 0.9725673109292984,
+      "num_tokens": 8523294.0,
+      "step": 3680
+    },
+    {
+      "epoch": 8.869040434520217,
+      "eval_entropy": 0.23945164404223473,
+      "eval_loss": 1.0145231485366821,
+      "eval_mean_token_accuracy": 0.8281202523895864,
+      "eval_num_tokens": 8523294.0,
+      "eval_runtime": 88.368,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 3680
+    },
+    {
+      "entropy": 0.09265611860901117,
+      "epoch": 8.91732045866023,
+      "grad_norm": 0.41092950105667114,
+      "learning_rate": 1.3375691105172625e-05,
+      "loss": 0.06975027322769164,
+      "mean_token_accuracy": 0.974636297672987,
+      "num_tokens": 8570358.0,
+      "step": 3700
+    },
+    {
+      "epoch": 8.91732045866023,
+      "eval_entropy": 0.239588919901446,
+      "eval_loss": 1.014059066772461,
+      "eval_mean_token_accuracy": 0.828028041660116,
+      "eval_num_tokens": 8570358.0,
+      "eval_runtime": 88.1039,
+      "eval_samples_per_second": 16.117,
+      "eval_steps_per_second": 2.02,
+      "step": 3700
+    },
+    {
+      "entropy": 0.0896465003490448,
+      "epoch": 8.965600482800241,
+      "grad_norm": 0.24000254273414612,
+      "learning_rate": 1.2228421385684305e-05,
+      "loss": 0.06743953824043274,
+      "mean_token_accuracy": 0.974006575345993,
+      "num_tokens": 8618246.0,
+      "step": 3720
+    },
+    {
+      "epoch": 8.965600482800241,
+      "eval_entropy": 0.2390461369846644,
+      "eval_loss": 1.011997938156128,
+      "eval_mean_token_accuracy": 0.8284533211354459,
+      "eval_num_tokens": 8618246.0,
+      "eval_runtime": 88.6202,
+      "eval_samples_per_second": 16.023,
+      "eval_steps_per_second": 2.009,
+      "step": 3720
+    },
+    {
+      "entropy": 0.09575109251520851,
+      "epoch": 9.012070006035003,
+      "grad_norm": 0.18088576197624207,
+      "learning_rate": 1.1130933515165429e-05,
+      "loss": 0.07295305132865906,
+      "mean_token_accuracy": 0.9739722857227573,
+      "num_tokens": 8661177.0,
+      "step": 3740
+    },
+    {
+      "epoch": 9.012070006035003,
+      "eval_entropy": 0.23728882387447892,
+      "eval_loss": 1.0134038925170898,
+      "eval_mean_token_accuracy": 0.8286508951294288,
+      "eval_num_tokens": 8661177.0,
+      "eval_runtime": 88.6434,
+      "eval_samples_per_second": 16.019,
+      "eval_steps_per_second": 2.008,
+      "step": 3740
+    },
+    {
+      "entropy": 0.08883201805874705,
+      "epoch": 9.060350030175016,
+      "grad_norm": 0.23120898008346558,
+      "learning_rate": 1.008353806970132e-05,
+      "loss": 0.06326142549514771,
+      "mean_token_accuracy": 0.9780020996928215,
+      "num_tokens": 8707602.0,
+      "step": 3760
+    },
+    {
+      "epoch": 9.060350030175016,
+      "eval_entropy": 0.23416905331142832,
+      "eval_loss": 1.0284100770950317,
+      "eval_mean_token_accuracy": 0.8284083213029283,
+      "eval_num_tokens": 8707602.0,
+      "eval_runtime": 88.2933,
+      "eval_samples_per_second": 16.083,
+      "eval_steps_per_second": 2.016,
+      "step": 3760
+    },
+    {
+      "entropy": 0.08861468276008963,
+      "epoch": 9.108630054315027,
+      "grad_norm": 0.17277656495571136,
+      "learning_rate": 9.086531449813489e-06,
+      "loss": 0.06727416515350342,
+      "mean_token_accuracy": 0.9760390736162663,
+      "num_tokens": 8751287.0,
+      "step": 3780
+    },
+    {
+      "epoch": 9.108630054315027,
+      "eval_entropy": 0.23370317386442355,
+      "eval_loss": 1.0381091833114624,
+      "eval_mean_token_accuracy": 0.8282668178670862,
+      "eval_num_tokens": 8751287.0,
+      "eval_runtime": 88.5738,
+      "eval_samples_per_second": 16.032,
+      "eval_steps_per_second": 2.01,
+      "step": 3780
+    },
+    {
+      "entropy": 0.09032954541034996,
+      "epoch": 9.15691007845504,
+      "grad_norm": 0.19481085240840912,
+      "learning_rate": 8.140195796581981e-06,
+      "loss": 0.06117786169052124,
+      "mean_token_accuracy": 0.9764454193413258,
+      "num_tokens": 8799982.0,
+      "step": 3800
+    },
+    {
+      "epoch": 9.15691007845504,
+      "eval_entropy": 0.23215936609868254,
+      "eval_loss": 1.0467767715454102,
+      "eval_mean_token_accuracy": 0.828405868806196,
+      "eval_num_tokens": 8799982.0,
+      "eval_runtime": 87.9551,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 3800
+    },
+    {
+      "entropy": 0.08824449330568314,
+      "epoch": 9.20519010259505,
+      "grad_norm": 0.20382793247699738,
+      "learning_rate": 7.2447989118025455e-06,
+      "loss": 0.06422553658485412,
+      "mean_token_accuracy": 0.9763121999800205,
+      "num_tokens": 8845536.0,
+      "step": 3820
+    },
+    {
+      "epoch": 9.20519010259505,
+      "eval_entropy": 0.23169661472352704,
+      "eval_loss": 1.0529043674468994,
+      "eval_mean_token_accuracy": 0.8282025610463003,
+      "eval_num_tokens": 8845536.0,
+      "eval_runtime": 88.5548,
+      "eval_samples_per_second": 16.035,
+      "eval_steps_per_second": 2.01,
+      "step": 3820
+    },
+    {
+      "entropy": 0.07768499720841646,
+      "epoch": 9.253470126735063,
+      "grad_norm": 0.11727919429540634,
+      "learning_rate": 6.400594182202084e-06,
+      "loss": 0.057279336452484134,
+      "mean_token_accuracy": 0.9795166745781898,
+      "num_tokens": 8896053.0,
+      "step": 3840
+    },
+    {
+      "epoch": 9.253470126735063,
+      "eval_entropy": 0.2301936373114586,
+      "eval_loss": 1.0576049089431763,
+      "eval_mean_token_accuracy": 0.8283417489421502,
+      "eval_num_tokens": 8896053.0,
+      "eval_runtime": 88.7854,
+      "eval_samples_per_second": 15.994,
+      "eval_steps_per_second": 2.005,
+      "step": 3840
+    },
+    {
+      "entropy": 0.09015200780704617,
+      "epoch": 9.301750150875076,
+      "grad_norm": 0.2089058756828308,
+      "learning_rate": 5.6078205077330705e-06,
+      "loss": 0.06626194715499878,
+      "mean_token_accuracy": 0.9747506938874722,
+      "num_tokens": 8940591.0,
+      "step": 3860
+    },
+    {
+      "epoch": 9.301750150875076,
+      "eval_entropy": 0.23024900676159377,
+      "eval_loss": 1.0599786043167114,
+      "eval_mean_token_accuracy": 0.8283051915383071,
+      "eval_num_tokens": 8940591.0,
+      "eval_runtime": 88.2963,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3860
+    },
+    {
+      "entropy": 0.08484094133600592,
+      "epoch": 9.350030175015087,
+      "grad_norm": 0.21646778285503387,
+      "learning_rate": 4.866702233967785e-06,
+      "loss": 0.05998039245605469,
+      "mean_token_accuracy": 0.9771868668496608,
+      "num_tokens": 8989035.0,
+      "step": 3880
+    },
+    {
+      "epoch": 9.350030175015087,
+      "eval_entropy": 0.2300723305578982,
+      "eval_loss": 1.0633587837219238,
+      "eval_mean_token_accuracy": 0.828371500701047,
+      "eval_num_tokens": 8989035.0,
+      "eval_runtime": 88.0233,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 3880
+    },
+    {
+      "entropy": 0.08905019680969417,
+      "epoch": 9.3983101991551,
+      "grad_norm": 0.21842406690120697,
+      "learning_rate": 4.177449088611095e-06,
+      "loss": 0.06666865348815917,
+      "mean_token_accuracy": 0.9753445640206337,
+      "num_tokens": 9033383.0,
+      "step": 3900
+    },
+    {
+      "epoch": 9.3983101991551,
+      "eval_entropy": 0.2295504841911659,
+      "eval_loss": 1.0662461519241333,
+      "eval_mean_token_accuracy": 0.8284122297603093,
+      "eval_num_tokens": 9033383.0,
+      "eval_runtime": 88.5525,
+      "eval_samples_per_second": 16.036,
+      "eval_steps_per_second": 2.01,
+      "step": 3900
+    },
+    {
+      "entropy": 0.08582521872594953,
+      "epoch": 9.446590223295111,
+      "grad_norm": 0.2000240683555603,
+      "learning_rate": 3.540256122150088e-06,
+      "loss": 0.06327944993972778,
+      "mean_token_accuracy": 0.9765432856976985,
+      "num_tokens": 9079672.0,
+      "step": 3920
+    },
+    {
+      "epoch": 9.446590223295111,
+      "eval_entropy": 0.2297678625818049,
+      "eval_loss": 1.0676169395446777,
+      "eval_mean_token_accuracy": 0.8283703879693921,
+      "eval_num_tokens": 9079672.0,
+      "eval_runtime": 87.6897,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 3920
+    },
+    {
+      "entropy": 0.09261773955076932,
+      "epoch": 9.494870247435124,
+      "grad_norm": 0.19104281067848206,
+      "learning_rate": 2.955303652656991e-06,
+      "loss": 0.06111966967582703,
+      "mean_token_accuracy": 0.9738784030079841,
+      "num_tokens": 9127647.0,
+      "step": 3940
+    },
+    {
+      "epoch": 9.494870247435124,
+      "eval_entropy": 0.22986975881490815,
+      "eval_loss": 1.0687925815582275,
+      "eval_mean_token_accuracy": 0.8282996858103892,
+      "eval_num_tokens": 9127647.0,
+      "eval_runtime": 88.2224,
+      "eval_samples_per_second": 16.096,
+      "eval_steps_per_second": 2.018,
+      "step": 3940
+    },
+    {
+      "entropy": 0.08347254800610245,
+      "epoch": 9.543150271575136,
+      "grad_norm": 0.18671369552612305,
+      "learning_rate": 2.422757214761356e-06,
+      "loss": 0.06384263038635254,
+      "mean_token_accuracy": 0.9770128890872002,
+      "num_tokens": 9172587.0,
+      "step": 3960
+    },
+    {
+      "epoch": 9.543150271575136,
+      "eval_entropy": 0.22975517851248217,
+      "eval_loss": 1.0701013803482056,
+      "eval_mean_token_accuracy": 0.8282244285840666,
+      "eval_num_tokens": 9172587.0,
+      "eval_runtime": 88.5794,
+      "eval_samples_per_second": 16.031,
+      "eval_steps_per_second": 2.009,
+      "step": 3960
+    },
+    {
+      "entropy": 0.08136763009242713,
+      "epoch": 9.591430295715147,
+      "grad_norm": 0.24811454117298126,
+      "learning_rate": 1.942767512805673e-06,
+      "loss": 0.06044899821281433,
+      "mean_token_accuracy": 0.9776685133576393,
+      "num_tokens": 9219278.0,
+      "step": 3980
+    },
+    {
+      "epoch": 9.591430295715147,
+      "eval_entropy": 0.22938149347064202,
+      "eval_loss": 1.0711463689804077,
+      "eval_mean_token_accuracy": 0.8282721410976367,
+      "eval_num_tokens": 9219278.0,
+      "eval_runtime": 88.4978,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 3980
+    },
+    {
+      "entropy": 0.09432941288687288,
+      "epoch": 9.63971031985516,
+      "grad_norm": 0.30733999609947205,
+      "learning_rate": 1.515470378197889e-06,
+      "loss": 0.0672213613986969,
+      "mean_token_accuracy": 0.9728612303733826,
+      "num_tokens": 9262346.0,
+      "step": 4000
+    },
+    {
+      "epoch": 9.63971031985516,
+      "eval_entropy": 0.22936245958121976,
+      "eval_loss": 1.071343183517456,
+      "eval_mean_token_accuracy": 0.8282827025049189,
+      "eval_num_tokens": 9262346.0,
+      "eval_runtime": 88.4428,
+      "eval_samples_per_second": 16.056,
+      "eval_steps_per_second": 2.013,
+      "step": 4000
+    },
+    {
+      "entropy": 0.08287148931995034,
+      "epoch": 9.687990343995171,
+      "grad_norm": 0.1976361870765686,
+      "learning_rate": 1.1409867309726256e-06,
+      "loss": 0.06318551301956177,
+      "mean_token_accuracy": 0.9770804084837437,
+      "num_tokens": 9309622.0,
+      "step": 4020
+    },
+    {
+      "epoch": 9.687990343995171,
+      "eval_entropy": 0.22929051920269314,
+      "eval_loss": 1.0711324214935303,
+      "eval_mean_token_accuracy": 0.8281597297513084,
+      "eval_num_tokens": 9309622.0,
+      "eval_runtime": 88.5243,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 4020
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.93548827168383e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4040/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4040/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4040/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4040/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4040/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4040/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4040/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4040/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4040/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4040/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4040/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4040/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4040/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4040/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..8494e1b2ab0ec89735e84590edb1a22a03f1e9d0
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4040/trainer_state.json
@@ -0,0 +1,4276 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 9.736270368135184,
+  "eval_steps": 20,
+  "global_step": 4040,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    },
+    {
+      "entropy": 0.1326121488586068,
+      "epoch": 7.181050090525045,
+      "grad_norm": 0.4843117296695709,
+      "learning_rate": 8.411870386507724e-05,
+      "loss": 0.09537842869758606,
+      "mean_token_accuracy": 0.9645151920616627,
+      "num_tokens": 6898444.0,
+      "step": 2980
+    },
+    {
+      "epoch": 7.181050090525045,
+      "eval_entropy": 0.27999477783280813,
+      "eval_loss": 0.9031037092208862,
+      "eval_mean_token_accuracy": 0.8261882204018282,
+      "eval_num_tokens": 6898444.0,
+      "eval_runtime": 88.5602,
+      "eval_samples_per_second": 16.034,
+      "eval_steps_per_second": 2.01,
+      "step": 2980
+    },
+    {
+      "entropy": 0.1156751093454659,
+      "epoch": 7.229330114665057,
+      "grad_norm": 0.5297922492027283,
+      "learning_rate": 8.149629351430326e-05,
+      "loss": 0.09203402996063233,
+      "mean_token_accuracy": 0.9689733698964119,
+      "num_tokens": 6947106.0,
+      "step": 3000
+    },
+    {
+      "epoch": 7.229330114665057,
+      "eval_entropy": 0.2740225845508361,
+      "eval_loss": 0.8876440525054932,
+      "eval_mean_token_accuracy": 0.8265116857678703,
+      "eval_num_tokens": 6947106.0,
+      "eval_runtime": 87.7971,
+      "eval_samples_per_second": 16.174,
+      "eval_steps_per_second": 2.027,
+      "step": 3000
+    },
+    {
+      "entropy": 0.12116114404052496,
+      "epoch": 7.27761013880507,
+      "grad_norm": 0.5467398166656494,
+      "learning_rate": 7.890406302404618e-05,
+      "loss": 0.0933056652545929,
+      "mean_token_accuracy": 0.9676455594599247,
+      "num_tokens": 6992784.0,
+      "step": 3020
+    },
+    {
+      "epoch": 7.27761013880507,
+      "eval_entropy": 0.27238342793804876,
+      "eval_loss": 0.9166117310523987,
+      "eval_mean_token_accuracy": 0.825499015242866,
+      "eval_num_tokens": 6992784.0,
+      "eval_runtime": 88.3276,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 3020
+    },
+    {
+      "entropy": 0.11715119928121567,
+      "epoch": 7.325890162945082,
+      "grad_norm": 0.4976702332496643,
+      "learning_rate": 7.634274596487176e-05,
+      "loss": 0.09449659585952759,
+      "mean_token_accuracy": 0.966506478190422,
+      "num_tokens": 7040482.0,
+      "step": 3040
+    },
+    {
+      "epoch": 7.325890162945082,
+      "eval_entropy": 0.2788975556077582,
+      "eval_loss": 0.8833065629005432,
+      "eval_mean_token_accuracy": 0.8269920596915684,
+      "eval_num_tokens": 7040482.0,
+      "eval_runtime": 88.9016,
+      "eval_samples_per_second": 15.973,
+      "eval_steps_per_second": 2.002,
+      "step": 3040
+    },
+    {
+      "entropy": 0.12565911412239075,
+      "epoch": 7.3741701870850935,
+      "grad_norm": 0.5728679895401001,
+      "learning_rate": 7.381306715921025e-05,
+      "loss": 0.09626244902610778,
+      "mean_token_accuracy": 0.9661444149911403,
+      "num_tokens": 7086261.0,
+      "step": 3060
+    },
+    {
+      "epoch": 7.3741701870850935,
+      "eval_entropy": 0.2762492011939542,
+      "eval_loss": 0.8961160778999329,
+      "eval_mean_token_accuracy": 0.826369910092836,
+      "eval_num_tokens": 7086261.0,
+      "eval_runtime": 88.5204,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 3060
+    },
+    {
+      "entropy": 0.12258567921817302,
+      "epoch": 7.422450211225105,
+      "grad_norm": 0.468985915184021,
+      "learning_rate": 7.131574247624031e-05,
+      "loss": 0.09120266437530518,
+      "mean_token_accuracy": 0.9674938447773457,
+      "num_tokens": 7133247.0,
+      "step": 3080
+    },
+    {
+      "epoch": 7.422450211225105,
+      "eval_entropy": 0.2688512086366,
+      "eval_loss": 0.9048654437065125,
+      "eval_mean_token_accuracy": 0.8265582302982888,
+      "eval_num_tokens": 7133247.0,
+      "eval_runtime": 88.164,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 3080
+    },
+    {
+      "entropy": 0.12070536343380808,
+      "epoch": 7.470730235365117,
+      "grad_norm": 0.5310841798782349,
+      "learning_rate": 6.885147862930688e-05,
+      "loss": 0.09386247992515565,
+      "mean_token_accuracy": 0.96664222702384,
+      "num_tokens": 7177810.0,
+      "step": 3100
+    },
+    {
+      "epoch": 7.470730235365117,
+      "eval_entropy": 0.26289632874593305,
+      "eval_loss": 0.902837336063385,
+      "eval_mean_token_accuracy": 0.828052184554968,
+      "eval_num_tokens": 7177810.0,
+      "eval_runtime": 88.1201,
+      "eval_samples_per_second": 16.114,
+      "eval_steps_per_second": 2.02,
+      "step": 3100
+    },
+    {
+      "entropy": 0.11322665181942285,
+      "epoch": 7.51901025950513,
+      "grad_norm": 0.6846067905426025,
+      "learning_rate": 6.642097297592908e-05,
+      "loss": 0.08772538304328918,
+      "mean_token_accuracy": 0.9691362135112286,
+      "num_tokens": 7225302.0,
+      "step": 3120
+    },
+    {
+      "epoch": 7.51901025950513,
+      "eval_entropy": 0.2670605613776807,
+      "eval_loss": 0.9130566120147705,
+      "eval_mean_token_accuracy": 0.826366903741708,
+      "eval_num_tokens": 7225302.0,
+      "eval_runtime": 88.6035,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3120
+    },
+    {
+      "entropy": 0.10985926580615342,
+      "epoch": 7.567290283645142,
+      "grad_norm": 0.4799748361110687,
+      "learning_rate": 6.402491332045683e-05,
+      "loss": 0.08672296404838561,
+      "mean_token_accuracy": 0.9698824048042297,
+      "num_tokens": 7274881.0,
+      "step": 3140
+    },
+    {
+      "epoch": 7.567290283645142,
+      "eval_entropy": 0.26657010866015146,
+      "eval_loss": 0.9056380987167358,
+      "eval_mean_token_accuracy": 0.8272326461384806,
+      "eval_num_tokens": 7274881.0,
+      "eval_runtime": 88.2967,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3140
+    },
+    {
+      "entropy": 0.11890473761595785,
+      "epoch": 7.615570307785154,
+      "grad_norm": 0.4081912934780121,
+      "learning_rate": 6.166397771942987e-05,
+      "loss": 0.09396827220916748,
+      "mean_token_accuracy": 0.967512384802103,
+      "num_tokens": 7321163.0,
+      "step": 3160
+    },
+    {
+      "epoch": 7.615570307785154,
+      "eval_entropy": 0.26364776575833226,
+      "eval_loss": 0.9204733967781067,
+      "eval_mean_token_accuracy": 0.827419729714983,
+      "eval_num_tokens": 7321163.0,
+      "eval_runtime": 88.3441,
+      "eval_samples_per_second": 16.074,
+      "eval_steps_per_second": 2.015,
+      "step": 3160
+    },
+    {
+      "entropy": 0.12209134344011545,
+      "epoch": 7.663850331925166,
+      "grad_norm": 0.5159774422645569,
+      "learning_rate": 5.93388342896962e-05,
+      "loss": 0.09292978644371033,
+      "mean_token_accuracy": 0.9658421002328396,
+      "num_tokens": 7366217.0,
+      "step": 3180
+    },
+    {
+      "epoch": 7.663850331925166,
+      "eval_entropy": 0.2714110068725736,
+      "eval_loss": 0.9143092036247253,
+      "eval_mean_token_accuracy": 0.8262251099843657,
+      "eval_num_tokens": 7366217.0,
+      "eval_runtime": 87.9635,
+      "eval_samples_per_second": 16.143,
+      "eval_steps_per_second": 2.024,
+      "step": 3180
+    },
+    {
+      "entropy": 0.11955609126016498,
+      "epoch": 7.712130356065178,
+      "grad_norm": 0.7005976438522339,
+      "learning_rate": 5.705014101934252e-05,
+      "loss": 0.09319776892662049,
+      "mean_token_accuracy": 0.9674513377249241,
+      "num_tokens": 7410887.0,
+      "step": 3200
+    },
+    {
+      "epoch": 7.712130356065178,
+      "eval_entropy": 0.2626291513442993,
+      "eval_loss": 0.91884446144104,
+      "eval_mean_token_accuracy": 0.827563955207889,
+      "eval_num_tokens": 7410887.0,
+      "eval_runtime": 87.8849,
+      "eval_samples_per_second": 16.157,
+      "eval_steps_per_second": 2.025,
+      "step": 3200
+    },
+    {
+      "entropy": 0.11767532899975777,
+      "epoch": 7.76041038020519,
+      "grad_norm": 0.4675362706184387,
+      "learning_rate": 5.479854558149187e-05,
+      "loss": 0.09038087725639343,
+      "mean_token_accuracy": 0.9669680051505566,
+      "num_tokens": 7457578.0,
+      "step": 3220
+    },
+    {
+      "epoch": 7.76041038020519,
+      "eval_entropy": 0.26581037203582486,
+      "eval_loss": 0.9146883487701416,
+      "eval_mean_token_accuracy": 0.8270345812433222,
+      "eval_num_tokens": 7457578.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 3220
+    },
+    {
+      "entropy": 0.11383061073720455,
+      "epoch": 7.808690404345202,
+      "grad_norm": 0.541247546672821,
+      "learning_rate": 5.258468515101918e-05,
+      "loss": 0.08942405581474304,
+      "mean_token_accuracy": 0.968398705124855,
+      "num_tokens": 7505761.0,
+      "step": 3240
+    },
+    {
+      "epoch": 7.808690404345202,
+      "eval_entropy": 0.26687998960861997,
+      "eval_loss": 0.9037085175514221,
+      "eval_mean_token_accuracy": 0.8274977863504646,
+      "eval_num_tokens": 7505761.0,
+      "eval_runtime": 88.5163,
+      "eval_samples_per_second": 16.042,
+      "eval_steps_per_second": 2.011,
+      "step": 3240
+    },
+    {
+      "entropy": 0.12004058873280883,
+      "epoch": 7.856970428485214,
+      "grad_norm": 0.5289552211761475,
+      "learning_rate": 5.040918622423899e-05,
+      "loss": 0.09064805507659912,
+      "mean_token_accuracy": 0.9673330299556255,
+      "num_tokens": 7552557.0,
+      "step": 3260
+    },
+    {
+      "epoch": 7.856970428485214,
+      "eval_entropy": 0.26288379251622085,
+      "eval_loss": 0.9146178364753723,
+      "eval_mean_token_accuracy": 0.8276662448149049,
+      "eval_num_tokens": 7552557.0,
+      "eval_runtime": 88.0723,
+      "eval_samples_per_second": 16.123,
+      "eval_steps_per_second": 2.021,
+      "step": 3260
+    },
+    {
+      "entropy": 0.1180689456872642,
+      "epoch": 7.905250452625227,
+      "grad_norm": 0.4754861891269684,
+      "learning_rate": 4.827266444161377e-05,
+      "loss": 0.09402655363082886,
+      "mean_token_accuracy": 0.9663361787796021,
+      "num_tokens": 7596998.0,
+      "step": 3280
+    },
+    {
+      "epoch": 7.905250452625227,
+      "eval_entropy": 0.261690996922134,
+      "eval_loss": 0.9120186567306519,
+      "eval_mean_token_accuracy": 0.8278995015647974,
+      "eval_num_tokens": 7596998.0,
+      "eval_runtime": 88.6573,
+      "eval_samples_per_second": 16.017,
+      "eval_steps_per_second": 2.008,
+      "step": 3280
+    },
+    {
+      "entropy": 0.1162164019420743,
+      "epoch": 7.953530476765239,
+      "grad_norm": 0.6203398108482361,
+      "learning_rate": 4.61757244135355e-05,
+      "loss": 0.08832768201828003,
+      "mean_token_accuracy": 0.9694220684468746,
+      "num_tokens": 7643288.0,
+      "step": 3300
+    },
+    {
+      "epoch": 7.953530476765239,
+      "eval_entropy": 0.26158678129817664,
+      "eval_loss": 0.9301361441612244,
+      "eval_mean_token_accuracy": 0.8269603885291668,
+      "eval_num_tokens": 7643288.0,
+      "eval_runtime": 88.7064,
+      "eval_samples_per_second": 16.008,
+      "eval_steps_per_second": 2.007,
+      "step": 3300
+    },
+    {
+      "entropy": 0.10877202204489089,
+      "epoch": 8.0,
+      "grad_norm": 3.058673143386841,
+      "learning_rate": 4.411895954922763e-05,
+      "loss": 0.09159615635871887,
+      "mean_token_accuracy": 0.9686878834451947,
+      "num_tokens": 7688608.0,
+      "step": 3320
+    },
+    {
+      "epoch": 8.0,
+      "eval_entropy": 0.2590438057197614,
+      "eval_loss": 0.9262065887451172,
+      "eval_mean_token_accuracy": 0.8277404546067956,
+      "eval_num_tokens": 7688608.0,
+      "eval_runtime": 87.7557,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 3320
+    },
+    {
+      "entropy": 0.09969092034734786,
+      "epoch": 8.048280024140013,
+      "grad_norm": 0.2561911642551422,
+      "learning_rate": 4.2102951888817644e-05,
+      "loss": 0.07228735685348511,
+      "mean_token_accuracy": 0.9743556551635265,
+      "num_tokens": 7732856.0,
+      "step": 3340
+    },
+    {
+      "epoch": 8.048280024140013,
+      "eval_entropy": 0.2462534454933713,
+      "eval_loss": 0.9755983352661133,
+      "eval_mean_token_accuracy": 0.827510209565752,
+      "eval_num_tokens": 7732856.0,
+      "eval_runtime": 88.0297,
+      "eval_samples_per_second": 16.131,
+      "eval_steps_per_second": 2.022,
+      "step": 3340
+    },
+    {
+      "entropy": 0.09477624730207027,
+      "epoch": 8.096560048280024,
+      "grad_norm": 0.3670356571674347,
+      "learning_rate": 4.012827193862633e-05,
+      "loss": 0.0686036467552185,
+      "mean_token_accuracy": 0.9743670664727688,
+      "num_tokens": 7778585.0,
+      "step": 3360
+    },
+    {
+      "epoch": 8.096560048280024,
+      "eval_entropy": 0.2440882951356052,
+      "eval_loss": 1.0118423700332642,
+      "eval_mean_token_accuracy": 0.8271241070849172,
+      "eval_num_tokens": 7778585.0,
+      "eval_runtime": 88.1364,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 3360
+    },
+    {
+      "entropy": 0.09417980071157217,
+      "epoch": 8.144840072420036,
+      "grad_norm": 0.5785111784934998,
+      "learning_rate": 3.819547850972133e-05,
+      "loss": 0.07100698947906495,
+      "mean_token_accuracy": 0.9737804099917412,
+      "num_tokens": 7824287.0,
+      "step": 3380
+    },
+    {
+      "epoch": 8.144840072420036,
+      "eval_entropy": 0.24603172860453637,
+      "eval_loss": 1.0021991729736328,
+      "eval_mean_token_accuracy": 0.827162122793412,
+      "eval_num_tokens": 7824287.0,
+      "eval_runtime": 88.4534,
+      "eval_samples_per_second": 16.054,
+      "eval_steps_per_second": 2.012,
+      "step": 3380
+    },
+    {
+      "entropy": 0.087222247896716,
+      "epoch": 8.193120096560047,
+      "grad_norm": 0.44848504662513733,
+      "learning_rate": 3.6305118559780393e-05,
+      "loss": 0.06325234174728393,
+      "mean_token_accuracy": 0.9769335232675076,
+      "num_tokens": 7874784.0,
+      "step": 3400
+    },
+    {
+      "epoch": 8.193120096560047,
+      "eval_entropy": 0.24544863905129807,
+      "eval_loss": 1.005008339881897,
+      "eval_mean_token_accuracy": 0.8273035240976998,
+      "eval_num_tokens": 7874784.0,
+      "eval_runtime": 88.6737,
+      "eval_samples_per_second": 16.014,
+      "eval_steps_per_second": 2.007,
+      "step": 3400
+    },
+    {
+      "entropy": 0.09037275062873959,
+      "epoch": 8.24140012070006,
+      "grad_norm": 0.23529155552387238,
+      "learning_rate": 3.4457727038308435e-05,
+      "loss": 0.06589569449424744,
+      "mean_token_accuracy": 0.9751470290124417,
+      "num_tokens": 7925240.0,
+      "step": 3420
+    },
+    {
+      "epoch": 8.24140012070006,
+      "eval_entropy": 0.24203423494368456,
+      "eval_loss": 1.0052824020385742,
+      "eval_mean_token_accuracy": 0.8275575503874361,
+      "eval_num_tokens": 7925240.0,
+      "eval_runtime": 88.3486,
+      "eval_samples_per_second": 16.073,
+      "eval_steps_per_second": 2.015,
+      "step": 3420
+    },
+    {
+      "entropy": 0.10373208015225828,
+      "epoch": 8.289680144840073,
+      "grad_norm": 0.24796408414840698,
+      "learning_rate": 3.2653826735253594e-05,
+      "loss": 0.07714771628379821,
+      "mean_token_accuracy": 0.9722891598939896,
+      "num_tokens": 7966685.0,
+      "step": 3440
+    },
+    {
+      "epoch": 8.289680144840073,
+      "eval_entropy": 0.23461157050025597,
+      "eval_loss": 1.012108325958252,
+      "eval_mean_token_accuracy": 0.82825900564033,
+      "eval_num_tokens": 7966685.0,
+      "eval_runtime": 88.1018,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 3440
+    },
+    {
+      "entropy": 0.08967954274266958,
+      "epoch": 8.337960168980084,
+      "grad_norm": 0.5314674973487854,
+      "learning_rate": 3.08939281330635e-05,
+      "loss": 0.06746266484260559,
+      "mean_token_accuracy": 0.9753133699297905,
+      "num_tokens": 8013656.0,
+      "step": 3460
+    },
+    {
+      "epoch": 8.337960168980084,
+      "eval_entropy": 0.23781998589467468,
+      "eval_loss": 1.0119913816452026,
+      "eval_mean_token_accuracy": 0.8273287157664139,
+      "eval_num_tokens": 8013656.0,
+      "eval_runtime": 88.6038,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3460
+    },
+    {
+      "entropy": 0.09280927330255509,
+      "epoch": 8.386240193120097,
+      "grad_norm": 0.30658936500549316,
+      "learning_rate": 2.9178529262224946e-05,
+      "loss": 0.06936421990394592,
+      "mean_token_accuracy": 0.9742105580866337,
+      "num_tokens": 8061108.0,
+      "step": 3480
+    },
+    {
+      "epoch": 8.386240193120097,
+      "eval_entropy": 0.23672426734747512,
+      "eval_loss": 1.0162041187286377,
+      "eval_mean_token_accuracy": 0.8278394281194451,
+      "eval_num_tokens": 8061108.0,
+      "eval_runtime": 88.6015,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 3480
+    },
+    {
+      "entropy": 0.09195122267119586,
+      "epoch": 8.43452021726011,
+      "grad_norm": 0.24512454867362976,
+      "learning_rate": 2.7508115560327256e-05,
+      "loss": 0.06996064782142639,
+      "mean_token_accuracy": 0.974104854464531,
+      "num_tokens": 8106924.0,
+      "step": 3500
+    },
+    {
+      "epoch": 8.43452021726011,
+      "eval_entropy": 0.23758245828781235,
+      "eval_loss": 1.0157747268676758,
+      "eval_mean_token_accuracy": 0.8278501579600773,
+      "eval_num_tokens": 8106924.0,
+      "eval_runtime": 88.5678,
+      "eval_samples_per_second": 16.033,
+      "eval_steps_per_second": 2.01,
+      "step": 3500
+    },
+    {
+      "entropy": 0.09518477590754629,
+      "epoch": 8.48280024140012,
+      "grad_norm": 0.2373228371143341,
+      "learning_rate": 2.5883159734688868e-05,
+      "loss": 0.06832797527313232,
+      "mean_token_accuracy": 0.9743806682527065,
+      "num_tokens": 8153070.0,
+      "step": 3520
+    },
+    {
+      "epoch": 8.48280024140012,
+      "eval_entropy": 0.2379308074210467,
+      "eval_loss": 1.0180530548095703,
+      "eval_mean_token_accuracy": 0.8282477052024241,
+      "eval_num_tokens": 8153070.0,
+      "eval_runtime": 88.4651,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 3520
+    },
+    {
+      "entropy": 0.0906269280705601,
+      "epoch": 8.531080265540133,
+      "grad_norm": 0.37602657079696655,
+      "learning_rate": 2.4304121628587173e-05,
+      "loss": 0.06921588778495788,
+      "mean_token_accuracy": 0.9745571479201317,
+      "num_tokens": 8200066.0,
+      "step": 3540
+    },
+    {
+      "epoch": 8.531080265540133,
+      "eval_entropy": 0.2357490769263064,
+      "eval_loss": 1.017139196395874,
+      "eval_mean_token_accuracy": 0.828713178299786,
+      "eval_num_tokens": 8200066.0,
+      "eval_runtime": 88.6424,
+      "eval_samples_per_second": 16.019,
+      "eval_steps_per_second": 2.008,
+      "step": 3540
+    },
+    {
+      "entropy": 0.09273334196768701,
+      "epoch": 8.579360289680144,
+      "grad_norm": 0.2978838086128235,
+      "learning_rate": 2.277144809112794e-05,
+      "loss": 0.06970887780189514,
+      "mean_token_accuracy": 0.9751374177634716,
+      "num_tokens": 8246157.0,
+      "step": 3560
+    },
+    {
+      "epoch": 8.579360289680144,
+      "eval_entropy": 0.23590933138065123,
+      "eval_loss": 1.019231915473938,
+      "eval_mean_token_accuracy": 0.8279163412163767,
+      "eval_num_tokens": 8246157.0,
+      "eval_runtime": 88.2691,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 3560
+    },
+    {
+      "entropy": 0.09854201921261847,
+      "epoch": 8.627640313820157,
+      "grad_norm": 0.5381243228912354,
+      "learning_rate": 2.1285572850792413e-05,
+      "loss": 0.07123501300811767,
+      "mean_token_accuracy": 0.9723988175392151,
+      "num_tokens": 8291752.0,
+      "step": 3580
+    },
+    {
+      "epoch": 8.627640313820157,
+      "eval_entropy": 0.23592024949494372,
+      "eval_loss": 1.020250678062439,
+      "eval_mean_token_accuracy": 0.8282466226079491,
+      "eval_num_tokens": 8291752.0,
+      "eval_runtime": 87.8939,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 3580
+    },
+    {
+      "entropy": 0.09092000788077712,
+      "epoch": 8.67592033796017,
+      "grad_norm": 0.29089561104774475,
+      "learning_rate": 1.9846916392697493e-05,
+      "loss": 0.06982720494270325,
+      "mean_token_accuracy": 0.9749995484948158,
+      "num_tokens": 8337651.0,
+      "step": 3600
+    },
+    {
+      "epoch": 8.67592033796017,
+      "eval_entropy": 0.23658913382318583,
+      "eval_loss": 1.019862413406372,
+      "eval_mean_token_accuracy": 0.8280127842104836,
+      "eval_num_tokens": 8337651.0,
+      "eval_runtime": 88.2855,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 3600
+    },
+    {
+      "entropy": 0.08708424470387399,
+      "epoch": 8.72420036210018,
+      "grad_norm": 0.29904839396476746,
+      "learning_rate": 1.8455885839603106e-05,
+      "loss": 0.06801370978355407,
+      "mean_token_accuracy": 0.9755801670253277,
+      "num_tokens": 8385330.0,
+      "step": 3620
+    },
+    {
+      "epoch": 8.72420036210018,
+      "eval_entropy": 0.23627035771862845,
+      "eval_loss": 1.0204569101333618,
+      "eval_mean_token_accuracy": 0.8279967894045155,
+      "eval_num_tokens": 8385330.0,
+      "eval_runtime": 88.1753,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 3620
+    },
+    {
+      "entropy": 0.0916255566291511,
+      "epoch": 8.772480386240193,
+      "grad_norm": 0.20571033656597137,
+      "learning_rate": 1.7112874836701502e-05,
+      "loss": 0.0692249894142151,
+      "mean_token_accuracy": 0.9745148062705994,
+      "num_tokens": 8432134.0,
+      "step": 3640
+    },
+    {
+      "epoch": 8.772480386240193,
+      "eval_entropy": 0.2340821772813797,
+      "eval_loss": 1.0212821960449219,
+      "eval_mean_token_accuracy": 0.828373622358515,
+      "eval_num_tokens": 8432134.0,
+      "eval_runtime": 87.2518,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 3640
+    },
+    {
+      "entropy": 0.0926868049427867,
+      "epoch": 8.820760410380204,
+      "grad_norm": 0.2618265450000763,
+      "learning_rate": 1.5818263440220203e-05,
+      "loss": 0.07338647246360779,
+      "mean_token_accuracy": 0.9724708944559097,
+      "num_tokens": 8476758.0,
+      "step": 3660
+    },
+    {
+      "epoch": 8.820760410380204,
+      "eval_entropy": 0.23637074575330433,
+      "eval_loss": 1.0186725854873657,
+      "eval_mean_token_accuracy": 0.8277707605549459,
+      "eval_num_tokens": 8476758.0,
+      "eval_runtime": 88.4695,
+      "eval_samples_per_second": 16.051,
+      "eval_steps_per_second": 2.012,
+      "step": 3660
+    },
+    {
+      "entropy": 0.09175782897509635,
+      "epoch": 8.869040434520217,
+      "grad_norm": 0.2889852821826935,
+      "learning_rate": 1.4572418009870604e-05,
+      "loss": 0.07066041231155396,
+      "mean_token_accuracy": 0.9725673109292984,
+      "num_tokens": 8523294.0,
+      "step": 3680
+    },
+    {
+      "epoch": 8.869040434520217,
+      "eval_entropy": 0.23945164404223473,
+      "eval_loss": 1.0145231485366821,
+      "eval_mean_token_accuracy": 0.8281202523895864,
+      "eval_num_tokens": 8523294.0,
+      "eval_runtime": 88.368,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 3680
+    },
+    {
+      "entropy": 0.09265611860901117,
+      "epoch": 8.91732045866023,
+      "grad_norm": 0.41092950105667114,
+      "learning_rate": 1.3375691105172625e-05,
+      "loss": 0.06975027322769164,
+      "mean_token_accuracy": 0.974636297672987,
+      "num_tokens": 8570358.0,
+      "step": 3700
+    },
+    {
+      "epoch": 8.91732045866023,
+      "eval_entropy": 0.239588919901446,
+      "eval_loss": 1.014059066772461,
+      "eval_mean_token_accuracy": 0.828028041660116,
+      "eval_num_tokens": 8570358.0,
+      "eval_runtime": 88.1039,
+      "eval_samples_per_second": 16.117,
+      "eval_steps_per_second": 2.02,
+      "step": 3700
+    },
+    {
+      "entropy": 0.0896465003490448,
+      "epoch": 8.965600482800241,
+      "grad_norm": 0.24000254273414612,
+      "learning_rate": 1.2228421385684305e-05,
+      "loss": 0.06743953824043274,
+      "mean_token_accuracy": 0.974006575345993,
+      "num_tokens": 8618246.0,
+      "step": 3720
+    },
+    {
+      "epoch": 8.965600482800241,
+      "eval_entropy": 0.2390461369846644,
+      "eval_loss": 1.011997938156128,
+      "eval_mean_token_accuracy": 0.8284533211354459,
+      "eval_num_tokens": 8618246.0,
+      "eval_runtime": 88.6202,
+      "eval_samples_per_second": 16.023,
+      "eval_steps_per_second": 2.009,
+      "step": 3720
+    },
+    {
+      "entropy": 0.09575109251520851,
+      "epoch": 9.012070006035003,
+      "grad_norm": 0.18088576197624207,
+      "learning_rate": 1.1130933515165429e-05,
+      "loss": 0.07295305132865906,
+      "mean_token_accuracy": 0.9739722857227573,
+      "num_tokens": 8661177.0,
+      "step": 3740
+    },
+    {
+      "epoch": 9.012070006035003,
+      "eval_entropy": 0.23728882387447892,
+      "eval_loss": 1.0134038925170898,
+      "eval_mean_token_accuracy": 0.8286508951294288,
+      "eval_num_tokens": 8661177.0,
+      "eval_runtime": 88.6434,
+      "eval_samples_per_second": 16.019,
+      "eval_steps_per_second": 2.008,
+      "step": 3740
+    },
+    {
+      "entropy": 0.08883201805874705,
+      "epoch": 9.060350030175016,
+      "grad_norm": 0.23120898008346558,
+      "learning_rate": 1.008353806970132e-05,
+      "loss": 0.06326142549514771,
+      "mean_token_accuracy": 0.9780020996928215,
+      "num_tokens": 8707602.0,
+      "step": 3760
+    },
+    {
+      "epoch": 9.060350030175016,
+      "eval_entropy": 0.23416905331142832,
+      "eval_loss": 1.0284100770950317,
+      "eval_mean_token_accuracy": 0.8284083213029283,
+      "eval_num_tokens": 8707602.0,
+      "eval_runtime": 88.2933,
+      "eval_samples_per_second": 16.083,
+      "eval_steps_per_second": 2.016,
+      "step": 3760
+    },
+    {
+      "entropy": 0.08861468276008963,
+      "epoch": 9.108630054315027,
+      "grad_norm": 0.17277656495571136,
+      "learning_rate": 9.086531449813489e-06,
+      "loss": 0.06727416515350342,
+      "mean_token_accuracy": 0.9760390736162663,
+      "num_tokens": 8751287.0,
+      "step": 3780
+    },
+    {
+      "epoch": 9.108630054315027,
+      "eval_entropy": 0.23370317386442355,
+      "eval_loss": 1.0381091833114624,
+      "eval_mean_token_accuracy": 0.8282668178670862,
+      "eval_num_tokens": 8751287.0,
+      "eval_runtime": 88.5738,
+      "eval_samples_per_second": 16.032,
+      "eval_steps_per_second": 2.01,
+      "step": 3780
+    },
+    {
+      "entropy": 0.09032954541034996,
+      "epoch": 9.15691007845504,
+      "grad_norm": 0.19481085240840912,
+      "learning_rate": 8.140195796581981e-06,
+      "loss": 0.06117786169052124,
+      "mean_token_accuracy": 0.9764454193413258,
+      "num_tokens": 8799982.0,
+      "step": 3800
+    },
+    {
+      "epoch": 9.15691007845504,
+      "eval_entropy": 0.23215936609868254,
+      "eval_loss": 1.0467767715454102,
+      "eval_mean_token_accuracy": 0.828405868806196,
+      "eval_num_tokens": 8799982.0,
+      "eval_runtime": 87.9551,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 3800
+    },
+    {
+      "entropy": 0.08824449330568314,
+      "epoch": 9.20519010259505,
+      "grad_norm": 0.20382793247699738,
+      "learning_rate": 7.2447989118025455e-06,
+      "loss": 0.06422553658485412,
+      "mean_token_accuracy": 0.9763121999800205,
+      "num_tokens": 8845536.0,
+      "step": 3820
+    },
+    {
+      "epoch": 9.20519010259505,
+      "eval_entropy": 0.23169661472352704,
+      "eval_loss": 1.0529043674468994,
+      "eval_mean_token_accuracy": 0.8282025610463003,
+      "eval_num_tokens": 8845536.0,
+      "eval_runtime": 88.5548,
+      "eval_samples_per_second": 16.035,
+      "eval_steps_per_second": 2.01,
+      "step": 3820
+    },
+    {
+      "entropy": 0.07768499720841646,
+      "epoch": 9.253470126735063,
+      "grad_norm": 0.11727919429540634,
+      "learning_rate": 6.400594182202084e-06,
+      "loss": 0.057279336452484134,
+      "mean_token_accuracy": 0.9795166745781898,
+      "num_tokens": 8896053.0,
+      "step": 3840
+    },
+    {
+      "epoch": 9.253470126735063,
+      "eval_entropy": 0.2301936373114586,
+      "eval_loss": 1.0576049089431763,
+      "eval_mean_token_accuracy": 0.8283417489421502,
+      "eval_num_tokens": 8896053.0,
+      "eval_runtime": 88.7854,
+      "eval_samples_per_second": 15.994,
+      "eval_steps_per_second": 2.005,
+      "step": 3840
+    },
+    {
+      "entropy": 0.09015200780704617,
+      "epoch": 9.301750150875076,
+      "grad_norm": 0.2089058756828308,
+      "learning_rate": 5.6078205077330705e-06,
+      "loss": 0.06626194715499878,
+      "mean_token_accuracy": 0.9747506938874722,
+      "num_tokens": 8940591.0,
+      "step": 3860
+    },
+    {
+      "epoch": 9.301750150875076,
+      "eval_entropy": 0.23024900676159377,
+      "eval_loss": 1.0599786043167114,
+      "eval_mean_token_accuracy": 0.8283051915383071,
+      "eval_num_tokens": 8940591.0,
+      "eval_runtime": 88.2963,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3860
+    },
+    {
+      "entropy": 0.08484094133600592,
+      "epoch": 9.350030175015087,
+      "grad_norm": 0.21646778285503387,
+      "learning_rate": 4.866702233967785e-06,
+      "loss": 0.05998039245605469,
+      "mean_token_accuracy": 0.9771868668496608,
+      "num_tokens": 8989035.0,
+      "step": 3880
+    },
+    {
+      "epoch": 9.350030175015087,
+      "eval_entropy": 0.2300723305578982,
+      "eval_loss": 1.0633587837219238,
+      "eval_mean_token_accuracy": 0.828371500701047,
+      "eval_num_tokens": 8989035.0,
+      "eval_runtime": 88.0233,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 3880
+    },
+    {
+      "entropy": 0.08905019680969417,
+      "epoch": 9.3983101991551,
+      "grad_norm": 0.21842406690120697,
+      "learning_rate": 4.177449088611095e-06,
+      "loss": 0.06666865348815917,
+      "mean_token_accuracy": 0.9753445640206337,
+      "num_tokens": 9033383.0,
+      "step": 3900
+    },
+    {
+      "epoch": 9.3983101991551,
+      "eval_entropy": 0.2295504841911659,
+      "eval_loss": 1.0662461519241333,
+      "eval_mean_token_accuracy": 0.8284122297603093,
+      "eval_num_tokens": 9033383.0,
+      "eval_runtime": 88.5525,
+      "eval_samples_per_second": 16.036,
+      "eval_steps_per_second": 2.01,
+      "step": 3900
+    },
+    {
+      "entropy": 0.08582521872594953,
+      "epoch": 9.446590223295111,
+      "grad_norm": 0.2000240683555603,
+      "learning_rate": 3.540256122150088e-06,
+      "loss": 0.06327944993972778,
+      "mean_token_accuracy": 0.9765432856976985,
+      "num_tokens": 9079672.0,
+      "step": 3920
+    },
+    {
+      "epoch": 9.446590223295111,
+      "eval_entropy": 0.2297678625818049,
+      "eval_loss": 1.0676169395446777,
+      "eval_mean_token_accuracy": 0.8283703879693921,
+      "eval_num_tokens": 9079672.0,
+      "eval_runtime": 87.6897,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 3920
+    },
+    {
+      "entropy": 0.09261773955076932,
+      "epoch": 9.494870247435124,
+      "grad_norm": 0.19104281067848206,
+      "learning_rate": 2.955303652656991e-06,
+      "loss": 0.06111966967582703,
+      "mean_token_accuracy": 0.9738784030079841,
+      "num_tokens": 9127647.0,
+      "step": 3940
+    },
+    {
+      "epoch": 9.494870247435124,
+      "eval_entropy": 0.22986975881490815,
+      "eval_loss": 1.0687925815582275,
+      "eval_mean_token_accuracy": 0.8282996858103892,
+      "eval_num_tokens": 9127647.0,
+      "eval_runtime": 88.2224,
+      "eval_samples_per_second": 16.096,
+      "eval_steps_per_second": 2.018,
+      "step": 3940
+    },
+    {
+      "entropy": 0.08347254800610245,
+      "epoch": 9.543150271575136,
+      "grad_norm": 0.18671369552612305,
+      "learning_rate": 2.422757214761356e-06,
+      "loss": 0.06384263038635254,
+      "mean_token_accuracy": 0.9770128890872002,
+      "num_tokens": 9172587.0,
+      "step": 3960
+    },
+    {
+      "epoch": 9.543150271575136,
+      "eval_entropy": 0.22975517851248217,
+      "eval_loss": 1.0701013803482056,
+      "eval_mean_token_accuracy": 0.8282244285840666,
+      "eval_num_tokens": 9172587.0,
+      "eval_runtime": 88.5794,
+      "eval_samples_per_second": 16.031,
+      "eval_steps_per_second": 2.009,
+      "step": 3960
+    },
+    {
+      "entropy": 0.08136763009242713,
+      "epoch": 9.591430295715147,
+      "grad_norm": 0.24811454117298126,
+      "learning_rate": 1.942767512805673e-06,
+      "loss": 0.06044899821281433,
+      "mean_token_accuracy": 0.9776685133576393,
+      "num_tokens": 9219278.0,
+      "step": 3980
+    },
+    {
+      "epoch": 9.591430295715147,
+      "eval_entropy": 0.22938149347064202,
+      "eval_loss": 1.0711463689804077,
+      "eval_mean_token_accuracy": 0.8282721410976367,
+      "eval_num_tokens": 9219278.0,
+      "eval_runtime": 88.4978,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 3980
+    },
+    {
+      "entropy": 0.09432941288687288,
+      "epoch": 9.63971031985516,
+      "grad_norm": 0.30733999609947205,
+      "learning_rate": 1.515470378197889e-06,
+      "loss": 0.0672213613986969,
+      "mean_token_accuracy": 0.9728612303733826,
+      "num_tokens": 9262346.0,
+      "step": 4000
+    },
+    {
+      "epoch": 9.63971031985516,
+      "eval_entropy": 0.22936245958121976,
+      "eval_loss": 1.071343183517456,
+      "eval_mean_token_accuracy": 0.8282827025049189,
+      "eval_num_tokens": 9262346.0,
+      "eval_runtime": 88.4428,
+      "eval_samples_per_second": 16.056,
+      "eval_steps_per_second": 2.013,
+      "step": 4000
+    },
+    {
+      "entropy": 0.08287148931995034,
+      "epoch": 9.687990343995171,
+      "grad_norm": 0.1976361870765686,
+      "learning_rate": 1.1409867309726256e-06,
+      "loss": 0.06318551301956177,
+      "mean_token_accuracy": 0.9770804084837437,
+      "num_tokens": 9309622.0,
+      "step": 4020
+    },
+    {
+      "epoch": 9.687990343995171,
+      "eval_entropy": 0.22929051920269314,
+      "eval_loss": 1.0711324214935303,
+      "eval_mean_token_accuracy": 0.8281597297513084,
+      "eval_num_tokens": 9309622.0,
+      "eval_runtime": 88.5243,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 4020
+    },
+    {
+      "entropy": 0.0907150611979887,
+      "epoch": 9.736270368135184,
+      "grad_norm": 0.16192017495632172,
+      "learning_rate": 8.194225455723828e-07,
+      "loss": 0.06589602828025817,
+      "mean_token_accuracy": 0.9738550461828709,
+      "num_tokens": 9354549.0,
+      "step": 4040
+    },
+    {
+      "epoch": 9.736270368135184,
+      "eval_entropy": 0.22913936391640244,
+      "eval_loss": 1.0715173482894897,
+      "eval_mean_token_accuracy": 0.8281301544623428,
+      "eval_num_tokens": 9354549.0,
+      "eval_runtime": 88.875,
+      "eval_samples_per_second": 15.978,
+      "eval_steps_per_second": 2.003,
+      "step": 4040
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.9548795845973606e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4060/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4060/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4060/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4060/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4060/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4060/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4060/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4060/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4060/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4060/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4060/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4060/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4060/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4060/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..45b45b5ca7473bdc57fcfdfbdbd9005c4b3b191f
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4060/trainer_state.json
@@ -0,0 +1,4297 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 9.784550392275197,
+  "eval_steps": 20,
+  "global_step": 4060,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    },
+    {
+      "entropy": 0.1326121488586068,
+      "epoch": 7.181050090525045,
+      "grad_norm": 0.4843117296695709,
+      "learning_rate": 8.411870386507724e-05,
+      "loss": 0.09537842869758606,
+      "mean_token_accuracy": 0.9645151920616627,
+      "num_tokens": 6898444.0,
+      "step": 2980
+    },
+    {
+      "epoch": 7.181050090525045,
+      "eval_entropy": 0.27999477783280813,
+      "eval_loss": 0.9031037092208862,
+      "eval_mean_token_accuracy": 0.8261882204018282,
+      "eval_num_tokens": 6898444.0,
+      "eval_runtime": 88.5602,
+      "eval_samples_per_second": 16.034,
+      "eval_steps_per_second": 2.01,
+      "step": 2980
+    },
+    {
+      "entropy": 0.1156751093454659,
+      "epoch": 7.229330114665057,
+      "grad_norm": 0.5297922492027283,
+      "learning_rate": 8.149629351430326e-05,
+      "loss": 0.09203402996063233,
+      "mean_token_accuracy": 0.9689733698964119,
+      "num_tokens": 6947106.0,
+      "step": 3000
+    },
+    {
+      "epoch": 7.229330114665057,
+      "eval_entropy": 0.2740225845508361,
+      "eval_loss": 0.8876440525054932,
+      "eval_mean_token_accuracy": 0.8265116857678703,
+      "eval_num_tokens": 6947106.0,
+      "eval_runtime": 87.7971,
+      "eval_samples_per_second": 16.174,
+      "eval_steps_per_second": 2.027,
+      "step": 3000
+    },
+    {
+      "entropy": 0.12116114404052496,
+      "epoch": 7.27761013880507,
+      "grad_norm": 0.5467398166656494,
+      "learning_rate": 7.890406302404618e-05,
+      "loss": 0.0933056652545929,
+      "mean_token_accuracy": 0.9676455594599247,
+      "num_tokens": 6992784.0,
+      "step": 3020
+    },
+    {
+      "epoch": 7.27761013880507,
+      "eval_entropy": 0.27238342793804876,
+      "eval_loss": 0.9166117310523987,
+      "eval_mean_token_accuracy": 0.825499015242866,
+      "eval_num_tokens": 6992784.0,
+      "eval_runtime": 88.3276,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 3020
+    },
+    {
+      "entropy": 0.11715119928121567,
+      "epoch": 7.325890162945082,
+      "grad_norm": 0.4976702332496643,
+      "learning_rate": 7.634274596487176e-05,
+      "loss": 0.09449659585952759,
+      "mean_token_accuracy": 0.966506478190422,
+      "num_tokens": 7040482.0,
+      "step": 3040
+    },
+    {
+      "epoch": 7.325890162945082,
+      "eval_entropy": 0.2788975556077582,
+      "eval_loss": 0.8833065629005432,
+      "eval_mean_token_accuracy": 0.8269920596915684,
+      "eval_num_tokens": 7040482.0,
+      "eval_runtime": 88.9016,
+      "eval_samples_per_second": 15.973,
+      "eval_steps_per_second": 2.002,
+      "step": 3040
+    },
+    {
+      "entropy": 0.12565911412239075,
+      "epoch": 7.3741701870850935,
+      "grad_norm": 0.5728679895401001,
+      "learning_rate": 7.381306715921025e-05,
+      "loss": 0.09626244902610778,
+      "mean_token_accuracy": 0.9661444149911403,
+      "num_tokens": 7086261.0,
+      "step": 3060
+    },
+    {
+      "epoch": 7.3741701870850935,
+      "eval_entropy": 0.2762492011939542,
+      "eval_loss": 0.8961160778999329,
+      "eval_mean_token_accuracy": 0.826369910092836,
+      "eval_num_tokens": 7086261.0,
+      "eval_runtime": 88.5204,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 3060
+    },
+    {
+      "entropy": 0.12258567921817302,
+      "epoch": 7.422450211225105,
+      "grad_norm": 0.468985915184021,
+      "learning_rate": 7.131574247624031e-05,
+      "loss": 0.09120266437530518,
+      "mean_token_accuracy": 0.9674938447773457,
+      "num_tokens": 7133247.0,
+      "step": 3080
+    },
+    {
+      "epoch": 7.422450211225105,
+      "eval_entropy": 0.2688512086366,
+      "eval_loss": 0.9048654437065125,
+      "eval_mean_token_accuracy": 0.8265582302982888,
+      "eval_num_tokens": 7133247.0,
+      "eval_runtime": 88.164,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 3080
+    },
+    {
+      "entropy": 0.12070536343380808,
+      "epoch": 7.470730235365117,
+      "grad_norm": 0.5310841798782349,
+      "learning_rate": 6.885147862930688e-05,
+      "loss": 0.09386247992515565,
+      "mean_token_accuracy": 0.96664222702384,
+      "num_tokens": 7177810.0,
+      "step": 3100
+    },
+    {
+      "epoch": 7.470730235365117,
+      "eval_entropy": 0.26289632874593305,
+      "eval_loss": 0.902837336063385,
+      "eval_mean_token_accuracy": 0.828052184554968,
+      "eval_num_tokens": 7177810.0,
+      "eval_runtime": 88.1201,
+      "eval_samples_per_second": 16.114,
+      "eval_steps_per_second": 2.02,
+      "step": 3100
+    },
+    {
+      "entropy": 0.11322665181942285,
+      "epoch": 7.51901025950513,
+      "grad_norm": 0.6846067905426025,
+      "learning_rate": 6.642097297592908e-05,
+      "loss": 0.08772538304328918,
+      "mean_token_accuracy": 0.9691362135112286,
+      "num_tokens": 7225302.0,
+      "step": 3120
+    },
+    {
+      "epoch": 7.51901025950513,
+      "eval_entropy": 0.2670605613776807,
+      "eval_loss": 0.9130566120147705,
+      "eval_mean_token_accuracy": 0.826366903741708,
+      "eval_num_tokens": 7225302.0,
+      "eval_runtime": 88.6035,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3120
+    },
+    {
+      "entropy": 0.10985926580615342,
+      "epoch": 7.567290283645142,
+      "grad_norm": 0.4799748361110687,
+      "learning_rate": 6.402491332045683e-05,
+      "loss": 0.08672296404838561,
+      "mean_token_accuracy": 0.9698824048042297,
+      "num_tokens": 7274881.0,
+      "step": 3140
+    },
+    {
+      "epoch": 7.567290283645142,
+      "eval_entropy": 0.26657010866015146,
+      "eval_loss": 0.9056380987167358,
+      "eval_mean_token_accuracy": 0.8272326461384806,
+      "eval_num_tokens": 7274881.0,
+      "eval_runtime": 88.2967,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3140
+    },
+    {
+      "entropy": 0.11890473761595785,
+      "epoch": 7.615570307785154,
+      "grad_norm": 0.4081912934780121,
+      "learning_rate": 6.166397771942987e-05,
+      "loss": 0.09396827220916748,
+      "mean_token_accuracy": 0.967512384802103,
+      "num_tokens": 7321163.0,
+      "step": 3160
+    },
+    {
+      "epoch": 7.615570307785154,
+      "eval_entropy": 0.26364776575833226,
+      "eval_loss": 0.9204733967781067,
+      "eval_mean_token_accuracy": 0.827419729714983,
+      "eval_num_tokens": 7321163.0,
+      "eval_runtime": 88.3441,
+      "eval_samples_per_second": 16.074,
+      "eval_steps_per_second": 2.015,
+      "step": 3160
+    },
+    {
+      "entropy": 0.12209134344011545,
+      "epoch": 7.663850331925166,
+      "grad_norm": 0.5159774422645569,
+      "learning_rate": 5.93388342896962e-05,
+      "loss": 0.09292978644371033,
+      "mean_token_accuracy": 0.9658421002328396,
+      "num_tokens": 7366217.0,
+      "step": 3180
+    },
+    {
+      "epoch": 7.663850331925166,
+      "eval_entropy": 0.2714110068725736,
+      "eval_loss": 0.9143092036247253,
+      "eval_mean_token_accuracy": 0.8262251099843657,
+      "eval_num_tokens": 7366217.0,
+      "eval_runtime": 87.9635,
+      "eval_samples_per_second": 16.143,
+      "eval_steps_per_second": 2.024,
+      "step": 3180
+    },
+    {
+      "entropy": 0.11955609126016498,
+      "epoch": 7.712130356065178,
+      "grad_norm": 0.7005976438522339,
+      "learning_rate": 5.705014101934252e-05,
+      "loss": 0.09319776892662049,
+      "mean_token_accuracy": 0.9674513377249241,
+      "num_tokens": 7410887.0,
+      "step": 3200
+    },
+    {
+      "epoch": 7.712130356065178,
+      "eval_entropy": 0.2626291513442993,
+      "eval_loss": 0.91884446144104,
+      "eval_mean_token_accuracy": 0.827563955207889,
+      "eval_num_tokens": 7410887.0,
+      "eval_runtime": 87.8849,
+      "eval_samples_per_second": 16.157,
+      "eval_steps_per_second": 2.025,
+      "step": 3200
+    },
+    {
+      "entropy": 0.11767532899975777,
+      "epoch": 7.76041038020519,
+      "grad_norm": 0.4675362706184387,
+      "learning_rate": 5.479854558149187e-05,
+      "loss": 0.09038087725639343,
+      "mean_token_accuracy": 0.9669680051505566,
+      "num_tokens": 7457578.0,
+      "step": 3220
+    },
+    {
+      "epoch": 7.76041038020519,
+      "eval_entropy": 0.26581037203582486,
+      "eval_loss": 0.9146883487701416,
+      "eval_mean_token_accuracy": 0.8270345812433222,
+      "eval_num_tokens": 7457578.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 3220
+    },
+    {
+      "entropy": 0.11383061073720455,
+      "epoch": 7.808690404345202,
+      "grad_norm": 0.541247546672821,
+      "learning_rate": 5.258468515101918e-05,
+      "loss": 0.08942405581474304,
+      "mean_token_accuracy": 0.968398705124855,
+      "num_tokens": 7505761.0,
+      "step": 3240
+    },
+    {
+      "epoch": 7.808690404345202,
+      "eval_entropy": 0.26687998960861997,
+      "eval_loss": 0.9037085175514221,
+      "eval_mean_token_accuracy": 0.8274977863504646,
+      "eval_num_tokens": 7505761.0,
+      "eval_runtime": 88.5163,
+      "eval_samples_per_second": 16.042,
+      "eval_steps_per_second": 2.011,
+      "step": 3240
+    },
+    {
+      "entropy": 0.12004058873280883,
+      "epoch": 7.856970428485214,
+      "grad_norm": 0.5289552211761475,
+      "learning_rate": 5.040918622423899e-05,
+      "loss": 0.09064805507659912,
+      "mean_token_accuracy": 0.9673330299556255,
+      "num_tokens": 7552557.0,
+      "step": 3260
+    },
+    {
+      "epoch": 7.856970428485214,
+      "eval_entropy": 0.26288379251622085,
+      "eval_loss": 0.9146178364753723,
+      "eval_mean_token_accuracy": 0.8276662448149049,
+      "eval_num_tokens": 7552557.0,
+      "eval_runtime": 88.0723,
+      "eval_samples_per_second": 16.123,
+      "eval_steps_per_second": 2.021,
+      "step": 3260
+    },
+    {
+      "entropy": 0.1180689456872642,
+      "epoch": 7.905250452625227,
+      "grad_norm": 0.4754861891269684,
+      "learning_rate": 4.827266444161377e-05,
+      "loss": 0.09402655363082886,
+      "mean_token_accuracy": 0.9663361787796021,
+      "num_tokens": 7596998.0,
+      "step": 3280
+    },
+    {
+      "epoch": 7.905250452625227,
+      "eval_entropy": 0.261690996922134,
+      "eval_loss": 0.9120186567306519,
+      "eval_mean_token_accuracy": 0.8278995015647974,
+      "eval_num_tokens": 7596998.0,
+      "eval_runtime": 88.6573,
+      "eval_samples_per_second": 16.017,
+      "eval_steps_per_second": 2.008,
+      "step": 3280
+    },
+    {
+      "entropy": 0.1162164019420743,
+      "epoch": 7.953530476765239,
+      "grad_norm": 0.6203398108482361,
+      "learning_rate": 4.61757244135355e-05,
+      "loss": 0.08832768201828003,
+      "mean_token_accuracy": 0.9694220684468746,
+      "num_tokens": 7643288.0,
+      "step": 3300
+    },
+    {
+      "epoch": 7.953530476765239,
+      "eval_entropy": 0.26158678129817664,
+      "eval_loss": 0.9301361441612244,
+      "eval_mean_token_accuracy": 0.8269603885291668,
+      "eval_num_tokens": 7643288.0,
+      "eval_runtime": 88.7064,
+      "eval_samples_per_second": 16.008,
+      "eval_steps_per_second": 2.007,
+      "step": 3300
+    },
+    {
+      "entropy": 0.10877202204489089,
+      "epoch": 8.0,
+      "grad_norm": 3.058673143386841,
+      "learning_rate": 4.411895954922763e-05,
+      "loss": 0.09159615635871887,
+      "mean_token_accuracy": 0.9686878834451947,
+      "num_tokens": 7688608.0,
+      "step": 3320
+    },
+    {
+      "epoch": 8.0,
+      "eval_entropy": 0.2590438057197614,
+      "eval_loss": 0.9262065887451172,
+      "eval_mean_token_accuracy": 0.8277404546067956,
+      "eval_num_tokens": 7688608.0,
+      "eval_runtime": 87.7557,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 3320
+    },
+    {
+      "entropy": 0.09969092034734786,
+      "epoch": 8.048280024140013,
+      "grad_norm": 0.2561911642551422,
+      "learning_rate": 4.2102951888817644e-05,
+      "loss": 0.07228735685348511,
+      "mean_token_accuracy": 0.9743556551635265,
+      "num_tokens": 7732856.0,
+      "step": 3340
+    },
+    {
+      "epoch": 8.048280024140013,
+      "eval_entropy": 0.2462534454933713,
+      "eval_loss": 0.9755983352661133,
+      "eval_mean_token_accuracy": 0.827510209565752,
+      "eval_num_tokens": 7732856.0,
+      "eval_runtime": 88.0297,
+      "eval_samples_per_second": 16.131,
+      "eval_steps_per_second": 2.022,
+      "step": 3340
+    },
+    {
+      "entropy": 0.09477624730207027,
+      "epoch": 8.096560048280024,
+      "grad_norm": 0.3670356571674347,
+      "learning_rate": 4.012827193862633e-05,
+      "loss": 0.0686036467552185,
+      "mean_token_accuracy": 0.9743670664727688,
+      "num_tokens": 7778585.0,
+      "step": 3360
+    },
+    {
+      "epoch": 8.096560048280024,
+      "eval_entropy": 0.2440882951356052,
+      "eval_loss": 1.0118423700332642,
+      "eval_mean_token_accuracy": 0.8271241070849172,
+      "eval_num_tokens": 7778585.0,
+      "eval_runtime": 88.1364,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 3360
+    },
+    {
+      "entropy": 0.09417980071157217,
+      "epoch": 8.144840072420036,
+      "grad_norm": 0.5785111784934998,
+      "learning_rate": 3.819547850972133e-05,
+      "loss": 0.07100698947906495,
+      "mean_token_accuracy": 0.9737804099917412,
+      "num_tokens": 7824287.0,
+      "step": 3380
+    },
+    {
+      "epoch": 8.144840072420036,
+      "eval_entropy": 0.24603172860453637,
+      "eval_loss": 1.0021991729736328,
+      "eval_mean_token_accuracy": 0.827162122793412,
+      "eval_num_tokens": 7824287.0,
+      "eval_runtime": 88.4534,
+      "eval_samples_per_second": 16.054,
+      "eval_steps_per_second": 2.012,
+      "step": 3380
+    },
+    {
+      "entropy": 0.087222247896716,
+      "epoch": 8.193120096560047,
+      "grad_norm": 0.44848504662513733,
+      "learning_rate": 3.6305118559780393e-05,
+      "loss": 0.06325234174728393,
+      "mean_token_accuracy": 0.9769335232675076,
+      "num_tokens": 7874784.0,
+      "step": 3400
+    },
+    {
+      "epoch": 8.193120096560047,
+      "eval_entropy": 0.24544863905129807,
+      "eval_loss": 1.005008339881897,
+      "eval_mean_token_accuracy": 0.8273035240976998,
+      "eval_num_tokens": 7874784.0,
+      "eval_runtime": 88.6737,
+      "eval_samples_per_second": 16.014,
+      "eval_steps_per_second": 2.007,
+      "step": 3400
+    },
+    {
+      "entropy": 0.09037275062873959,
+      "epoch": 8.24140012070006,
+      "grad_norm": 0.23529155552387238,
+      "learning_rate": 3.4457727038308435e-05,
+      "loss": 0.06589569449424744,
+      "mean_token_accuracy": 0.9751470290124417,
+      "num_tokens": 7925240.0,
+      "step": 3420
+    },
+    {
+      "epoch": 8.24140012070006,
+      "eval_entropy": 0.24203423494368456,
+      "eval_loss": 1.0052824020385742,
+      "eval_mean_token_accuracy": 0.8275575503874361,
+      "eval_num_tokens": 7925240.0,
+      "eval_runtime": 88.3486,
+      "eval_samples_per_second": 16.073,
+      "eval_steps_per_second": 2.015,
+      "step": 3420
+    },
+    {
+      "entropy": 0.10373208015225828,
+      "epoch": 8.289680144840073,
+      "grad_norm": 0.24796408414840698,
+      "learning_rate": 3.2653826735253594e-05,
+      "loss": 0.07714771628379821,
+      "mean_token_accuracy": 0.9722891598939896,
+      "num_tokens": 7966685.0,
+      "step": 3440
+    },
+    {
+      "epoch": 8.289680144840073,
+      "eval_entropy": 0.23461157050025597,
+      "eval_loss": 1.012108325958252,
+      "eval_mean_token_accuracy": 0.82825900564033,
+      "eval_num_tokens": 7966685.0,
+      "eval_runtime": 88.1018,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 3440
+    },
+    {
+      "entropy": 0.08967954274266958,
+      "epoch": 8.337960168980084,
+      "grad_norm": 0.5314674973487854,
+      "learning_rate": 3.08939281330635e-05,
+      "loss": 0.06746266484260559,
+      "mean_token_accuracy": 0.9753133699297905,
+      "num_tokens": 8013656.0,
+      "step": 3460
+    },
+    {
+      "epoch": 8.337960168980084,
+      "eval_entropy": 0.23781998589467468,
+      "eval_loss": 1.0119913816452026,
+      "eval_mean_token_accuracy": 0.8273287157664139,
+      "eval_num_tokens": 8013656.0,
+      "eval_runtime": 88.6038,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3460
+    },
+    {
+      "entropy": 0.09280927330255509,
+      "epoch": 8.386240193120097,
+      "grad_norm": 0.30658936500549316,
+      "learning_rate": 2.9178529262224946e-05,
+      "loss": 0.06936421990394592,
+      "mean_token_accuracy": 0.9742105580866337,
+      "num_tokens": 8061108.0,
+      "step": 3480
+    },
+    {
+      "epoch": 8.386240193120097,
+      "eval_entropy": 0.23672426734747512,
+      "eval_loss": 1.0162041187286377,
+      "eval_mean_token_accuracy": 0.8278394281194451,
+      "eval_num_tokens": 8061108.0,
+      "eval_runtime": 88.6015,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 3480
+    },
+    {
+      "entropy": 0.09195122267119586,
+      "epoch": 8.43452021726011,
+      "grad_norm": 0.24512454867362976,
+      "learning_rate": 2.7508115560327256e-05,
+      "loss": 0.06996064782142639,
+      "mean_token_accuracy": 0.974104854464531,
+      "num_tokens": 8106924.0,
+      "step": 3500
+    },
+    {
+      "epoch": 8.43452021726011,
+      "eval_entropy": 0.23758245828781235,
+      "eval_loss": 1.0157747268676758,
+      "eval_mean_token_accuracy": 0.8278501579600773,
+      "eval_num_tokens": 8106924.0,
+      "eval_runtime": 88.5678,
+      "eval_samples_per_second": 16.033,
+      "eval_steps_per_second": 2.01,
+      "step": 3500
+    },
+    {
+      "entropy": 0.09518477590754629,
+      "epoch": 8.48280024140012,
+      "grad_norm": 0.2373228371143341,
+      "learning_rate": 2.5883159734688868e-05,
+      "loss": 0.06832797527313232,
+      "mean_token_accuracy": 0.9743806682527065,
+      "num_tokens": 8153070.0,
+      "step": 3520
+    },
+    {
+      "epoch": 8.48280024140012,
+      "eval_entropy": 0.2379308074210467,
+      "eval_loss": 1.0180530548095703,
+      "eval_mean_token_accuracy": 0.8282477052024241,
+      "eval_num_tokens": 8153070.0,
+      "eval_runtime": 88.4651,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 3520
+    },
+    {
+      "entropy": 0.0906269280705601,
+      "epoch": 8.531080265540133,
+      "grad_norm": 0.37602657079696655,
+      "learning_rate": 2.4304121628587173e-05,
+      "loss": 0.06921588778495788,
+      "mean_token_accuracy": 0.9745571479201317,
+      "num_tokens": 8200066.0,
+      "step": 3540
+    },
+    {
+      "epoch": 8.531080265540133,
+      "eval_entropy": 0.2357490769263064,
+      "eval_loss": 1.017139196395874,
+      "eval_mean_token_accuracy": 0.828713178299786,
+      "eval_num_tokens": 8200066.0,
+      "eval_runtime": 88.6424,
+      "eval_samples_per_second": 16.019,
+      "eval_steps_per_second": 2.008,
+      "step": 3540
+    },
+    {
+      "entropy": 0.09273334196768701,
+      "epoch": 8.579360289680144,
+      "grad_norm": 0.2978838086128235,
+      "learning_rate": 2.277144809112794e-05,
+      "loss": 0.06970887780189514,
+      "mean_token_accuracy": 0.9751374177634716,
+      "num_tokens": 8246157.0,
+      "step": 3560
+    },
+    {
+      "epoch": 8.579360289680144,
+      "eval_entropy": 0.23590933138065123,
+      "eval_loss": 1.019231915473938,
+      "eval_mean_token_accuracy": 0.8279163412163767,
+      "eval_num_tokens": 8246157.0,
+      "eval_runtime": 88.2691,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 3560
+    },
+    {
+      "entropy": 0.09854201921261847,
+      "epoch": 8.627640313820157,
+      "grad_norm": 0.5381243228912354,
+      "learning_rate": 2.1285572850792413e-05,
+      "loss": 0.07123501300811767,
+      "mean_token_accuracy": 0.9723988175392151,
+      "num_tokens": 8291752.0,
+      "step": 3580
+    },
+    {
+      "epoch": 8.627640313820157,
+      "eval_entropy": 0.23592024949494372,
+      "eval_loss": 1.020250678062439,
+      "eval_mean_token_accuracy": 0.8282466226079491,
+      "eval_num_tokens": 8291752.0,
+      "eval_runtime": 87.8939,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 3580
+    },
+    {
+      "entropy": 0.09092000788077712,
+      "epoch": 8.67592033796017,
+      "grad_norm": 0.29089561104774475,
+      "learning_rate": 1.9846916392697493e-05,
+      "loss": 0.06982720494270325,
+      "mean_token_accuracy": 0.9749995484948158,
+      "num_tokens": 8337651.0,
+      "step": 3600
+    },
+    {
+      "epoch": 8.67592033796017,
+      "eval_entropy": 0.23658913382318583,
+      "eval_loss": 1.019862413406372,
+      "eval_mean_token_accuracy": 0.8280127842104836,
+      "eval_num_tokens": 8337651.0,
+      "eval_runtime": 88.2855,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 3600
+    },
+    {
+      "entropy": 0.08708424470387399,
+      "epoch": 8.72420036210018,
+      "grad_norm": 0.29904839396476746,
+      "learning_rate": 1.8455885839603106e-05,
+      "loss": 0.06801370978355407,
+      "mean_token_accuracy": 0.9755801670253277,
+      "num_tokens": 8385330.0,
+      "step": 3620
+    },
+    {
+      "epoch": 8.72420036210018,
+      "eval_entropy": 0.23627035771862845,
+      "eval_loss": 1.0204569101333618,
+      "eval_mean_token_accuracy": 0.8279967894045155,
+      "eval_num_tokens": 8385330.0,
+      "eval_runtime": 88.1753,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 3620
+    },
+    {
+      "entropy": 0.0916255566291511,
+      "epoch": 8.772480386240193,
+      "grad_norm": 0.20571033656597137,
+      "learning_rate": 1.7112874836701502e-05,
+      "loss": 0.0692249894142151,
+      "mean_token_accuracy": 0.9745148062705994,
+      "num_tokens": 8432134.0,
+      "step": 3640
+    },
+    {
+      "epoch": 8.772480386240193,
+      "eval_entropy": 0.2340821772813797,
+      "eval_loss": 1.0212821960449219,
+      "eval_mean_token_accuracy": 0.828373622358515,
+      "eval_num_tokens": 8432134.0,
+      "eval_runtime": 87.2518,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 3640
+    },
+    {
+      "entropy": 0.0926868049427867,
+      "epoch": 8.820760410380204,
+      "grad_norm": 0.2618265450000763,
+      "learning_rate": 1.5818263440220203e-05,
+      "loss": 0.07338647246360779,
+      "mean_token_accuracy": 0.9724708944559097,
+      "num_tokens": 8476758.0,
+      "step": 3660
+    },
+    {
+      "epoch": 8.820760410380204,
+      "eval_entropy": 0.23637074575330433,
+      "eval_loss": 1.0186725854873657,
+      "eval_mean_token_accuracy": 0.8277707605549459,
+      "eval_num_tokens": 8476758.0,
+      "eval_runtime": 88.4695,
+      "eval_samples_per_second": 16.051,
+      "eval_steps_per_second": 2.012,
+      "step": 3660
+    },
+    {
+      "entropy": 0.09175782897509635,
+      "epoch": 8.869040434520217,
+      "grad_norm": 0.2889852821826935,
+      "learning_rate": 1.4572418009870604e-05,
+      "loss": 0.07066041231155396,
+      "mean_token_accuracy": 0.9725673109292984,
+      "num_tokens": 8523294.0,
+      "step": 3680
+    },
+    {
+      "epoch": 8.869040434520217,
+      "eval_entropy": 0.23945164404223473,
+      "eval_loss": 1.0145231485366821,
+      "eval_mean_token_accuracy": 0.8281202523895864,
+      "eval_num_tokens": 8523294.0,
+      "eval_runtime": 88.368,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 3680
+    },
+    {
+      "entropy": 0.09265611860901117,
+      "epoch": 8.91732045866023,
+      "grad_norm": 0.41092950105667114,
+      "learning_rate": 1.3375691105172625e-05,
+      "loss": 0.06975027322769164,
+      "mean_token_accuracy": 0.974636297672987,
+      "num_tokens": 8570358.0,
+      "step": 3700
+    },
+    {
+      "epoch": 8.91732045866023,
+      "eval_entropy": 0.239588919901446,
+      "eval_loss": 1.014059066772461,
+      "eval_mean_token_accuracy": 0.828028041660116,
+      "eval_num_tokens": 8570358.0,
+      "eval_runtime": 88.1039,
+      "eval_samples_per_second": 16.117,
+      "eval_steps_per_second": 2.02,
+      "step": 3700
+    },
+    {
+      "entropy": 0.0896465003490448,
+      "epoch": 8.965600482800241,
+      "grad_norm": 0.24000254273414612,
+      "learning_rate": 1.2228421385684305e-05,
+      "loss": 0.06743953824043274,
+      "mean_token_accuracy": 0.974006575345993,
+      "num_tokens": 8618246.0,
+      "step": 3720
+    },
+    {
+      "epoch": 8.965600482800241,
+      "eval_entropy": 0.2390461369846644,
+      "eval_loss": 1.011997938156128,
+      "eval_mean_token_accuracy": 0.8284533211354459,
+      "eval_num_tokens": 8618246.0,
+      "eval_runtime": 88.6202,
+      "eval_samples_per_second": 16.023,
+      "eval_steps_per_second": 2.009,
+      "step": 3720
+    },
+    {
+      "entropy": 0.09575109251520851,
+      "epoch": 9.012070006035003,
+      "grad_norm": 0.18088576197624207,
+      "learning_rate": 1.1130933515165429e-05,
+      "loss": 0.07295305132865906,
+      "mean_token_accuracy": 0.9739722857227573,
+      "num_tokens": 8661177.0,
+      "step": 3740
+    },
+    {
+      "epoch": 9.012070006035003,
+      "eval_entropy": 0.23728882387447892,
+      "eval_loss": 1.0134038925170898,
+      "eval_mean_token_accuracy": 0.8286508951294288,
+      "eval_num_tokens": 8661177.0,
+      "eval_runtime": 88.6434,
+      "eval_samples_per_second": 16.019,
+      "eval_steps_per_second": 2.008,
+      "step": 3740
+    },
+    {
+      "entropy": 0.08883201805874705,
+      "epoch": 9.060350030175016,
+      "grad_norm": 0.23120898008346558,
+      "learning_rate": 1.008353806970132e-05,
+      "loss": 0.06326142549514771,
+      "mean_token_accuracy": 0.9780020996928215,
+      "num_tokens": 8707602.0,
+      "step": 3760
+    },
+    {
+      "epoch": 9.060350030175016,
+      "eval_entropy": 0.23416905331142832,
+      "eval_loss": 1.0284100770950317,
+      "eval_mean_token_accuracy": 0.8284083213029283,
+      "eval_num_tokens": 8707602.0,
+      "eval_runtime": 88.2933,
+      "eval_samples_per_second": 16.083,
+      "eval_steps_per_second": 2.016,
+      "step": 3760
+    },
+    {
+      "entropy": 0.08861468276008963,
+      "epoch": 9.108630054315027,
+      "grad_norm": 0.17277656495571136,
+      "learning_rate": 9.086531449813489e-06,
+      "loss": 0.06727416515350342,
+      "mean_token_accuracy": 0.9760390736162663,
+      "num_tokens": 8751287.0,
+      "step": 3780
+    },
+    {
+      "epoch": 9.108630054315027,
+      "eval_entropy": 0.23370317386442355,
+      "eval_loss": 1.0381091833114624,
+      "eval_mean_token_accuracy": 0.8282668178670862,
+      "eval_num_tokens": 8751287.0,
+      "eval_runtime": 88.5738,
+      "eval_samples_per_second": 16.032,
+      "eval_steps_per_second": 2.01,
+      "step": 3780
+    },
+    {
+      "entropy": 0.09032954541034996,
+      "epoch": 9.15691007845504,
+      "grad_norm": 0.19481085240840912,
+      "learning_rate": 8.140195796581981e-06,
+      "loss": 0.06117786169052124,
+      "mean_token_accuracy": 0.9764454193413258,
+      "num_tokens": 8799982.0,
+      "step": 3800
+    },
+    {
+      "epoch": 9.15691007845504,
+      "eval_entropy": 0.23215936609868254,
+      "eval_loss": 1.0467767715454102,
+      "eval_mean_token_accuracy": 0.828405868806196,
+      "eval_num_tokens": 8799982.0,
+      "eval_runtime": 87.9551,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 3800
+    },
+    {
+      "entropy": 0.08824449330568314,
+      "epoch": 9.20519010259505,
+      "grad_norm": 0.20382793247699738,
+      "learning_rate": 7.2447989118025455e-06,
+      "loss": 0.06422553658485412,
+      "mean_token_accuracy": 0.9763121999800205,
+      "num_tokens": 8845536.0,
+      "step": 3820
+    },
+    {
+      "epoch": 9.20519010259505,
+      "eval_entropy": 0.23169661472352704,
+      "eval_loss": 1.0529043674468994,
+      "eval_mean_token_accuracy": 0.8282025610463003,
+      "eval_num_tokens": 8845536.0,
+      "eval_runtime": 88.5548,
+      "eval_samples_per_second": 16.035,
+      "eval_steps_per_second": 2.01,
+      "step": 3820
+    },
+    {
+      "entropy": 0.07768499720841646,
+      "epoch": 9.253470126735063,
+      "grad_norm": 0.11727919429540634,
+      "learning_rate": 6.400594182202084e-06,
+      "loss": 0.057279336452484134,
+      "mean_token_accuracy": 0.9795166745781898,
+      "num_tokens": 8896053.0,
+      "step": 3840
+    },
+    {
+      "epoch": 9.253470126735063,
+      "eval_entropy": 0.2301936373114586,
+      "eval_loss": 1.0576049089431763,
+      "eval_mean_token_accuracy": 0.8283417489421502,
+      "eval_num_tokens": 8896053.0,
+      "eval_runtime": 88.7854,
+      "eval_samples_per_second": 15.994,
+      "eval_steps_per_second": 2.005,
+      "step": 3840
+    },
+    {
+      "entropy": 0.09015200780704617,
+      "epoch": 9.301750150875076,
+      "grad_norm": 0.2089058756828308,
+      "learning_rate": 5.6078205077330705e-06,
+      "loss": 0.06626194715499878,
+      "mean_token_accuracy": 0.9747506938874722,
+      "num_tokens": 8940591.0,
+      "step": 3860
+    },
+    {
+      "epoch": 9.301750150875076,
+      "eval_entropy": 0.23024900676159377,
+      "eval_loss": 1.0599786043167114,
+      "eval_mean_token_accuracy": 0.8283051915383071,
+      "eval_num_tokens": 8940591.0,
+      "eval_runtime": 88.2963,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3860
+    },
+    {
+      "entropy": 0.08484094133600592,
+      "epoch": 9.350030175015087,
+      "grad_norm": 0.21646778285503387,
+      "learning_rate": 4.866702233967785e-06,
+      "loss": 0.05998039245605469,
+      "mean_token_accuracy": 0.9771868668496608,
+      "num_tokens": 8989035.0,
+      "step": 3880
+    },
+    {
+      "epoch": 9.350030175015087,
+      "eval_entropy": 0.2300723305578982,
+      "eval_loss": 1.0633587837219238,
+      "eval_mean_token_accuracy": 0.828371500701047,
+      "eval_num_tokens": 8989035.0,
+      "eval_runtime": 88.0233,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 3880
+    },
+    {
+      "entropy": 0.08905019680969417,
+      "epoch": 9.3983101991551,
+      "grad_norm": 0.21842406690120697,
+      "learning_rate": 4.177449088611095e-06,
+      "loss": 0.06666865348815917,
+      "mean_token_accuracy": 0.9753445640206337,
+      "num_tokens": 9033383.0,
+      "step": 3900
+    },
+    {
+      "epoch": 9.3983101991551,
+      "eval_entropy": 0.2295504841911659,
+      "eval_loss": 1.0662461519241333,
+      "eval_mean_token_accuracy": 0.8284122297603093,
+      "eval_num_tokens": 9033383.0,
+      "eval_runtime": 88.5525,
+      "eval_samples_per_second": 16.036,
+      "eval_steps_per_second": 2.01,
+      "step": 3900
+    },
+    {
+      "entropy": 0.08582521872594953,
+      "epoch": 9.446590223295111,
+      "grad_norm": 0.2000240683555603,
+      "learning_rate": 3.540256122150088e-06,
+      "loss": 0.06327944993972778,
+      "mean_token_accuracy": 0.9765432856976985,
+      "num_tokens": 9079672.0,
+      "step": 3920
+    },
+    {
+      "epoch": 9.446590223295111,
+      "eval_entropy": 0.2297678625818049,
+      "eval_loss": 1.0676169395446777,
+      "eval_mean_token_accuracy": 0.8283703879693921,
+      "eval_num_tokens": 9079672.0,
+      "eval_runtime": 87.6897,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 3920
+    },
+    {
+      "entropy": 0.09261773955076932,
+      "epoch": 9.494870247435124,
+      "grad_norm": 0.19104281067848206,
+      "learning_rate": 2.955303652656991e-06,
+      "loss": 0.06111966967582703,
+      "mean_token_accuracy": 0.9738784030079841,
+      "num_tokens": 9127647.0,
+      "step": 3940
+    },
+    {
+      "epoch": 9.494870247435124,
+      "eval_entropy": 0.22986975881490815,
+      "eval_loss": 1.0687925815582275,
+      "eval_mean_token_accuracy": 0.8282996858103892,
+      "eval_num_tokens": 9127647.0,
+      "eval_runtime": 88.2224,
+      "eval_samples_per_second": 16.096,
+      "eval_steps_per_second": 2.018,
+      "step": 3940
+    },
+    {
+      "entropy": 0.08347254800610245,
+      "epoch": 9.543150271575136,
+      "grad_norm": 0.18671369552612305,
+      "learning_rate": 2.422757214761356e-06,
+      "loss": 0.06384263038635254,
+      "mean_token_accuracy": 0.9770128890872002,
+      "num_tokens": 9172587.0,
+      "step": 3960
+    },
+    {
+      "epoch": 9.543150271575136,
+      "eval_entropy": 0.22975517851248217,
+      "eval_loss": 1.0701013803482056,
+      "eval_mean_token_accuracy": 0.8282244285840666,
+      "eval_num_tokens": 9172587.0,
+      "eval_runtime": 88.5794,
+      "eval_samples_per_second": 16.031,
+      "eval_steps_per_second": 2.009,
+      "step": 3960
+    },
+    {
+      "entropy": 0.08136763009242713,
+      "epoch": 9.591430295715147,
+      "grad_norm": 0.24811454117298126,
+      "learning_rate": 1.942767512805673e-06,
+      "loss": 0.06044899821281433,
+      "mean_token_accuracy": 0.9776685133576393,
+      "num_tokens": 9219278.0,
+      "step": 3980
+    },
+    {
+      "epoch": 9.591430295715147,
+      "eval_entropy": 0.22938149347064202,
+      "eval_loss": 1.0711463689804077,
+      "eval_mean_token_accuracy": 0.8282721410976367,
+      "eval_num_tokens": 9219278.0,
+      "eval_runtime": 88.4978,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 3980
+    },
+    {
+      "entropy": 0.09432941288687288,
+      "epoch": 9.63971031985516,
+      "grad_norm": 0.30733999609947205,
+      "learning_rate": 1.515470378197889e-06,
+      "loss": 0.0672213613986969,
+      "mean_token_accuracy": 0.9728612303733826,
+      "num_tokens": 9262346.0,
+      "step": 4000
+    },
+    {
+      "epoch": 9.63971031985516,
+      "eval_entropy": 0.22936245958121976,
+      "eval_loss": 1.071343183517456,
+      "eval_mean_token_accuracy": 0.8282827025049189,
+      "eval_num_tokens": 9262346.0,
+      "eval_runtime": 88.4428,
+      "eval_samples_per_second": 16.056,
+      "eval_steps_per_second": 2.013,
+      "step": 4000
+    },
+    {
+      "entropy": 0.08287148931995034,
+      "epoch": 9.687990343995171,
+      "grad_norm": 0.1976361870765686,
+      "learning_rate": 1.1409867309726256e-06,
+      "loss": 0.06318551301956177,
+      "mean_token_accuracy": 0.9770804084837437,
+      "num_tokens": 9309622.0,
+      "step": 4020
+    },
+    {
+      "epoch": 9.687990343995171,
+      "eval_entropy": 0.22929051920269314,
+      "eval_loss": 1.0711324214935303,
+      "eval_mean_token_accuracy": 0.8281597297513084,
+      "eval_num_tokens": 9309622.0,
+      "eval_runtime": 88.5243,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 4020
+    },
+    {
+      "entropy": 0.0907150611979887,
+      "epoch": 9.736270368135184,
+      "grad_norm": 0.16192017495632172,
+      "learning_rate": 8.194225455723828e-07,
+      "loss": 0.06589602828025817,
+      "mean_token_accuracy": 0.9738550461828709,
+      "num_tokens": 9354549.0,
+      "step": 4040
+    },
+    {
+      "epoch": 9.736270368135184,
+      "eval_entropy": 0.22913936391640244,
+      "eval_loss": 1.0715173482894897,
+      "eval_mean_token_accuracy": 0.8281301544623428,
+      "eval_num_tokens": 9354549.0,
+      "eval_runtime": 88.875,
+      "eval_samples_per_second": 15.978,
+      "eval_steps_per_second": 2.003,
+      "step": 4040
+    },
+    {
+      "entropy": 0.09003224167972804,
+      "epoch": 9.784550392275197,
+      "grad_norm": 0.2092711478471756,
+      "learning_rate": 5.508688208578523e-07,
+      "loss": 0.06655729413032532,
+      "mean_token_accuracy": 0.9750426560640335,
+      "num_tokens": 9399394.0,
+      "step": 4060
+    },
+    {
+      "epoch": 9.784550392275197,
+      "eval_entropy": 0.2291239777140403,
+      "eval_loss": 1.0715563297271729,
+      "eval_mean_token_accuracy": 0.8281808595978812,
+      "eval_num_tokens": 9399394.0,
+      "eval_runtime": 88.4328,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 4060
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.9741849322224845e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4080/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4080/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4080/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4080/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4080/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4080/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4080/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4080/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4080/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4080/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4080/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4080/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4080/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4080/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..58147bdf060a374b5723e250d839deedcd64b1c2
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4080/trainer_state.json
@@ -0,0 +1,4318 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 9.832830416415208,
+  "eval_steps": 20,
+  "global_step": 4080,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    },
+    {
+      "entropy": 0.1326121488586068,
+      "epoch": 7.181050090525045,
+      "grad_norm": 0.4843117296695709,
+      "learning_rate": 8.411870386507724e-05,
+      "loss": 0.09537842869758606,
+      "mean_token_accuracy": 0.9645151920616627,
+      "num_tokens": 6898444.0,
+      "step": 2980
+    },
+    {
+      "epoch": 7.181050090525045,
+      "eval_entropy": 0.27999477783280813,
+      "eval_loss": 0.9031037092208862,
+      "eval_mean_token_accuracy": 0.8261882204018282,
+      "eval_num_tokens": 6898444.0,
+      "eval_runtime": 88.5602,
+      "eval_samples_per_second": 16.034,
+      "eval_steps_per_second": 2.01,
+      "step": 2980
+    },
+    {
+      "entropy": 0.1156751093454659,
+      "epoch": 7.229330114665057,
+      "grad_norm": 0.5297922492027283,
+      "learning_rate": 8.149629351430326e-05,
+      "loss": 0.09203402996063233,
+      "mean_token_accuracy": 0.9689733698964119,
+      "num_tokens": 6947106.0,
+      "step": 3000
+    },
+    {
+      "epoch": 7.229330114665057,
+      "eval_entropy": 0.2740225845508361,
+      "eval_loss": 0.8876440525054932,
+      "eval_mean_token_accuracy": 0.8265116857678703,
+      "eval_num_tokens": 6947106.0,
+      "eval_runtime": 87.7971,
+      "eval_samples_per_second": 16.174,
+      "eval_steps_per_second": 2.027,
+      "step": 3000
+    },
+    {
+      "entropy": 0.12116114404052496,
+      "epoch": 7.27761013880507,
+      "grad_norm": 0.5467398166656494,
+      "learning_rate": 7.890406302404618e-05,
+      "loss": 0.0933056652545929,
+      "mean_token_accuracy": 0.9676455594599247,
+      "num_tokens": 6992784.0,
+      "step": 3020
+    },
+    {
+      "epoch": 7.27761013880507,
+      "eval_entropy": 0.27238342793804876,
+      "eval_loss": 0.9166117310523987,
+      "eval_mean_token_accuracy": 0.825499015242866,
+      "eval_num_tokens": 6992784.0,
+      "eval_runtime": 88.3276,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 3020
+    },
+    {
+      "entropy": 0.11715119928121567,
+      "epoch": 7.325890162945082,
+      "grad_norm": 0.4976702332496643,
+      "learning_rate": 7.634274596487176e-05,
+      "loss": 0.09449659585952759,
+      "mean_token_accuracy": 0.966506478190422,
+      "num_tokens": 7040482.0,
+      "step": 3040
+    },
+    {
+      "epoch": 7.325890162945082,
+      "eval_entropy": 0.2788975556077582,
+      "eval_loss": 0.8833065629005432,
+      "eval_mean_token_accuracy": 0.8269920596915684,
+      "eval_num_tokens": 7040482.0,
+      "eval_runtime": 88.9016,
+      "eval_samples_per_second": 15.973,
+      "eval_steps_per_second": 2.002,
+      "step": 3040
+    },
+    {
+      "entropy": 0.12565911412239075,
+      "epoch": 7.3741701870850935,
+      "grad_norm": 0.5728679895401001,
+      "learning_rate": 7.381306715921025e-05,
+      "loss": 0.09626244902610778,
+      "mean_token_accuracy": 0.9661444149911403,
+      "num_tokens": 7086261.0,
+      "step": 3060
+    },
+    {
+      "epoch": 7.3741701870850935,
+      "eval_entropy": 0.2762492011939542,
+      "eval_loss": 0.8961160778999329,
+      "eval_mean_token_accuracy": 0.826369910092836,
+      "eval_num_tokens": 7086261.0,
+      "eval_runtime": 88.5204,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 3060
+    },
+    {
+      "entropy": 0.12258567921817302,
+      "epoch": 7.422450211225105,
+      "grad_norm": 0.468985915184021,
+      "learning_rate": 7.131574247624031e-05,
+      "loss": 0.09120266437530518,
+      "mean_token_accuracy": 0.9674938447773457,
+      "num_tokens": 7133247.0,
+      "step": 3080
+    },
+    {
+      "epoch": 7.422450211225105,
+      "eval_entropy": 0.2688512086366,
+      "eval_loss": 0.9048654437065125,
+      "eval_mean_token_accuracy": 0.8265582302982888,
+      "eval_num_tokens": 7133247.0,
+      "eval_runtime": 88.164,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 3080
+    },
+    {
+      "entropy": 0.12070536343380808,
+      "epoch": 7.470730235365117,
+      "grad_norm": 0.5310841798782349,
+      "learning_rate": 6.885147862930688e-05,
+      "loss": 0.09386247992515565,
+      "mean_token_accuracy": 0.96664222702384,
+      "num_tokens": 7177810.0,
+      "step": 3100
+    },
+    {
+      "epoch": 7.470730235365117,
+      "eval_entropy": 0.26289632874593305,
+      "eval_loss": 0.902837336063385,
+      "eval_mean_token_accuracy": 0.828052184554968,
+      "eval_num_tokens": 7177810.0,
+      "eval_runtime": 88.1201,
+      "eval_samples_per_second": 16.114,
+      "eval_steps_per_second": 2.02,
+      "step": 3100
+    },
+    {
+      "entropy": 0.11322665181942285,
+      "epoch": 7.51901025950513,
+      "grad_norm": 0.6846067905426025,
+      "learning_rate": 6.642097297592908e-05,
+      "loss": 0.08772538304328918,
+      "mean_token_accuracy": 0.9691362135112286,
+      "num_tokens": 7225302.0,
+      "step": 3120
+    },
+    {
+      "epoch": 7.51901025950513,
+      "eval_entropy": 0.2670605613776807,
+      "eval_loss": 0.9130566120147705,
+      "eval_mean_token_accuracy": 0.826366903741708,
+      "eval_num_tokens": 7225302.0,
+      "eval_runtime": 88.6035,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3120
+    },
+    {
+      "entropy": 0.10985926580615342,
+      "epoch": 7.567290283645142,
+      "grad_norm": 0.4799748361110687,
+      "learning_rate": 6.402491332045683e-05,
+      "loss": 0.08672296404838561,
+      "mean_token_accuracy": 0.9698824048042297,
+      "num_tokens": 7274881.0,
+      "step": 3140
+    },
+    {
+      "epoch": 7.567290283645142,
+      "eval_entropy": 0.26657010866015146,
+      "eval_loss": 0.9056380987167358,
+      "eval_mean_token_accuracy": 0.8272326461384806,
+      "eval_num_tokens": 7274881.0,
+      "eval_runtime": 88.2967,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3140
+    },
+    {
+      "entropy": 0.11890473761595785,
+      "epoch": 7.615570307785154,
+      "grad_norm": 0.4081912934780121,
+      "learning_rate": 6.166397771942987e-05,
+      "loss": 0.09396827220916748,
+      "mean_token_accuracy": 0.967512384802103,
+      "num_tokens": 7321163.0,
+      "step": 3160
+    },
+    {
+      "epoch": 7.615570307785154,
+      "eval_entropy": 0.26364776575833226,
+      "eval_loss": 0.9204733967781067,
+      "eval_mean_token_accuracy": 0.827419729714983,
+      "eval_num_tokens": 7321163.0,
+      "eval_runtime": 88.3441,
+      "eval_samples_per_second": 16.074,
+      "eval_steps_per_second": 2.015,
+      "step": 3160
+    },
+    {
+      "entropy": 0.12209134344011545,
+      "epoch": 7.663850331925166,
+      "grad_norm": 0.5159774422645569,
+      "learning_rate": 5.93388342896962e-05,
+      "loss": 0.09292978644371033,
+      "mean_token_accuracy": 0.9658421002328396,
+      "num_tokens": 7366217.0,
+      "step": 3180
+    },
+    {
+      "epoch": 7.663850331925166,
+      "eval_entropy": 0.2714110068725736,
+      "eval_loss": 0.9143092036247253,
+      "eval_mean_token_accuracy": 0.8262251099843657,
+      "eval_num_tokens": 7366217.0,
+      "eval_runtime": 87.9635,
+      "eval_samples_per_second": 16.143,
+      "eval_steps_per_second": 2.024,
+      "step": 3180
+    },
+    {
+      "entropy": 0.11955609126016498,
+      "epoch": 7.712130356065178,
+      "grad_norm": 0.7005976438522339,
+      "learning_rate": 5.705014101934252e-05,
+      "loss": 0.09319776892662049,
+      "mean_token_accuracy": 0.9674513377249241,
+      "num_tokens": 7410887.0,
+      "step": 3200
+    },
+    {
+      "epoch": 7.712130356065178,
+      "eval_entropy": 0.2626291513442993,
+      "eval_loss": 0.91884446144104,
+      "eval_mean_token_accuracy": 0.827563955207889,
+      "eval_num_tokens": 7410887.0,
+      "eval_runtime": 87.8849,
+      "eval_samples_per_second": 16.157,
+      "eval_steps_per_second": 2.025,
+      "step": 3200
+    },
+    {
+      "entropy": 0.11767532899975777,
+      "epoch": 7.76041038020519,
+      "grad_norm": 0.4675362706184387,
+      "learning_rate": 5.479854558149187e-05,
+      "loss": 0.09038087725639343,
+      "mean_token_accuracy": 0.9669680051505566,
+      "num_tokens": 7457578.0,
+      "step": 3220
+    },
+    {
+      "epoch": 7.76041038020519,
+      "eval_entropy": 0.26581037203582486,
+      "eval_loss": 0.9146883487701416,
+      "eval_mean_token_accuracy": 0.8270345812433222,
+      "eval_num_tokens": 7457578.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 3220
+    },
+    {
+      "entropy": 0.11383061073720455,
+      "epoch": 7.808690404345202,
+      "grad_norm": 0.541247546672821,
+      "learning_rate": 5.258468515101918e-05,
+      "loss": 0.08942405581474304,
+      "mean_token_accuracy": 0.968398705124855,
+      "num_tokens": 7505761.0,
+      "step": 3240
+    },
+    {
+      "epoch": 7.808690404345202,
+      "eval_entropy": 0.26687998960861997,
+      "eval_loss": 0.9037085175514221,
+      "eval_mean_token_accuracy": 0.8274977863504646,
+      "eval_num_tokens": 7505761.0,
+      "eval_runtime": 88.5163,
+      "eval_samples_per_second": 16.042,
+      "eval_steps_per_second": 2.011,
+      "step": 3240
+    },
+    {
+      "entropy": 0.12004058873280883,
+      "epoch": 7.856970428485214,
+      "grad_norm": 0.5289552211761475,
+      "learning_rate": 5.040918622423899e-05,
+      "loss": 0.09064805507659912,
+      "mean_token_accuracy": 0.9673330299556255,
+      "num_tokens": 7552557.0,
+      "step": 3260
+    },
+    {
+      "epoch": 7.856970428485214,
+      "eval_entropy": 0.26288379251622085,
+      "eval_loss": 0.9146178364753723,
+      "eval_mean_token_accuracy": 0.8276662448149049,
+      "eval_num_tokens": 7552557.0,
+      "eval_runtime": 88.0723,
+      "eval_samples_per_second": 16.123,
+      "eval_steps_per_second": 2.021,
+      "step": 3260
+    },
+    {
+      "entropy": 0.1180689456872642,
+      "epoch": 7.905250452625227,
+      "grad_norm": 0.4754861891269684,
+      "learning_rate": 4.827266444161377e-05,
+      "loss": 0.09402655363082886,
+      "mean_token_accuracy": 0.9663361787796021,
+      "num_tokens": 7596998.0,
+      "step": 3280
+    },
+    {
+      "epoch": 7.905250452625227,
+      "eval_entropy": 0.261690996922134,
+      "eval_loss": 0.9120186567306519,
+      "eval_mean_token_accuracy": 0.8278995015647974,
+      "eval_num_tokens": 7596998.0,
+      "eval_runtime": 88.6573,
+      "eval_samples_per_second": 16.017,
+      "eval_steps_per_second": 2.008,
+      "step": 3280
+    },
+    {
+      "entropy": 0.1162164019420743,
+      "epoch": 7.953530476765239,
+      "grad_norm": 0.6203398108482361,
+      "learning_rate": 4.61757244135355e-05,
+      "loss": 0.08832768201828003,
+      "mean_token_accuracy": 0.9694220684468746,
+      "num_tokens": 7643288.0,
+      "step": 3300
+    },
+    {
+      "epoch": 7.953530476765239,
+      "eval_entropy": 0.26158678129817664,
+      "eval_loss": 0.9301361441612244,
+      "eval_mean_token_accuracy": 0.8269603885291668,
+      "eval_num_tokens": 7643288.0,
+      "eval_runtime": 88.7064,
+      "eval_samples_per_second": 16.008,
+      "eval_steps_per_second": 2.007,
+      "step": 3300
+    },
+    {
+      "entropy": 0.10877202204489089,
+      "epoch": 8.0,
+      "grad_norm": 3.058673143386841,
+      "learning_rate": 4.411895954922763e-05,
+      "loss": 0.09159615635871887,
+      "mean_token_accuracy": 0.9686878834451947,
+      "num_tokens": 7688608.0,
+      "step": 3320
+    },
+    {
+      "epoch": 8.0,
+      "eval_entropy": 0.2590438057197614,
+      "eval_loss": 0.9262065887451172,
+      "eval_mean_token_accuracy": 0.8277404546067956,
+      "eval_num_tokens": 7688608.0,
+      "eval_runtime": 87.7557,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 3320
+    },
+    {
+      "entropy": 0.09969092034734786,
+      "epoch": 8.048280024140013,
+      "grad_norm": 0.2561911642551422,
+      "learning_rate": 4.2102951888817644e-05,
+      "loss": 0.07228735685348511,
+      "mean_token_accuracy": 0.9743556551635265,
+      "num_tokens": 7732856.0,
+      "step": 3340
+    },
+    {
+      "epoch": 8.048280024140013,
+      "eval_entropy": 0.2462534454933713,
+      "eval_loss": 0.9755983352661133,
+      "eval_mean_token_accuracy": 0.827510209565752,
+      "eval_num_tokens": 7732856.0,
+      "eval_runtime": 88.0297,
+      "eval_samples_per_second": 16.131,
+      "eval_steps_per_second": 2.022,
+      "step": 3340
+    },
+    {
+      "entropy": 0.09477624730207027,
+      "epoch": 8.096560048280024,
+      "grad_norm": 0.3670356571674347,
+      "learning_rate": 4.012827193862633e-05,
+      "loss": 0.0686036467552185,
+      "mean_token_accuracy": 0.9743670664727688,
+      "num_tokens": 7778585.0,
+      "step": 3360
+    },
+    {
+      "epoch": 8.096560048280024,
+      "eval_entropy": 0.2440882951356052,
+      "eval_loss": 1.0118423700332642,
+      "eval_mean_token_accuracy": 0.8271241070849172,
+      "eval_num_tokens": 7778585.0,
+      "eval_runtime": 88.1364,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 3360
+    },
+    {
+      "entropy": 0.09417980071157217,
+      "epoch": 8.144840072420036,
+      "grad_norm": 0.5785111784934998,
+      "learning_rate": 3.819547850972133e-05,
+      "loss": 0.07100698947906495,
+      "mean_token_accuracy": 0.9737804099917412,
+      "num_tokens": 7824287.0,
+      "step": 3380
+    },
+    {
+      "epoch": 8.144840072420036,
+      "eval_entropy": 0.24603172860453637,
+      "eval_loss": 1.0021991729736328,
+      "eval_mean_token_accuracy": 0.827162122793412,
+      "eval_num_tokens": 7824287.0,
+      "eval_runtime": 88.4534,
+      "eval_samples_per_second": 16.054,
+      "eval_steps_per_second": 2.012,
+      "step": 3380
+    },
+    {
+      "entropy": 0.087222247896716,
+      "epoch": 8.193120096560047,
+      "grad_norm": 0.44848504662513733,
+      "learning_rate": 3.6305118559780393e-05,
+      "loss": 0.06325234174728393,
+      "mean_token_accuracy": 0.9769335232675076,
+      "num_tokens": 7874784.0,
+      "step": 3400
+    },
+    {
+      "epoch": 8.193120096560047,
+      "eval_entropy": 0.24544863905129807,
+      "eval_loss": 1.005008339881897,
+      "eval_mean_token_accuracy": 0.8273035240976998,
+      "eval_num_tokens": 7874784.0,
+      "eval_runtime": 88.6737,
+      "eval_samples_per_second": 16.014,
+      "eval_steps_per_second": 2.007,
+      "step": 3400
+    },
+    {
+      "entropy": 0.09037275062873959,
+      "epoch": 8.24140012070006,
+      "grad_norm": 0.23529155552387238,
+      "learning_rate": 3.4457727038308435e-05,
+      "loss": 0.06589569449424744,
+      "mean_token_accuracy": 0.9751470290124417,
+      "num_tokens": 7925240.0,
+      "step": 3420
+    },
+    {
+      "epoch": 8.24140012070006,
+      "eval_entropy": 0.24203423494368456,
+      "eval_loss": 1.0052824020385742,
+      "eval_mean_token_accuracy": 0.8275575503874361,
+      "eval_num_tokens": 7925240.0,
+      "eval_runtime": 88.3486,
+      "eval_samples_per_second": 16.073,
+      "eval_steps_per_second": 2.015,
+      "step": 3420
+    },
+    {
+      "entropy": 0.10373208015225828,
+      "epoch": 8.289680144840073,
+      "grad_norm": 0.24796408414840698,
+      "learning_rate": 3.2653826735253594e-05,
+      "loss": 0.07714771628379821,
+      "mean_token_accuracy": 0.9722891598939896,
+      "num_tokens": 7966685.0,
+      "step": 3440
+    },
+    {
+      "epoch": 8.289680144840073,
+      "eval_entropy": 0.23461157050025597,
+      "eval_loss": 1.012108325958252,
+      "eval_mean_token_accuracy": 0.82825900564033,
+      "eval_num_tokens": 7966685.0,
+      "eval_runtime": 88.1018,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 3440
+    },
+    {
+      "entropy": 0.08967954274266958,
+      "epoch": 8.337960168980084,
+      "grad_norm": 0.5314674973487854,
+      "learning_rate": 3.08939281330635e-05,
+      "loss": 0.06746266484260559,
+      "mean_token_accuracy": 0.9753133699297905,
+      "num_tokens": 8013656.0,
+      "step": 3460
+    },
+    {
+      "epoch": 8.337960168980084,
+      "eval_entropy": 0.23781998589467468,
+      "eval_loss": 1.0119913816452026,
+      "eval_mean_token_accuracy": 0.8273287157664139,
+      "eval_num_tokens": 8013656.0,
+      "eval_runtime": 88.6038,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3460
+    },
+    {
+      "entropy": 0.09280927330255509,
+      "epoch": 8.386240193120097,
+      "grad_norm": 0.30658936500549316,
+      "learning_rate": 2.9178529262224946e-05,
+      "loss": 0.06936421990394592,
+      "mean_token_accuracy": 0.9742105580866337,
+      "num_tokens": 8061108.0,
+      "step": 3480
+    },
+    {
+      "epoch": 8.386240193120097,
+      "eval_entropy": 0.23672426734747512,
+      "eval_loss": 1.0162041187286377,
+      "eval_mean_token_accuracy": 0.8278394281194451,
+      "eval_num_tokens": 8061108.0,
+      "eval_runtime": 88.6015,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 3480
+    },
+    {
+      "entropy": 0.09195122267119586,
+      "epoch": 8.43452021726011,
+      "grad_norm": 0.24512454867362976,
+      "learning_rate": 2.7508115560327256e-05,
+      "loss": 0.06996064782142639,
+      "mean_token_accuracy": 0.974104854464531,
+      "num_tokens": 8106924.0,
+      "step": 3500
+    },
+    {
+      "epoch": 8.43452021726011,
+      "eval_entropy": 0.23758245828781235,
+      "eval_loss": 1.0157747268676758,
+      "eval_mean_token_accuracy": 0.8278501579600773,
+      "eval_num_tokens": 8106924.0,
+      "eval_runtime": 88.5678,
+      "eval_samples_per_second": 16.033,
+      "eval_steps_per_second": 2.01,
+      "step": 3500
+    },
+    {
+      "entropy": 0.09518477590754629,
+      "epoch": 8.48280024140012,
+      "grad_norm": 0.2373228371143341,
+      "learning_rate": 2.5883159734688868e-05,
+      "loss": 0.06832797527313232,
+      "mean_token_accuracy": 0.9743806682527065,
+      "num_tokens": 8153070.0,
+      "step": 3520
+    },
+    {
+      "epoch": 8.48280024140012,
+      "eval_entropy": 0.2379308074210467,
+      "eval_loss": 1.0180530548095703,
+      "eval_mean_token_accuracy": 0.8282477052024241,
+      "eval_num_tokens": 8153070.0,
+      "eval_runtime": 88.4651,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 3520
+    },
+    {
+      "entropy": 0.0906269280705601,
+      "epoch": 8.531080265540133,
+      "grad_norm": 0.37602657079696655,
+      "learning_rate": 2.4304121628587173e-05,
+      "loss": 0.06921588778495788,
+      "mean_token_accuracy": 0.9745571479201317,
+      "num_tokens": 8200066.0,
+      "step": 3540
+    },
+    {
+      "epoch": 8.531080265540133,
+      "eval_entropy": 0.2357490769263064,
+      "eval_loss": 1.017139196395874,
+      "eval_mean_token_accuracy": 0.828713178299786,
+      "eval_num_tokens": 8200066.0,
+      "eval_runtime": 88.6424,
+      "eval_samples_per_second": 16.019,
+      "eval_steps_per_second": 2.008,
+      "step": 3540
+    },
+    {
+      "entropy": 0.09273334196768701,
+      "epoch": 8.579360289680144,
+      "grad_norm": 0.2978838086128235,
+      "learning_rate": 2.277144809112794e-05,
+      "loss": 0.06970887780189514,
+      "mean_token_accuracy": 0.9751374177634716,
+      "num_tokens": 8246157.0,
+      "step": 3560
+    },
+    {
+      "epoch": 8.579360289680144,
+      "eval_entropy": 0.23590933138065123,
+      "eval_loss": 1.019231915473938,
+      "eval_mean_token_accuracy": 0.8279163412163767,
+      "eval_num_tokens": 8246157.0,
+      "eval_runtime": 88.2691,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 3560
+    },
+    {
+      "entropy": 0.09854201921261847,
+      "epoch": 8.627640313820157,
+      "grad_norm": 0.5381243228912354,
+      "learning_rate": 2.1285572850792413e-05,
+      "loss": 0.07123501300811767,
+      "mean_token_accuracy": 0.9723988175392151,
+      "num_tokens": 8291752.0,
+      "step": 3580
+    },
+    {
+      "epoch": 8.627640313820157,
+      "eval_entropy": 0.23592024949494372,
+      "eval_loss": 1.020250678062439,
+      "eval_mean_token_accuracy": 0.8282466226079491,
+      "eval_num_tokens": 8291752.0,
+      "eval_runtime": 87.8939,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 3580
+    },
+    {
+      "entropy": 0.09092000788077712,
+      "epoch": 8.67592033796017,
+      "grad_norm": 0.29089561104774475,
+      "learning_rate": 1.9846916392697493e-05,
+      "loss": 0.06982720494270325,
+      "mean_token_accuracy": 0.9749995484948158,
+      "num_tokens": 8337651.0,
+      "step": 3600
+    },
+    {
+      "epoch": 8.67592033796017,
+      "eval_entropy": 0.23658913382318583,
+      "eval_loss": 1.019862413406372,
+      "eval_mean_token_accuracy": 0.8280127842104836,
+      "eval_num_tokens": 8337651.0,
+      "eval_runtime": 88.2855,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 3600
+    },
+    {
+      "entropy": 0.08708424470387399,
+      "epoch": 8.72420036210018,
+      "grad_norm": 0.29904839396476746,
+      "learning_rate": 1.8455885839603106e-05,
+      "loss": 0.06801370978355407,
+      "mean_token_accuracy": 0.9755801670253277,
+      "num_tokens": 8385330.0,
+      "step": 3620
+    },
+    {
+      "epoch": 8.72420036210018,
+      "eval_entropy": 0.23627035771862845,
+      "eval_loss": 1.0204569101333618,
+      "eval_mean_token_accuracy": 0.8279967894045155,
+      "eval_num_tokens": 8385330.0,
+      "eval_runtime": 88.1753,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 3620
+    },
+    {
+      "entropy": 0.0916255566291511,
+      "epoch": 8.772480386240193,
+      "grad_norm": 0.20571033656597137,
+      "learning_rate": 1.7112874836701502e-05,
+      "loss": 0.0692249894142151,
+      "mean_token_accuracy": 0.9745148062705994,
+      "num_tokens": 8432134.0,
+      "step": 3640
+    },
+    {
+      "epoch": 8.772480386240193,
+      "eval_entropy": 0.2340821772813797,
+      "eval_loss": 1.0212821960449219,
+      "eval_mean_token_accuracy": 0.828373622358515,
+      "eval_num_tokens": 8432134.0,
+      "eval_runtime": 87.2518,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 3640
+    },
+    {
+      "entropy": 0.0926868049427867,
+      "epoch": 8.820760410380204,
+      "grad_norm": 0.2618265450000763,
+      "learning_rate": 1.5818263440220203e-05,
+      "loss": 0.07338647246360779,
+      "mean_token_accuracy": 0.9724708944559097,
+      "num_tokens": 8476758.0,
+      "step": 3660
+    },
+    {
+      "epoch": 8.820760410380204,
+      "eval_entropy": 0.23637074575330433,
+      "eval_loss": 1.0186725854873657,
+      "eval_mean_token_accuracy": 0.8277707605549459,
+      "eval_num_tokens": 8476758.0,
+      "eval_runtime": 88.4695,
+      "eval_samples_per_second": 16.051,
+      "eval_steps_per_second": 2.012,
+      "step": 3660
+    },
+    {
+      "entropy": 0.09175782897509635,
+      "epoch": 8.869040434520217,
+      "grad_norm": 0.2889852821826935,
+      "learning_rate": 1.4572418009870604e-05,
+      "loss": 0.07066041231155396,
+      "mean_token_accuracy": 0.9725673109292984,
+      "num_tokens": 8523294.0,
+      "step": 3680
+    },
+    {
+      "epoch": 8.869040434520217,
+      "eval_entropy": 0.23945164404223473,
+      "eval_loss": 1.0145231485366821,
+      "eval_mean_token_accuracy": 0.8281202523895864,
+      "eval_num_tokens": 8523294.0,
+      "eval_runtime": 88.368,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 3680
+    },
+    {
+      "entropy": 0.09265611860901117,
+      "epoch": 8.91732045866023,
+      "grad_norm": 0.41092950105667114,
+      "learning_rate": 1.3375691105172625e-05,
+      "loss": 0.06975027322769164,
+      "mean_token_accuracy": 0.974636297672987,
+      "num_tokens": 8570358.0,
+      "step": 3700
+    },
+    {
+      "epoch": 8.91732045866023,
+      "eval_entropy": 0.239588919901446,
+      "eval_loss": 1.014059066772461,
+      "eval_mean_token_accuracy": 0.828028041660116,
+      "eval_num_tokens": 8570358.0,
+      "eval_runtime": 88.1039,
+      "eval_samples_per_second": 16.117,
+      "eval_steps_per_second": 2.02,
+      "step": 3700
+    },
+    {
+      "entropy": 0.0896465003490448,
+      "epoch": 8.965600482800241,
+      "grad_norm": 0.24000254273414612,
+      "learning_rate": 1.2228421385684305e-05,
+      "loss": 0.06743953824043274,
+      "mean_token_accuracy": 0.974006575345993,
+      "num_tokens": 8618246.0,
+      "step": 3720
+    },
+    {
+      "epoch": 8.965600482800241,
+      "eval_entropy": 0.2390461369846644,
+      "eval_loss": 1.011997938156128,
+      "eval_mean_token_accuracy": 0.8284533211354459,
+      "eval_num_tokens": 8618246.0,
+      "eval_runtime": 88.6202,
+      "eval_samples_per_second": 16.023,
+      "eval_steps_per_second": 2.009,
+      "step": 3720
+    },
+    {
+      "entropy": 0.09575109251520851,
+      "epoch": 9.012070006035003,
+      "grad_norm": 0.18088576197624207,
+      "learning_rate": 1.1130933515165429e-05,
+      "loss": 0.07295305132865906,
+      "mean_token_accuracy": 0.9739722857227573,
+      "num_tokens": 8661177.0,
+      "step": 3740
+    },
+    {
+      "epoch": 9.012070006035003,
+      "eval_entropy": 0.23728882387447892,
+      "eval_loss": 1.0134038925170898,
+      "eval_mean_token_accuracy": 0.8286508951294288,
+      "eval_num_tokens": 8661177.0,
+      "eval_runtime": 88.6434,
+      "eval_samples_per_second": 16.019,
+      "eval_steps_per_second": 2.008,
+      "step": 3740
+    },
+    {
+      "entropy": 0.08883201805874705,
+      "epoch": 9.060350030175016,
+      "grad_norm": 0.23120898008346558,
+      "learning_rate": 1.008353806970132e-05,
+      "loss": 0.06326142549514771,
+      "mean_token_accuracy": 0.9780020996928215,
+      "num_tokens": 8707602.0,
+      "step": 3760
+    },
+    {
+      "epoch": 9.060350030175016,
+      "eval_entropy": 0.23416905331142832,
+      "eval_loss": 1.0284100770950317,
+      "eval_mean_token_accuracy": 0.8284083213029283,
+      "eval_num_tokens": 8707602.0,
+      "eval_runtime": 88.2933,
+      "eval_samples_per_second": 16.083,
+      "eval_steps_per_second": 2.016,
+      "step": 3760
+    },
+    {
+      "entropy": 0.08861468276008963,
+      "epoch": 9.108630054315027,
+      "grad_norm": 0.17277656495571136,
+      "learning_rate": 9.086531449813489e-06,
+      "loss": 0.06727416515350342,
+      "mean_token_accuracy": 0.9760390736162663,
+      "num_tokens": 8751287.0,
+      "step": 3780
+    },
+    {
+      "epoch": 9.108630054315027,
+      "eval_entropy": 0.23370317386442355,
+      "eval_loss": 1.0381091833114624,
+      "eval_mean_token_accuracy": 0.8282668178670862,
+      "eval_num_tokens": 8751287.0,
+      "eval_runtime": 88.5738,
+      "eval_samples_per_second": 16.032,
+      "eval_steps_per_second": 2.01,
+      "step": 3780
+    },
+    {
+      "entropy": 0.09032954541034996,
+      "epoch": 9.15691007845504,
+      "grad_norm": 0.19481085240840912,
+      "learning_rate": 8.140195796581981e-06,
+      "loss": 0.06117786169052124,
+      "mean_token_accuracy": 0.9764454193413258,
+      "num_tokens": 8799982.0,
+      "step": 3800
+    },
+    {
+      "epoch": 9.15691007845504,
+      "eval_entropy": 0.23215936609868254,
+      "eval_loss": 1.0467767715454102,
+      "eval_mean_token_accuracy": 0.828405868806196,
+      "eval_num_tokens": 8799982.0,
+      "eval_runtime": 87.9551,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 3800
+    },
+    {
+      "entropy": 0.08824449330568314,
+      "epoch": 9.20519010259505,
+      "grad_norm": 0.20382793247699738,
+      "learning_rate": 7.2447989118025455e-06,
+      "loss": 0.06422553658485412,
+      "mean_token_accuracy": 0.9763121999800205,
+      "num_tokens": 8845536.0,
+      "step": 3820
+    },
+    {
+      "epoch": 9.20519010259505,
+      "eval_entropy": 0.23169661472352704,
+      "eval_loss": 1.0529043674468994,
+      "eval_mean_token_accuracy": 0.8282025610463003,
+      "eval_num_tokens": 8845536.0,
+      "eval_runtime": 88.5548,
+      "eval_samples_per_second": 16.035,
+      "eval_steps_per_second": 2.01,
+      "step": 3820
+    },
+    {
+      "entropy": 0.07768499720841646,
+      "epoch": 9.253470126735063,
+      "grad_norm": 0.11727919429540634,
+      "learning_rate": 6.400594182202084e-06,
+      "loss": 0.057279336452484134,
+      "mean_token_accuracy": 0.9795166745781898,
+      "num_tokens": 8896053.0,
+      "step": 3840
+    },
+    {
+      "epoch": 9.253470126735063,
+      "eval_entropy": 0.2301936373114586,
+      "eval_loss": 1.0576049089431763,
+      "eval_mean_token_accuracy": 0.8283417489421502,
+      "eval_num_tokens": 8896053.0,
+      "eval_runtime": 88.7854,
+      "eval_samples_per_second": 15.994,
+      "eval_steps_per_second": 2.005,
+      "step": 3840
+    },
+    {
+      "entropy": 0.09015200780704617,
+      "epoch": 9.301750150875076,
+      "grad_norm": 0.2089058756828308,
+      "learning_rate": 5.6078205077330705e-06,
+      "loss": 0.06626194715499878,
+      "mean_token_accuracy": 0.9747506938874722,
+      "num_tokens": 8940591.0,
+      "step": 3860
+    },
+    {
+      "epoch": 9.301750150875076,
+      "eval_entropy": 0.23024900676159377,
+      "eval_loss": 1.0599786043167114,
+      "eval_mean_token_accuracy": 0.8283051915383071,
+      "eval_num_tokens": 8940591.0,
+      "eval_runtime": 88.2963,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3860
+    },
+    {
+      "entropy": 0.08484094133600592,
+      "epoch": 9.350030175015087,
+      "grad_norm": 0.21646778285503387,
+      "learning_rate": 4.866702233967785e-06,
+      "loss": 0.05998039245605469,
+      "mean_token_accuracy": 0.9771868668496608,
+      "num_tokens": 8989035.0,
+      "step": 3880
+    },
+    {
+      "epoch": 9.350030175015087,
+      "eval_entropy": 0.2300723305578982,
+      "eval_loss": 1.0633587837219238,
+      "eval_mean_token_accuracy": 0.828371500701047,
+      "eval_num_tokens": 8989035.0,
+      "eval_runtime": 88.0233,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 3880
+    },
+    {
+      "entropy": 0.08905019680969417,
+      "epoch": 9.3983101991551,
+      "grad_norm": 0.21842406690120697,
+      "learning_rate": 4.177449088611095e-06,
+      "loss": 0.06666865348815917,
+      "mean_token_accuracy": 0.9753445640206337,
+      "num_tokens": 9033383.0,
+      "step": 3900
+    },
+    {
+      "epoch": 9.3983101991551,
+      "eval_entropy": 0.2295504841911659,
+      "eval_loss": 1.0662461519241333,
+      "eval_mean_token_accuracy": 0.8284122297603093,
+      "eval_num_tokens": 9033383.0,
+      "eval_runtime": 88.5525,
+      "eval_samples_per_second": 16.036,
+      "eval_steps_per_second": 2.01,
+      "step": 3900
+    },
+    {
+      "entropy": 0.08582521872594953,
+      "epoch": 9.446590223295111,
+      "grad_norm": 0.2000240683555603,
+      "learning_rate": 3.540256122150088e-06,
+      "loss": 0.06327944993972778,
+      "mean_token_accuracy": 0.9765432856976985,
+      "num_tokens": 9079672.0,
+      "step": 3920
+    },
+    {
+      "epoch": 9.446590223295111,
+      "eval_entropy": 0.2297678625818049,
+      "eval_loss": 1.0676169395446777,
+      "eval_mean_token_accuracy": 0.8283703879693921,
+      "eval_num_tokens": 9079672.0,
+      "eval_runtime": 87.6897,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 3920
+    },
+    {
+      "entropy": 0.09261773955076932,
+      "epoch": 9.494870247435124,
+      "grad_norm": 0.19104281067848206,
+      "learning_rate": 2.955303652656991e-06,
+      "loss": 0.06111966967582703,
+      "mean_token_accuracy": 0.9738784030079841,
+      "num_tokens": 9127647.0,
+      "step": 3940
+    },
+    {
+      "epoch": 9.494870247435124,
+      "eval_entropy": 0.22986975881490815,
+      "eval_loss": 1.0687925815582275,
+      "eval_mean_token_accuracy": 0.8282996858103892,
+      "eval_num_tokens": 9127647.0,
+      "eval_runtime": 88.2224,
+      "eval_samples_per_second": 16.096,
+      "eval_steps_per_second": 2.018,
+      "step": 3940
+    },
+    {
+      "entropy": 0.08347254800610245,
+      "epoch": 9.543150271575136,
+      "grad_norm": 0.18671369552612305,
+      "learning_rate": 2.422757214761356e-06,
+      "loss": 0.06384263038635254,
+      "mean_token_accuracy": 0.9770128890872002,
+      "num_tokens": 9172587.0,
+      "step": 3960
+    },
+    {
+      "epoch": 9.543150271575136,
+      "eval_entropy": 0.22975517851248217,
+      "eval_loss": 1.0701013803482056,
+      "eval_mean_token_accuracy": 0.8282244285840666,
+      "eval_num_tokens": 9172587.0,
+      "eval_runtime": 88.5794,
+      "eval_samples_per_second": 16.031,
+      "eval_steps_per_second": 2.009,
+      "step": 3960
+    },
+    {
+      "entropy": 0.08136763009242713,
+      "epoch": 9.591430295715147,
+      "grad_norm": 0.24811454117298126,
+      "learning_rate": 1.942767512805673e-06,
+      "loss": 0.06044899821281433,
+      "mean_token_accuracy": 0.9776685133576393,
+      "num_tokens": 9219278.0,
+      "step": 3980
+    },
+    {
+      "epoch": 9.591430295715147,
+      "eval_entropy": 0.22938149347064202,
+      "eval_loss": 1.0711463689804077,
+      "eval_mean_token_accuracy": 0.8282721410976367,
+      "eval_num_tokens": 9219278.0,
+      "eval_runtime": 88.4978,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 3980
+    },
+    {
+      "entropy": 0.09432941288687288,
+      "epoch": 9.63971031985516,
+      "grad_norm": 0.30733999609947205,
+      "learning_rate": 1.515470378197889e-06,
+      "loss": 0.0672213613986969,
+      "mean_token_accuracy": 0.9728612303733826,
+      "num_tokens": 9262346.0,
+      "step": 4000
+    },
+    {
+      "epoch": 9.63971031985516,
+      "eval_entropy": 0.22936245958121976,
+      "eval_loss": 1.071343183517456,
+      "eval_mean_token_accuracy": 0.8282827025049189,
+      "eval_num_tokens": 9262346.0,
+      "eval_runtime": 88.4428,
+      "eval_samples_per_second": 16.056,
+      "eval_steps_per_second": 2.013,
+      "step": 4000
+    },
+    {
+      "entropy": 0.08287148931995034,
+      "epoch": 9.687990343995171,
+      "grad_norm": 0.1976361870765686,
+      "learning_rate": 1.1409867309726256e-06,
+      "loss": 0.06318551301956177,
+      "mean_token_accuracy": 0.9770804084837437,
+      "num_tokens": 9309622.0,
+      "step": 4020
+    },
+    {
+      "epoch": 9.687990343995171,
+      "eval_entropy": 0.22929051920269314,
+      "eval_loss": 1.0711324214935303,
+      "eval_mean_token_accuracy": 0.8281597297513084,
+      "eval_num_tokens": 9309622.0,
+      "eval_runtime": 88.5243,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 4020
+    },
+    {
+      "entropy": 0.0907150611979887,
+      "epoch": 9.736270368135184,
+      "grad_norm": 0.16192017495632172,
+      "learning_rate": 8.194225455723828e-07,
+      "loss": 0.06589602828025817,
+      "mean_token_accuracy": 0.9738550461828709,
+      "num_tokens": 9354549.0,
+      "step": 4040
+    },
+    {
+      "epoch": 9.736270368135184,
+      "eval_entropy": 0.22913936391640244,
+      "eval_loss": 1.0715173482894897,
+      "eval_mean_token_accuracy": 0.8281301544623428,
+      "eval_num_tokens": 9354549.0,
+      "eval_runtime": 88.875,
+      "eval_samples_per_second": 15.978,
+      "eval_steps_per_second": 2.003,
+      "step": 4040
+    },
+    {
+      "entropy": 0.09003224167972804,
+      "epoch": 9.784550392275197,
+      "grad_norm": 0.2092711478471756,
+      "learning_rate": 5.508688208578523e-07,
+      "loss": 0.06655729413032532,
+      "mean_token_accuracy": 0.9750426560640335,
+      "num_tokens": 9399394.0,
+      "step": 4060
+    },
+    {
+      "epoch": 9.784550392275197,
+      "eval_entropy": 0.2291239777140403,
+      "eval_loss": 1.0715563297271729,
+      "eval_mean_token_accuracy": 0.8281808595978812,
+      "eval_num_tokens": 9399394.0,
+      "eval_runtime": 88.4328,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 4060
+    },
+    {
+      "entropy": 0.08719871481880545,
+      "epoch": 9.832830416415208,
+      "grad_norm": 0.17966455221176147,
+      "learning_rate": 3.3540155435645076e-07,
+      "loss": 0.05917760729789734,
+      "mean_token_accuracy": 0.9752349570393563,
+      "num_tokens": 9449483.0,
+      "step": 4080
+    },
+    {
+      "epoch": 9.832830416415208,
+      "eval_entropy": 0.2291418021816886,
+      "eval_loss": 1.0714553594589233,
+      "eval_mean_token_accuracy": 0.8281870403986299,
+      "eval_num_tokens": 9449483.0,
+      "eval_runtime": 88.2612,
+      "eval_samples_per_second": 16.089,
+      "eval_steps_per_second": 2.017,
+      "step": 4080
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.996149940607304e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4100/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4100/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4100/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4100/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4100/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4100/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4100/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4100/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4100/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4100/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4100/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4100/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4100/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4100/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..d174b19caf00a3dc8026cf81a3cf34f38ee46ad5
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4100/trainer_state.json
@@ -0,0 +1,4339 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 9.88111044055522,
+  "eval_steps": 20,
+  "global_step": 4100,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    },
+    {
+      "entropy": 0.1326121488586068,
+      "epoch": 7.181050090525045,
+      "grad_norm": 0.4843117296695709,
+      "learning_rate": 8.411870386507724e-05,
+      "loss": 0.09537842869758606,
+      "mean_token_accuracy": 0.9645151920616627,
+      "num_tokens": 6898444.0,
+      "step": 2980
+    },
+    {
+      "epoch": 7.181050090525045,
+      "eval_entropy": 0.27999477783280813,
+      "eval_loss": 0.9031037092208862,
+      "eval_mean_token_accuracy": 0.8261882204018282,
+      "eval_num_tokens": 6898444.0,
+      "eval_runtime": 88.5602,
+      "eval_samples_per_second": 16.034,
+      "eval_steps_per_second": 2.01,
+      "step": 2980
+    },
+    {
+      "entropy": 0.1156751093454659,
+      "epoch": 7.229330114665057,
+      "grad_norm": 0.5297922492027283,
+      "learning_rate": 8.149629351430326e-05,
+      "loss": 0.09203402996063233,
+      "mean_token_accuracy": 0.9689733698964119,
+      "num_tokens": 6947106.0,
+      "step": 3000
+    },
+    {
+      "epoch": 7.229330114665057,
+      "eval_entropy": 0.2740225845508361,
+      "eval_loss": 0.8876440525054932,
+      "eval_mean_token_accuracy": 0.8265116857678703,
+      "eval_num_tokens": 6947106.0,
+      "eval_runtime": 87.7971,
+      "eval_samples_per_second": 16.174,
+      "eval_steps_per_second": 2.027,
+      "step": 3000
+    },
+    {
+      "entropy": 0.12116114404052496,
+      "epoch": 7.27761013880507,
+      "grad_norm": 0.5467398166656494,
+      "learning_rate": 7.890406302404618e-05,
+      "loss": 0.0933056652545929,
+      "mean_token_accuracy": 0.9676455594599247,
+      "num_tokens": 6992784.0,
+      "step": 3020
+    },
+    {
+      "epoch": 7.27761013880507,
+      "eval_entropy": 0.27238342793804876,
+      "eval_loss": 0.9166117310523987,
+      "eval_mean_token_accuracy": 0.825499015242866,
+      "eval_num_tokens": 6992784.0,
+      "eval_runtime": 88.3276,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 3020
+    },
+    {
+      "entropy": 0.11715119928121567,
+      "epoch": 7.325890162945082,
+      "grad_norm": 0.4976702332496643,
+      "learning_rate": 7.634274596487176e-05,
+      "loss": 0.09449659585952759,
+      "mean_token_accuracy": 0.966506478190422,
+      "num_tokens": 7040482.0,
+      "step": 3040
+    },
+    {
+      "epoch": 7.325890162945082,
+      "eval_entropy": 0.2788975556077582,
+      "eval_loss": 0.8833065629005432,
+      "eval_mean_token_accuracy": 0.8269920596915684,
+      "eval_num_tokens": 7040482.0,
+      "eval_runtime": 88.9016,
+      "eval_samples_per_second": 15.973,
+      "eval_steps_per_second": 2.002,
+      "step": 3040
+    },
+    {
+      "entropy": 0.12565911412239075,
+      "epoch": 7.3741701870850935,
+      "grad_norm": 0.5728679895401001,
+      "learning_rate": 7.381306715921025e-05,
+      "loss": 0.09626244902610778,
+      "mean_token_accuracy": 0.9661444149911403,
+      "num_tokens": 7086261.0,
+      "step": 3060
+    },
+    {
+      "epoch": 7.3741701870850935,
+      "eval_entropy": 0.2762492011939542,
+      "eval_loss": 0.8961160778999329,
+      "eval_mean_token_accuracy": 0.826369910092836,
+      "eval_num_tokens": 7086261.0,
+      "eval_runtime": 88.5204,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 3060
+    },
+    {
+      "entropy": 0.12258567921817302,
+      "epoch": 7.422450211225105,
+      "grad_norm": 0.468985915184021,
+      "learning_rate": 7.131574247624031e-05,
+      "loss": 0.09120266437530518,
+      "mean_token_accuracy": 0.9674938447773457,
+      "num_tokens": 7133247.0,
+      "step": 3080
+    },
+    {
+      "epoch": 7.422450211225105,
+      "eval_entropy": 0.2688512086366,
+      "eval_loss": 0.9048654437065125,
+      "eval_mean_token_accuracy": 0.8265582302982888,
+      "eval_num_tokens": 7133247.0,
+      "eval_runtime": 88.164,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 3080
+    },
+    {
+      "entropy": 0.12070536343380808,
+      "epoch": 7.470730235365117,
+      "grad_norm": 0.5310841798782349,
+      "learning_rate": 6.885147862930688e-05,
+      "loss": 0.09386247992515565,
+      "mean_token_accuracy": 0.96664222702384,
+      "num_tokens": 7177810.0,
+      "step": 3100
+    },
+    {
+      "epoch": 7.470730235365117,
+      "eval_entropy": 0.26289632874593305,
+      "eval_loss": 0.902837336063385,
+      "eval_mean_token_accuracy": 0.828052184554968,
+      "eval_num_tokens": 7177810.0,
+      "eval_runtime": 88.1201,
+      "eval_samples_per_second": 16.114,
+      "eval_steps_per_second": 2.02,
+      "step": 3100
+    },
+    {
+      "entropy": 0.11322665181942285,
+      "epoch": 7.51901025950513,
+      "grad_norm": 0.6846067905426025,
+      "learning_rate": 6.642097297592908e-05,
+      "loss": 0.08772538304328918,
+      "mean_token_accuracy": 0.9691362135112286,
+      "num_tokens": 7225302.0,
+      "step": 3120
+    },
+    {
+      "epoch": 7.51901025950513,
+      "eval_entropy": 0.2670605613776807,
+      "eval_loss": 0.9130566120147705,
+      "eval_mean_token_accuracy": 0.826366903741708,
+      "eval_num_tokens": 7225302.0,
+      "eval_runtime": 88.6035,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3120
+    },
+    {
+      "entropy": 0.10985926580615342,
+      "epoch": 7.567290283645142,
+      "grad_norm": 0.4799748361110687,
+      "learning_rate": 6.402491332045683e-05,
+      "loss": 0.08672296404838561,
+      "mean_token_accuracy": 0.9698824048042297,
+      "num_tokens": 7274881.0,
+      "step": 3140
+    },
+    {
+      "epoch": 7.567290283645142,
+      "eval_entropy": 0.26657010866015146,
+      "eval_loss": 0.9056380987167358,
+      "eval_mean_token_accuracy": 0.8272326461384806,
+      "eval_num_tokens": 7274881.0,
+      "eval_runtime": 88.2967,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3140
+    },
+    {
+      "entropy": 0.11890473761595785,
+      "epoch": 7.615570307785154,
+      "grad_norm": 0.4081912934780121,
+      "learning_rate": 6.166397771942987e-05,
+      "loss": 0.09396827220916748,
+      "mean_token_accuracy": 0.967512384802103,
+      "num_tokens": 7321163.0,
+      "step": 3160
+    },
+    {
+      "epoch": 7.615570307785154,
+      "eval_entropy": 0.26364776575833226,
+      "eval_loss": 0.9204733967781067,
+      "eval_mean_token_accuracy": 0.827419729714983,
+      "eval_num_tokens": 7321163.0,
+      "eval_runtime": 88.3441,
+      "eval_samples_per_second": 16.074,
+      "eval_steps_per_second": 2.015,
+      "step": 3160
+    },
+    {
+      "entropy": 0.12209134344011545,
+      "epoch": 7.663850331925166,
+      "grad_norm": 0.5159774422645569,
+      "learning_rate": 5.93388342896962e-05,
+      "loss": 0.09292978644371033,
+      "mean_token_accuracy": 0.9658421002328396,
+      "num_tokens": 7366217.0,
+      "step": 3180
+    },
+    {
+      "epoch": 7.663850331925166,
+      "eval_entropy": 0.2714110068725736,
+      "eval_loss": 0.9143092036247253,
+      "eval_mean_token_accuracy": 0.8262251099843657,
+      "eval_num_tokens": 7366217.0,
+      "eval_runtime": 87.9635,
+      "eval_samples_per_second": 16.143,
+      "eval_steps_per_second": 2.024,
+      "step": 3180
+    },
+    {
+      "entropy": 0.11955609126016498,
+      "epoch": 7.712130356065178,
+      "grad_norm": 0.7005976438522339,
+      "learning_rate": 5.705014101934252e-05,
+      "loss": 0.09319776892662049,
+      "mean_token_accuracy": 0.9674513377249241,
+      "num_tokens": 7410887.0,
+      "step": 3200
+    },
+    {
+      "epoch": 7.712130356065178,
+      "eval_entropy": 0.2626291513442993,
+      "eval_loss": 0.91884446144104,
+      "eval_mean_token_accuracy": 0.827563955207889,
+      "eval_num_tokens": 7410887.0,
+      "eval_runtime": 87.8849,
+      "eval_samples_per_second": 16.157,
+      "eval_steps_per_second": 2.025,
+      "step": 3200
+    },
+    {
+      "entropy": 0.11767532899975777,
+      "epoch": 7.76041038020519,
+      "grad_norm": 0.4675362706184387,
+      "learning_rate": 5.479854558149187e-05,
+      "loss": 0.09038087725639343,
+      "mean_token_accuracy": 0.9669680051505566,
+      "num_tokens": 7457578.0,
+      "step": 3220
+    },
+    {
+      "epoch": 7.76041038020519,
+      "eval_entropy": 0.26581037203582486,
+      "eval_loss": 0.9146883487701416,
+      "eval_mean_token_accuracy": 0.8270345812433222,
+      "eval_num_tokens": 7457578.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 3220
+    },
+    {
+      "entropy": 0.11383061073720455,
+      "epoch": 7.808690404345202,
+      "grad_norm": 0.541247546672821,
+      "learning_rate": 5.258468515101918e-05,
+      "loss": 0.08942405581474304,
+      "mean_token_accuracy": 0.968398705124855,
+      "num_tokens": 7505761.0,
+      "step": 3240
+    },
+    {
+      "epoch": 7.808690404345202,
+      "eval_entropy": 0.26687998960861997,
+      "eval_loss": 0.9037085175514221,
+      "eval_mean_token_accuracy": 0.8274977863504646,
+      "eval_num_tokens": 7505761.0,
+      "eval_runtime": 88.5163,
+      "eval_samples_per_second": 16.042,
+      "eval_steps_per_second": 2.011,
+      "step": 3240
+    },
+    {
+      "entropy": 0.12004058873280883,
+      "epoch": 7.856970428485214,
+      "grad_norm": 0.5289552211761475,
+      "learning_rate": 5.040918622423899e-05,
+      "loss": 0.09064805507659912,
+      "mean_token_accuracy": 0.9673330299556255,
+      "num_tokens": 7552557.0,
+      "step": 3260
+    },
+    {
+      "epoch": 7.856970428485214,
+      "eval_entropy": 0.26288379251622085,
+      "eval_loss": 0.9146178364753723,
+      "eval_mean_token_accuracy": 0.8276662448149049,
+      "eval_num_tokens": 7552557.0,
+      "eval_runtime": 88.0723,
+      "eval_samples_per_second": 16.123,
+      "eval_steps_per_second": 2.021,
+      "step": 3260
+    },
+    {
+      "entropy": 0.1180689456872642,
+      "epoch": 7.905250452625227,
+      "grad_norm": 0.4754861891269684,
+      "learning_rate": 4.827266444161377e-05,
+      "loss": 0.09402655363082886,
+      "mean_token_accuracy": 0.9663361787796021,
+      "num_tokens": 7596998.0,
+      "step": 3280
+    },
+    {
+      "epoch": 7.905250452625227,
+      "eval_entropy": 0.261690996922134,
+      "eval_loss": 0.9120186567306519,
+      "eval_mean_token_accuracy": 0.8278995015647974,
+      "eval_num_tokens": 7596998.0,
+      "eval_runtime": 88.6573,
+      "eval_samples_per_second": 16.017,
+      "eval_steps_per_second": 2.008,
+      "step": 3280
+    },
+    {
+      "entropy": 0.1162164019420743,
+      "epoch": 7.953530476765239,
+      "grad_norm": 0.6203398108482361,
+      "learning_rate": 4.61757244135355e-05,
+      "loss": 0.08832768201828003,
+      "mean_token_accuracy": 0.9694220684468746,
+      "num_tokens": 7643288.0,
+      "step": 3300
+    },
+    {
+      "epoch": 7.953530476765239,
+      "eval_entropy": 0.26158678129817664,
+      "eval_loss": 0.9301361441612244,
+      "eval_mean_token_accuracy": 0.8269603885291668,
+      "eval_num_tokens": 7643288.0,
+      "eval_runtime": 88.7064,
+      "eval_samples_per_second": 16.008,
+      "eval_steps_per_second": 2.007,
+      "step": 3300
+    },
+    {
+      "entropy": 0.10877202204489089,
+      "epoch": 8.0,
+      "grad_norm": 3.058673143386841,
+      "learning_rate": 4.411895954922763e-05,
+      "loss": 0.09159615635871887,
+      "mean_token_accuracy": 0.9686878834451947,
+      "num_tokens": 7688608.0,
+      "step": 3320
+    },
+    {
+      "epoch": 8.0,
+      "eval_entropy": 0.2590438057197614,
+      "eval_loss": 0.9262065887451172,
+      "eval_mean_token_accuracy": 0.8277404546067956,
+      "eval_num_tokens": 7688608.0,
+      "eval_runtime": 87.7557,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 3320
+    },
+    {
+      "entropy": 0.09969092034734786,
+      "epoch": 8.048280024140013,
+      "grad_norm": 0.2561911642551422,
+      "learning_rate": 4.2102951888817644e-05,
+      "loss": 0.07228735685348511,
+      "mean_token_accuracy": 0.9743556551635265,
+      "num_tokens": 7732856.0,
+      "step": 3340
+    },
+    {
+      "epoch": 8.048280024140013,
+      "eval_entropy": 0.2462534454933713,
+      "eval_loss": 0.9755983352661133,
+      "eval_mean_token_accuracy": 0.827510209565752,
+      "eval_num_tokens": 7732856.0,
+      "eval_runtime": 88.0297,
+      "eval_samples_per_second": 16.131,
+      "eval_steps_per_second": 2.022,
+      "step": 3340
+    },
+    {
+      "entropy": 0.09477624730207027,
+      "epoch": 8.096560048280024,
+      "grad_norm": 0.3670356571674347,
+      "learning_rate": 4.012827193862633e-05,
+      "loss": 0.0686036467552185,
+      "mean_token_accuracy": 0.9743670664727688,
+      "num_tokens": 7778585.0,
+      "step": 3360
+    },
+    {
+      "epoch": 8.096560048280024,
+      "eval_entropy": 0.2440882951356052,
+      "eval_loss": 1.0118423700332642,
+      "eval_mean_token_accuracy": 0.8271241070849172,
+      "eval_num_tokens": 7778585.0,
+      "eval_runtime": 88.1364,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 3360
+    },
+    {
+      "entropy": 0.09417980071157217,
+      "epoch": 8.144840072420036,
+      "grad_norm": 0.5785111784934998,
+      "learning_rate": 3.819547850972133e-05,
+      "loss": 0.07100698947906495,
+      "mean_token_accuracy": 0.9737804099917412,
+      "num_tokens": 7824287.0,
+      "step": 3380
+    },
+    {
+      "epoch": 8.144840072420036,
+      "eval_entropy": 0.24603172860453637,
+      "eval_loss": 1.0021991729736328,
+      "eval_mean_token_accuracy": 0.827162122793412,
+      "eval_num_tokens": 7824287.0,
+      "eval_runtime": 88.4534,
+      "eval_samples_per_second": 16.054,
+      "eval_steps_per_second": 2.012,
+      "step": 3380
+    },
+    {
+      "entropy": 0.087222247896716,
+      "epoch": 8.193120096560047,
+      "grad_norm": 0.44848504662513733,
+      "learning_rate": 3.6305118559780393e-05,
+      "loss": 0.06325234174728393,
+      "mean_token_accuracy": 0.9769335232675076,
+      "num_tokens": 7874784.0,
+      "step": 3400
+    },
+    {
+      "epoch": 8.193120096560047,
+      "eval_entropy": 0.24544863905129807,
+      "eval_loss": 1.005008339881897,
+      "eval_mean_token_accuracy": 0.8273035240976998,
+      "eval_num_tokens": 7874784.0,
+      "eval_runtime": 88.6737,
+      "eval_samples_per_second": 16.014,
+      "eval_steps_per_second": 2.007,
+      "step": 3400
+    },
+    {
+      "entropy": 0.09037275062873959,
+      "epoch": 8.24140012070006,
+      "grad_norm": 0.23529155552387238,
+      "learning_rate": 3.4457727038308435e-05,
+      "loss": 0.06589569449424744,
+      "mean_token_accuracy": 0.9751470290124417,
+      "num_tokens": 7925240.0,
+      "step": 3420
+    },
+    {
+      "epoch": 8.24140012070006,
+      "eval_entropy": 0.24203423494368456,
+      "eval_loss": 1.0052824020385742,
+      "eval_mean_token_accuracy": 0.8275575503874361,
+      "eval_num_tokens": 7925240.0,
+      "eval_runtime": 88.3486,
+      "eval_samples_per_second": 16.073,
+      "eval_steps_per_second": 2.015,
+      "step": 3420
+    },
+    {
+      "entropy": 0.10373208015225828,
+      "epoch": 8.289680144840073,
+      "grad_norm": 0.24796408414840698,
+      "learning_rate": 3.2653826735253594e-05,
+      "loss": 0.07714771628379821,
+      "mean_token_accuracy": 0.9722891598939896,
+      "num_tokens": 7966685.0,
+      "step": 3440
+    },
+    {
+      "epoch": 8.289680144840073,
+      "eval_entropy": 0.23461157050025597,
+      "eval_loss": 1.012108325958252,
+      "eval_mean_token_accuracy": 0.82825900564033,
+      "eval_num_tokens": 7966685.0,
+      "eval_runtime": 88.1018,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 3440
+    },
+    {
+      "entropy": 0.08967954274266958,
+      "epoch": 8.337960168980084,
+      "grad_norm": 0.5314674973487854,
+      "learning_rate": 3.08939281330635e-05,
+      "loss": 0.06746266484260559,
+      "mean_token_accuracy": 0.9753133699297905,
+      "num_tokens": 8013656.0,
+      "step": 3460
+    },
+    {
+      "epoch": 8.337960168980084,
+      "eval_entropy": 0.23781998589467468,
+      "eval_loss": 1.0119913816452026,
+      "eval_mean_token_accuracy": 0.8273287157664139,
+      "eval_num_tokens": 8013656.0,
+      "eval_runtime": 88.6038,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3460
+    },
+    {
+      "entropy": 0.09280927330255509,
+      "epoch": 8.386240193120097,
+      "grad_norm": 0.30658936500549316,
+      "learning_rate": 2.9178529262224946e-05,
+      "loss": 0.06936421990394592,
+      "mean_token_accuracy": 0.9742105580866337,
+      "num_tokens": 8061108.0,
+      "step": 3480
+    },
+    {
+      "epoch": 8.386240193120097,
+      "eval_entropy": 0.23672426734747512,
+      "eval_loss": 1.0162041187286377,
+      "eval_mean_token_accuracy": 0.8278394281194451,
+      "eval_num_tokens": 8061108.0,
+      "eval_runtime": 88.6015,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 3480
+    },
+    {
+      "entropy": 0.09195122267119586,
+      "epoch": 8.43452021726011,
+      "grad_norm": 0.24512454867362976,
+      "learning_rate": 2.7508115560327256e-05,
+      "loss": 0.06996064782142639,
+      "mean_token_accuracy": 0.974104854464531,
+      "num_tokens": 8106924.0,
+      "step": 3500
+    },
+    {
+      "epoch": 8.43452021726011,
+      "eval_entropy": 0.23758245828781235,
+      "eval_loss": 1.0157747268676758,
+      "eval_mean_token_accuracy": 0.8278501579600773,
+      "eval_num_tokens": 8106924.0,
+      "eval_runtime": 88.5678,
+      "eval_samples_per_second": 16.033,
+      "eval_steps_per_second": 2.01,
+      "step": 3500
+    },
+    {
+      "entropy": 0.09518477590754629,
+      "epoch": 8.48280024140012,
+      "grad_norm": 0.2373228371143341,
+      "learning_rate": 2.5883159734688868e-05,
+      "loss": 0.06832797527313232,
+      "mean_token_accuracy": 0.9743806682527065,
+      "num_tokens": 8153070.0,
+      "step": 3520
+    },
+    {
+      "epoch": 8.48280024140012,
+      "eval_entropy": 0.2379308074210467,
+      "eval_loss": 1.0180530548095703,
+      "eval_mean_token_accuracy": 0.8282477052024241,
+      "eval_num_tokens": 8153070.0,
+      "eval_runtime": 88.4651,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 3520
+    },
+    {
+      "entropy": 0.0906269280705601,
+      "epoch": 8.531080265540133,
+      "grad_norm": 0.37602657079696655,
+      "learning_rate": 2.4304121628587173e-05,
+      "loss": 0.06921588778495788,
+      "mean_token_accuracy": 0.9745571479201317,
+      "num_tokens": 8200066.0,
+      "step": 3540
+    },
+    {
+      "epoch": 8.531080265540133,
+      "eval_entropy": 0.2357490769263064,
+      "eval_loss": 1.017139196395874,
+      "eval_mean_token_accuracy": 0.828713178299786,
+      "eval_num_tokens": 8200066.0,
+      "eval_runtime": 88.6424,
+      "eval_samples_per_second": 16.019,
+      "eval_steps_per_second": 2.008,
+      "step": 3540
+    },
+    {
+      "entropy": 0.09273334196768701,
+      "epoch": 8.579360289680144,
+      "grad_norm": 0.2978838086128235,
+      "learning_rate": 2.277144809112794e-05,
+      "loss": 0.06970887780189514,
+      "mean_token_accuracy": 0.9751374177634716,
+      "num_tokens": 8246157.0,
+      "step": 3560
+    },
+    {
+      "epoch": 8.579360289680144,
+      "eval_entropy": 0.23590933138065123,
+      "eval_loss": 1.019231915473938,
+      "eval_mean_token_accuracy": 0.8279163412163767,
+      "eval_num_tokens": 8246157.0,
+      "eval_runtime": 88.2691,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 3560
+    },
+    {
+      "entropy": 0.09854201921261847,
+      "epoch": 8.627640313820157,
+      "grad_norm": 0.5381243228912354,
+      "learning_rate": 2.1285572850792413e-05,
+      "loss": 0.07123501300811767,
+      "mean_token_accuracy": 0.9723988175392151,
+      "num_tokens": 8291752.0,
+      "step": 3580
+    },
+    {
+      "epoch": 8.627640313820157,
+      "eval_entropy": 0.23592024949494372,
+      "eval_loss": 1.020250678062439,
+      "eval_mean_token_accuracy": 0.8282466226079491,
+      "eval_num_tokens": 8291752.0,
+      "eval_runtime": 87.8939,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 3580
+    },
+    {
+      "entropy": 0.09092000788077712,
+      "epoch": 8.67592033796017,
+      "grad_norm": 0.29089561104774475,
+      "learning_rate": 1.9846916392697493e-05,
+      "loss": 0.06982720494270325,
+      "mean_token_accuracy": 0.9749995484948158,
+      "num_tokens": 8337651.0,
+      "step": 3600
+    },
+    {
+      "epoch": 8.67592033796017,
+      "eval_entropy": 0.23658913382318583,
+      "eval_loss": 1.019862413406372,
+      "eval_mean_token_accuracy": 0.8280127842104836,
+      "eval_num_tokens": 8337651.0,
+      "eval_runtime": 88.2855,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 3600
+    },
+    {
+      "entropy": 0.08708424470387399,
+      "epoch": 8.72420036210018,
+      "grad_norm": 0.29904839396476746,
+      "learning_rate": 1.8455885839603106e-05,
+      "loss": 0.06801370978355407,
+      "mean_token_accuracy": 0.9755801670253277,
+      "num_tokens": 8385330.0,
+      "step": 3620
+    },
+    {
+      "epoch": 8.72420036210018,
+      "eval_entropy": 0.23627035771862845,
+      "eval_loss": 1.0204569101333618,
+      "eval_mean_token_accuracy": 0.8279967894045155,
+      "eval_num_tokens": 8385330.0,
+      "eval_runtime": 88.1753,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 3620
+    },
+    {
+      "entropy": 0.0916255566291511,
+      "epoch": 8.772480386240193,
+      "grad_norm": 0.20571033656597137,
+      "learning_rate": 1.7112874836701502e-05,
+      "loss": 0.0692249894142151,
+      "mean_token_accuracy": 0.9745148062705994,
+      "num_tokens": 8432134.0,
+      "step": 3640
+    },
+    {
+      "epoch": 8.772480386240193,
+      "eval_entropy": 0.2340821772813797,
+      "eval_loss": 1.0212821960449219,
+      "eval_mean_token_accuracy": 0.828373622358515,
+      "eval_num_tokens": 8432134.0,
+      "eval_runtime": 87.2518,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 3640
+    },
+    {
+      "entropy": 0.0926868049427867,
+      "epoch": 8.820760410380204,
+      "grad_norm": 0.2618265450000763,
+      "learning_rate": 1.5818263440220203e-05,
+      "loss": 0.07338647246360779,
+      "mean_token_accuracy": 0.9724708944559097,
+      "num_tokens": 8476758.0,
+      "step": 3660
+    },
+    {
+      "epoch": 8.820760410380204,
+      "eval_entropy": 0.23637074575330433,
+      "eval_loss": 1.0186725854873657,
+      "eval_mean_token_accuracy": 0.8277707605549459,
+      "eval_num_tokens": 8476758.0,
+      "eval_runtime": 88.4695,
+      "eval_samples_per_second": 16.051,
+      "eval_steps_per_second": 2.012,
+      "step": 3660
+    },
+    {
+      "entropy": 0.09175782897509635,
+      "epoch": 8.869040434520217,
+      "grad_norm": 0.2889852821826935,
+      "learning_rate": 1.4572418009870604e-05,
+      "loss": 0.07066041231155396,
+      "mean_token_accuracy": 0.9725673109292984,
+      "num_tokens": 8523294.0,
+      "step": 3680
+    },
+    {
+      "epoch": 8.869040434520217,
+      "eval_entropy": 0.23945164404223473,
+      "eval_loss": 1.0145231485366821,
+      "eval_mean_token_accuracy": 0.8281202523895864,
+      "eval_num_tokens": 8523294.0,
+      "eval_runtime": 88.368,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 3680
+    },
+    {
+      "entropy": 0.09265611860901117,
+      "epoch": 8.91732045866023,
+      "grad_norm": 0.41092950105667114,
+      "learning_rate": 1.3375691105172625e-05,
+      "loss": 0.06975027322769164,
+      "mean_token_accuracy": 0.974636297672987,
+      "num_tokens": 8570358.0,
+      "step": 3700
+    },
+    {
+      "epoch": 8.91732045866023,
+      "eval_entropy": 0.239588919901446,
+      "eval_loss": 1.014059066772461,
+      "eval_mean_token_accuracy": 0.828028041660116,
+      "eval_num_tokens": 8570358.0,
+      "eval_runtime": 88.1039,
+      "eval_samples_per_second": 16.117,
+      "eval_steps_per_second": 2.02,
+      "step": 3700
+    },
+    {
+      "entropy": 0.0896465003490448,
+      "epoch": 8.965600482800241,
+      "grad_norm": 0.24000254273414612,
+      "learning_rate": 1.2228421385684305e-05,
+      "loss": 0.06743953824043274,
+      "mean_token_accuracy": 0.974006575345993,
+      "num_tokens": 8618246.0,
+      "step": 3720
+    },
+    {
+      "epoch": 8.965600482800241,
+      "eval_entropy": 0.2390461369846644,
+      "eval_loss": 1.011997938156128,
+      "eval_mean_token_accuracy": 0.8284533211354459,
+      "eval_num_tokens": 8618246.0,
+      "eval_runtime": 88.6202,
+      "eval_samples_per_second": 16.023,
+      "eval_steps_per_second": 2.009,
+      "step": 3720
+    },
+    {
+      "entropy": 0.09575109251520851,
+      "epoch": 9.012070006035003,
+      "grad_norm": 0.18088576197624207,
+      "learning_rate": 1.1130933515165429e-05,
+      "loss": 0.07295305132865906,
+      "mean_token_accuracy": 0.9739722857227573,
+      "num_tokens": 8661177.0,
+      "step": 3740
+    },
+    {
+      "epoch": 9.012070006035003,
+      "eval_entropy": 0.23728882387447892,
+      "eval_loss": 1.0134038925170898,
+      "eval_mean_token_accuracy": 0.8286508951294288,
+      "eval_num_tokens": 8661177.0,
+      "eval_runtime": 88.6434,
+      "eval_samples_per_second": 16.019,
+      "eval_steps_per_second": 2.008,
+      "step": 3740
+    },
+    {
+      "entropy": 0.08883201805874705,
+      "epoch": 9.060350030175016,
+      "grad_norm": 0.23120898008346558,
+      "learning_rate": 1.008353806970132e-05,
+      "loss": 0.06326142549514771,
+      "mean_token_accuracy": 0.9780020996928215,
+      "num_tokens": 8707602.0,
+      "step": 3760
+    },
+    {
+      "epoch": 9.060350030175016,
+      "eval_entropy": 0.23416905331142832,
+      "eval_loss": 1.0284100770950317,
+      "eval_mean_token_accuracy": 0.8284083213029283,
+      "eval_num_tokens": 8707602.0,
+      "eval_runtime": 88.2933,
+      "eval_samples_per_second": 16.083,
+      "eval_steps_per_second": 2.016,
+      "step": 3760
+    },
+    {
+      "entropy": 0.08861468276008963,
+      "epoch": 9.108630054315027,
+      "grad_norm": 0.17277656495571136,
+      "learning_rate": 9.086531449813489e-06,
+      "loss": 0.06727416515350342,
+      "mean_token_accuracy": 0.9760390736162663,
+      "num_tokens": 8751287.0,
+      "step": 3780
+    },
+    {
+      "epoch": 9.108630054315027,
+      "eval_entropy": 0.23370317386442355,
+      "eval_loss": 1.0381091833114624,
+      "eval_mean_token_accuracy": 0.8282668178670862,
+      "eval_num_tokens": 8751287.0,
+      "eval_runtime": 88.5738,
+      "eval_samples_per_second": 16.032,
+      "eval_steps_per_second": 2.01,
+      "step": 3780
+    },
+    {
+      "entropy": 0.09032954541034996,
+      "epoch": 9.15691007845504,
+      "grad_norm": 0.19481085240840912,
+      "learning_rate": 8.140195796581981e-06,
+      "loss": 0.06117786169052124,
+      "mean_token_accuracy": 0.9764454193413258,
+      "num_tokens": 8799982.0,
+      "step": 3800
+    },
+    {
+      "epoch": 9.15691007845504,
+      "eval_entropy": 0.23215936609868254,
+      "eval_loss": 1.0467767715454102,
+      "eval_mean_token_accuracy": 0.828405868806196,
+      "eval_num_tokens": 8799982.0,
+      "eval_runtime": 87.9551,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 3800
+    },
+    {
+      "entropy": 0.08824449330568314,
+      "epoch": 9.20519010259505,
+      "grad_norm": 0.20382793247699738,
+      "learning_rate": 7.2447989118025455e-06,
+      "loss": 0.06422553658485412,
+      "mean_token_accuracy": 0.9763121999800205,
+      "num_tokens": 8845536.0,
+      "step": 3820
+    },
+    {
+      "epoch": 9.20519010259505,
+      "eval_entropy": 0.23169661472352704,
+      "eval_loss": 1.0529043674468994,
+      "eval_mean_token_accuracy": 0.8282025610463003,
+      "eval_num_tokens": 8845536.0,
+      "eval_runtime": 88.5548,
+      "eval_samples_per_second": 16.035,
+      "eval_steps_per_second": 2.01,
+      "step": 3820
+    },
+    {
+      "entropy": 0.07768499720841646,
+      "epoch": 9.253470126735063,
+      "grad_norm": 0.11727919429540634,
+      "learning_rate": 6.400594182202084e-06,
+      "loss": 0.057279336452484134,
+      "mean_token_accuracy": 0.9795166745781898,
+      "num_tokens": 8896053.0,
+      "step": 3840
+    },
+    {
+      "epoch": 9.253470126735063,
+      "eval_entropy": 0.2301936373114586,
+      "eval_loss": 1.0576049089431763,
+      "eval_mean_token_accuracy": 0.8283417489421502,
+      "eval_num_tokens": 8896053.0,
+      "eval_runtime": 88.7854,
+      "eval_samples_per_second": 15.994,
+      "eval_steps_per_second": 2.005,
+      "step": 3840
+    },
+    {
+      "entropy": 0.09015200780704617,
+      "epoch": 9.301750150875076,
+      "grad_norm": 0.2089058756828308,
+      "learning_rate": 5.6078205077330705e-06,
+      "loss": 0.06626194715499878,
+      "mean_token_accuracy": 0.9747506938874722,
+      "num_tokens": 8940591.0,
+      "step": 3860
+    },
+    {
+      "epoch": 9.301750150875076,
+      "eval_entropy": 0.23024900676159377,
+      "eval_loss": 1.0599786043167114,
+      "eval_mean_token_accuracy": 0.8283051915383071,
+      "eval_num_tokens": 8940591.0,
+      "eval_runtime": 88.2963,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3860
+    },
+    {
+      "entropy": 0.08484094133600592,
+      "epoch": 9.350030175015087,
+      "grad_norm": 0.21646778285503387,
+      "learning_rate": 4.866702233967785e-06,
+      "loss": 0.05998039245605469,
+      "mean_token_accuracy": 0.9771868668496608,
+      "num_tokens": 8989035.0,
+      "step": 3880
+    },
+    {
+      "epoch": 9.350030175015087,
+      "eval_entropy": 0.2300723305578982,
+      "eval_loss": 1.0633587837219238,
+      "eval_mean_token_accuracy": 0.828371500701047,
+      "eval_num_tokens": 8989035.0,
+      "eval_runtime": 88.0233,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 3880
+    },
+    {
+      "entropy": 0.08905019680969417,
+      "epoch": 9.3983101991551,
+      "grad_norm": 0.21842406690120697,
+      "learning_rate": 4.177449088611095e-06,
+      "loss": 0.06666865348815917,
+      "mean_token_accuracy": 0.9753445640206337,
+      "num_tokens": 9033383.0,
+      "step": 3900
+    },
+    {
+      "epoch": 9.3983101991551,
+      "eval_entropy": 0.2295504841911659,
+      "eval_loss": 1.0662461519241333,
+      "eval_mean_token_accuracy": 0.8284122297603093,
+      "eval_num_tokens": 9033383.0,
+      "eval_runtime": 88.5525,
+      "eval_samples_per_second": 16.036,
+      "eval_steps_per_second": 2.01,
+      "step": 3900
+    },
+    {
+      "entropy": 0.08582521872594953,
+      "epoch": 9.446590223295111,
+      "grad_norm": 0.2000240683555603,
+      "learning_rate": 3.540256122150088e-06,
+      "loss": 0.06327944993972778,
+      "mean_token_accuracy": 0.9765432856976985,
+      "num_tokens": 9079672.0,
+      "step": 3920
+    },
+    {
+      "epoch": 9.446590223295111,
+      "eval_entropy": 0.2297678625818049,
+      "eval_loss": 1.0676169395446777,
+      "eval_mean_token_accuracy": 0.8283703879693921,
+      "eval_num_tokens": 9079672.0,
+      "eval_runtime": 87.6897,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 3920
+    },
+    {
+      "entropy": 0.09261773955076932,
+      "epoch": 9.494870247435124,
+      "grad_norm": 0.19104281067848206,
+      "learning_rate": 2.955303652656991e-06,
+      "loss": 0.06111966967582703,
+      "mean_token_accuracy": 0.9738784030079841,
+      "num_tokens": 9127647.0,
+      "step": 3940
+    },
+    {
+      "epoch": 9.494870247435124,
+      "eval_entropy": 0.22986975881490815,
+      "eval_loss": 1.0687925815582275,
+      "eval_mean_token_accuracy": 0.8282996858103892,
+      "eval_num_tokens": 9127647.0,
+      "eval_runtime": 88.2224,
+      "eval_samples_per_second": 16.096,
+      "eval_steps_per_second": 2.018,
+      "step": 3940
+    },
+    {
+      "entropy": 0.08347254800610245,
+      "epoch": 9.543150271575136,
+      "grad_norm": 0.18671369552612305,
+      "learning_rate": 2.422757214761356e-06,
+      "loss": 0.06384263038635254,
+      "mean_token_accuracy": 0.9770128890872002,
+      "num_tokens": 9172587.0,
+      "step": 3960
+    },
+    {
+      "epoch": 9.543150271575136,
+      "eval_entropy": 0.22975517851248217,
+      "eval_loss": 1.0701013803482056,
+      "eval_mean_token_accuracy": 0.8282244285840666,
+      "eval_num_tokens": 9172587.0,
+      "eval_runtime": 88.5794,
+      "eval_samples_per_second": 16.031,
+      "eval_steps_per_second": 2.009,
+      "step": 3960
+    },
+    {
+      "entropy": 0.08136763009242713,
+      "epoch": 9.591430295715147,
+      "grad_norm": 0.24811454117298126,
+      "learning_rate": 1.942767512805673e-06,
+      "loss": 0.06044899821281433,
+      "mean_token_accuracy": 0.9776685133576393,
+      "num_tokens": 9219278.0,
+      "step": 3980
+    },
+    {
+      "epoch": 9.591430295715147,
+      "eval_entropy": 0.22938149347064202,
+      "eval_loss": 1.0711463689804077,
+      "eval_mean_token_accuracy": 0.8282721410976367,
+      "eval_num_tokens": 9219278.0,
+      "eval_runtime": 88.4978,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 3980
+    },
+    {
+      "entropy": 0.09432941288687288,
+      "epoch": 9.63971031985516,
+      "grad_norm": 0.30733999609947205,
+      "learning_rate": 1.515470378197889e-06,
+      "loss": 0.0672213613986969,
+      "mean_token_accuracy": 0.9728612303733826,
+      "num_tokens": 9262346.0,
+      "step": 4000
+    },
+    {
+      "epoch": 9.63971031985516,
+      "eval_entropy": 0.22936245958121976,
+      "eval_loss": 1.071343183517456,
+      "eval_mean_token_accuracy": 0.8282827025049189,
+      "eval_num_tokens": 9262346.0,
+      "eval_runtime": 88.4428,
+      "eval_samples_per_second": 16.056,
+      "eval_steps_per_second": 2.013,
+      "step": 4000
+    },
+    {
+      "entropy": 0.08287148931995034,
+      "epoch": 9.687990343995171,
+      "grad_norm": 0.1976361870765686,
+      "learning_rate": 1.1409867309726256e-06,
+      "loss": 0.06318551301956177,
+      "mean_token_accuracy": 0.9770804084837437,
+      "num_tokens": 9309622.0,
+      "step": 4020
+    },
+    {
+      "epoch": 9.687990343995171,
+      "eval_entropy": 0.22929051920269314,
+      "eval_loss": 1.0711324214935303,
+      "eval_mean_token_accuracy": 0.8281597297513084,
+      "eval_num_tokens": 9309622.0,
+      "eval_runtime": 88.5243,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 4020
+    },
+    {
+      "entropy": 0.0907150611979887,
+      "epoch": 9.736270368135184,
+      "grad_norm": 0.16192017495632172,
+      "learning_rate": 8.194225455723828e-07,
+      "loss": 0.06589602828025817,
+      "mean_token_accuracy": 0.9738550461828709,
+      "num_tokens": 9354549.0,
+      "step": 4040
+    },
+    {
+      "epoch": 9.736270368135184,
+      "eval_entropy": 0.22913936391640244,
+      "eval_loss": 1.0715173482894897,
+      "eval_mean_token_accuracy": 0.8281301544623428,
+      "eval_num_tokens": 9354549.0,
+      "eval_runtime": 88.875,
+      "eval_samples_per_second": 15.978,
+      "eval_steps_per_second": 2.003,
+      "step": 4040
+    },
+    {
+      "entropy": 0.09003224167972804,
+      "epoch": 9.784550392275197,
+      "grad_norm": 0.2092711478471756,
+      "learning_rate": 5.508688208578523e-07,
+      "loss": 0.06655729413032532,
+      "mean_token_accuracy": 0.9750426560640335,
+      "num_tokens": 9399394.0,
+      "step": 4060
+    },
+    {
+      "epoch": 9.784550392275197,
+      "eval_entropy": 0.2291239777140403,
+      "eval_loss": 1.0715563297271729,
+      "eval_mean_token_accuracy": 0.8281808595978812,
+      "eval_num_tokens": 9399394.0,
+      "eval_runtime": 88.4328,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 4060
+    },
+    {
+      "entropy": 0.08719871481880545,
+      "epoch": 9.832830416415208,
+      "grad_norm": 0.17966455221176147,
+      "learning_rate": 3.3540155435645076e-07,
+      "loss": 0.05917760729789734,
+      "mean_token_accuracy": 0.9752349570393563,
+      "num_tokens": 9449483.0,
+      "step": 4080
+    },
+    {
+      "epoch": 9.832830416415208,
+      "eval_entropy": 0.2291418021816886,
+      "eval_loss": 1.0714553594589233,
+      "eval_mean_token_accuracy": 0.8281870403986299,
+      "eval_num_tokens": 9449483.0,
+      "eval_runtime": 88.2612,
+      "eval_samples_per_second": 16.089,
+      "eval_steps_per_second": 2.017,
+      "step": 4080
+    },
+    {
+      "entropy": 0.07887464743107557,
+      "epoch": 9.88111044055522,
+      "grad_norm": 0.4397202134132385,
+      "learning_rate": 1.7308172075587924e-07,
+      "loss": 0.060457843542099,
+      "mean_token_accuracy": 0.9779011741280556,
+      "num_tokens": 9497731.0,
+      "step": 4100
+    },
+    {
+      "epoch": 9.88111044055522,
+      "eval_entropy": 0.2292038949855258,
+      "eval_loss": 1.0717124938964844,
+      "eval_mean_token_accuracy": 0.8282261353530241,
+      "eval_num_tokens": 9497731.0,
+      "eval_runtime": 88.5793,
+      "eval_samples_per_second": 16.031,
+      "eval_steps_per_second": 2.009,
+      "step": 4100
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 4.016731609606226e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4120/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4120/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4120/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4120/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4120/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4120/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4120/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4120/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4120/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4120/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4120/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4120/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4120/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4120/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..3a166eca37890dcd8d5c06784b45779bbdb0fb67
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4120/trainer_state.json
@@ -0,0 +1,4360 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 9.929390464695233,
+  "eval_steps": 20,
+  "global_step": 4120,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    },
+    {
+      "entropy": 0.1326121488586068,
+      "epoch": 7.181050090525045,
+      "grad_norm": 0.4843117296695709,
+      "learning_rate": 8.411870386507724e-05,
+      "loss": 0.09537842869758606,
+      "mean_token_accuracy": 0.9645151920616627,
+      "num_tokens": 6898444.0,
+      "step": 2980
+    },
+    {
+      "epoch": 7.181050090525045,
+      "eval_entropy": 0.27999477783280813,
+      "eval_loss": 0.9031037092208862,
+      "eval_mean_token_accuracy": 0.8261882204018282,
+      "eval_num_tokens": 6898444.0,
+      "eval_runtime": 88.5602,
+      "eval_samples_per_second": 16.034,
+      "eval_steps_per_second": 2.01,
+      "step": 2980
+    },
+    {
+      "entropy": 0.1156751093454659,
+      "epoch": 7.229330114665057,
+      "grad_norm": 0.5297922492027283,
+      "learning_rate": 8.149629351430326e-05,
+      "loss": 0.09203402996063233,
+      "mean_token_accuracy": 0.9689733698964119,
+      "num_tokens": 6947106.0,
+      "step": 3000
+    },
+    {
+      "epoch": 7.229330114665057,
+      "eval_entropy": 0.2740225845508361,
+      "eval_loss": 0.8876440525054932,
+      "eval_mean_token_accuracy": 0.8265116857678703,
+      "eval_num_tokens": 6947106.0,
+      "eval_runtime": 87.7971,
+      "eval_samples_per_second": 16.174,
+      "eval_steps_per_second": 2.027,
+      "step": 3000
+    },
+    {
+      "entropy": 0.12116114404052496,
+      "epoch": 7.27761013880507,
+      "grad_norm": 0.5467398166656494,
+      "learning_rate": 7.890406302404618e-05,
+      "loss": 0.0933056652545929,
+      "mean_token_accuracy": 0.9676455594599247,
+      "num_tokens": 6992784.0,
+      "step": 3020
+    },
+    {
+      "epoch": 7.27761013880507,
+      "eval_entropy": 0.27238342793804876,
+      "eval_loss": 0.9166117310523987,
+      "eval_mean_token_accuracy": 0.825499015242866,
+      "eval_num_tokens": 6992784.0,
+      "eval_runtime": 88.3276,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 3020
+    },
+    {
+      "entropy": 0.11715119928121567,
+      "epoch": 7.325890162945082,
+      "grad_norm": 0.4976702332496643,
+      "learning_rate": 7.634274596487176e-05,
+      "loss": 0.09449659585952759,
+      "mean_token_accuracy": 0.966506478190422,
+      "num_tokens": 7040482.0,
+      "step": 3040
+    },
+    {
+      "epoch": 7.325890162945082,
+      "eval_entropy": 0.2788975556077582,
+      "eval_loss": 0.8833065629005432,
+      "eval_mean_token_accuracy": 0.8269920596915684,
+      "eval_num_tokens": 7040482.0,
+      "eval_runtime": 88.9016,
+      "eval_samples_per_second": 15.973,
+      "eval_steps_per_second": 2.002,
+      "step": 3040
+    },
+    {
+      "entropy": 0.12565911412239075,
+      "epoch": 7.3741701870850935,
+      "grad_norm": 0.5728679895401001,
+      "learning_rate": 7.381306715921025e-05,
+      "loss": 0.09626244902610778,
+      "mean_token_accuracy": 0.9661444149911403,
+      "num_tokens": 7086261.0,
+      "step": 3060
+    },
+    {
+      "epoch": 7.3741701870850935,
+      "eval_entropy": 0.2762492011939542,
+      "eval_loss": 0.8961160778999329,
+      "eval_mean_token_accuracy": 0.826369910092836,
+      "eval_num_tokens": 7086261.0,
+      "eval_runtime": 88.5204,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 3060
+    },
+    {
+      "entropy": 0.12258567921817302,
+      "epoch": 7.422450211225105,
+      "grad_norm": 0.468985915184021,
+      "learning_rate": 7.131574247624031e-05,
+      "loss": 0.09120266437530518,
+      "mean_token_accuracy": 0.9674938447773457,
+      "num_tokens": 7133247.0,
+      "step": 3080
+    },
+    {
+      "epoch": 7.422450211225105,
+      "eval_entropy": 0.2688512086366,
+      "eval_loss": 0.9048654437065125,
+      "eval_mean_token_accuracy": 0.8265582302982888,
+      "eval_num_tokens": 7133247.0,
+      "eval_runtime": 88.164,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 3080
+    },
+    {
+      "entropy": 0.12070536343380808,
+      "epoch": 7.470730235365117,
+      "grad_norm": 0.5310841798782349,
+      "learning_rate": 6.885147862930688e-05,
+      "loss": 0.09386247992515565,
+      "mean_token_accuracy": 0.96664222702384,
+      "num_tokens": 7177810.0,
+      "step": 3100
+    },
+    {
+      "epoch": 7.470730235365117,
+      "eval_entropy": 0.26289632874593305,
+      "eval_loss": 0.902837336063385,
+      "eval_mean_token_accuracy": 0.828052184554968,
+      "eval_num_tokens": 7177810.0,
+      "eval_runtime": 88.1201,
+      "eval_samples_per_second": 16.114,
+      "eval_steps_per_second": 2.02,
+      "step": 3100
+    },
+    {
+      "entropy": 0.11322665181942285,
+      "epoch": 7.51901025950513,
+      "grad_norm": 0.6846067905426025,
+      "learning_rate": 6.642097297592908e-05,
+      "loss": 0.08772538304328918,
+      "mean_token_accuracy": 0.9691362135112286,
+      "num_tokens": 7225302.0,
+      "step": 3120
+    },
+    {
+      "epoch": 7.51901025950513,
+      "eval_entropy": 0.2670605613776807,
+      "eval_loss": 0.9130566120147705,
+      "eval_mean_token_accuracy": 0.826366903741708,
+      "eval_num_tokens": 7225302.0,
+      "eval_runtime": 88.6035,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3120
+    },
+    {
+      "entropy": 0.10985926580615342,
+      "epoch": 7.567290283645142,
+      "grad_norm": 0.4799748361110687,
+      "learning_rate": 6.402491332045683e-05,
+      "loss": 0.08672296404838561,
+      "mean_token_accuracy": 0.9698824048042297,
+      "num_tokens": 7274881.0,
+      "step": 3140
+    },
+    {
+      "epoch": 7.567290283645142,
+      "eval_entropy": 0.26657010866015146,
+      "eval_loss": 0.9056380987167358,
+      "eval_mean_token_accuracy": 0.8272326461384806,
+      "eval_num_tokens": 7274881.0,
+      "eval_runtime": 88.2967,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3140
+    },
+    {
+      "entropy": 0.11890473761595785,
+      "epoch": 7.615570307785154,
+      "grad_norm": 0.4081912934780121,
+      "learning_rate": 6.166397771942987e-05,
+      "loss": 0.09396827220916748,
+      "mean_token_accuracy": 0.967512384802103,
+      "num_tokens": 7321163.0,
+      "step": 3160
+    },
+    {
+      "epoch": 7.615570307785154,
+      "eval_entropy": 0.26364776575833226,
+      "eval_loss": 0.9204733967781067,
+      "eval_mean_token_accuracy": 0.827419729714983,
+      "eval_num_tokens": 7321163.0,
+      "eval_runtime": 88.3441,
+      "eval_samples_per_second": 16.074,
+      "eval_steps_per_second": 2.015,
+      "step": 3160
+    },
+    {
+      "entropy": 0.12209134344011545,
+      "epoch": 7.663850331925166,
+      "grad_norm": 0.5159774422645569,
+      "learning_rate": 5.93388342896962e-05,
+      "loss": 0.09292978644371033,
+      "mean_token_accuracy": 0.9658421002328396,
+      "num_tokens": 7366217.0,
+      "step": 3180
+    },
+    {
+      "epoch": 7.663850331925166,
+      "eval_entropy": 0.2714110068725736,
+      "eval_loss": 0.9143092036247253,
+      "eval_mean_token_accuracy": 0.8262251099843657,
+      "eval_num_tokens": 7366217.0,
+      "eval_runtime": 87.9635,
+      "eval_samples_per_second": 16.143,
+      "eval_steps_per_second": 2.024,
+      "step": 3180
+    },
+    {
+      "entropy": 0.11955609126016498,
+      "epoch": 7.712130356065178,
+      "grad_norm": 0.7005976438522339,
+      "learning_rate": 5.705014101934252e-05,
+      "loss": 0.09319776892662049,
+      "mean_token_accuracy": 0.9674513377249241,
+      "num_tokens": 7410887.0,
+      "step": 3200
+    },
+    {
+      "epoch": 7.712130356065178,
+      "eval_entropy": 0.2626291513442993,
+      "eval_loss": 0.91884446144104,
+      "eval_mean_token_accuracy": 0.827563955207889,
+      "eval_num_tokens": 7410887.0,
+      "eval_runtime": 87.8849,
+      "eval_samples_per_second": 16.157,
+      "eval_steps_per_second": 2.025,
+      "step": 3200
+    },
+    {
+      "entropy": 0.11767532899975777,
+      "epoch": 7.76041038020519,
+      "grad_norm": 0.4675362706184387,
+      "learning_rate": 5.479854558149187e-05,
+      "loss": 0.09038087725639343,
+      "mean_token_accuracy": 0.9669680051505566,
+      "num_tokens": 7457578.0,
+      "step": 3220
+    },
+    {
+      "epoch": 7.76041038020519,
+      "eval_entropy": 0.26581037203582486,
+      "eval_loss": 0.9146883487701416,
+      "eval_mean_token_accuracy": 0.8270345812433222,
+      "eval_num_tokens": 7457578.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 3220
+    },
+    {
+      "entropy": 0.11383061073720455,
+      "epoch": 7.808690404345202,
+      "grad_norm": 0.541247546672821,
+      "learning_rate": 5.258468515101918e-05,
+      "loss": 0.08942405581474304,
+      "mean_token_accuracy": 0.968398705124855,
+      "num_tokens": 7505761.0,
+      "step": 3240
+    },
+    {
+      "epoch": 7.808690404345202,
+      "eval_entropy": 0.26687998960861997,
+      "eval_loss": 0.9037085175514221,
+      "eval_mean_token_accuracy": 0.8274977863504646,
+      "eval_num_tokens": 7505761.0,
+      "eval_runtime": 88.5163,
+      "eval_samples_per_second": 16.042,
+      "eval_steps_per_second": 2.011,
+      "step": 3240
+    },
+    {
+      "entropy": 0.12004058873280883,
+      "epoch": 7.856970428485214,
+      "grad_norm": 0.5289552211761475,
+      "learning_rate": 5.040918622423899e-05,
+      "loss": 0.09064805507659912,
+      "mean_token_accuracy": 0.9673330299556255,
+      "num_tokens": 7552557.0,
+      "step": 3260
+    },
+    {
+      "epoch": 7.856970428485214,
+      "eval_entropy": 0.26288379251622085,
+      "eval_loss": 0.9146178364753723,
+      "eval_mean_token_accuracy": 0.8276662448149049,
+      "eval_num_tokens": 7552557.0,
+      "eval_runtime": 88.0723,
+      "eval_samples_per_second": 16.123,
+      "eval_steps_per_second": 2.021,
+      "step": 3260
+    },
+    {
+      "entropy": 0.1180689456872642,
+      "epoch": 7.905250452625227,
+      "grad_norm": 0.4754861891269684,
+      "learning_rate": 4.827266444161377e-05,
+      "loss": 0.09402655363082886,
+      "mean_token_accuracy": 0.9663361787796021,
+      "num_tokens": 7596998.0,
+      "step": 3280
+    },
+    {
+      "epoch": 7.905250452625227,
+      "eval_entropy": 0.261690996922134,
+      "eval_loss": 0.9120186567306519,
+      "eval_mean_token_accuracy": 0.8278995015647974,
+      "eval_num_tokens": 7596998.0,
+      "eval_runtime": 88.6573,
+      "eval_samples_per_second": 16.017,
+      "eval_steps_per_second": 2.008,
+      "step": 3280
+    },
+    {
+      "entropy": 0.1162164019420743,
+      "epoch": 7.953530476765239,
+      "grad_norm": 0.6203398108482361,
+      "learning_rate": 4.61757244135355e-05,
+      "loss": 0.08832768201828003,
+      "mean_token_accuracy": 0.9694220684468746,
+      "num_tokens": 7643288.0,
+      "step": 3300
+    },
+    {
+      "epoch": 7.953530476765239,
+      "eval_entropy": 0.26158678129817664,
+      "eval_loss": 0.9301361441612244,
+      "eval_mean_token_accuracy": 0.8269603885291668,
+      "eval_num_tokens": 7643288.0,
+      "eval_runtime": 88.7064,
+      "eval_samples_per_second": 16.008,
+      "eval_steps_per_second": 2.007,
+      "step": 3300
+    },
+    {
+      "entropy": 0.10877202204489089,
+      "epoch": 8.0,
+      "grad_norm": 3.058673143386841,
+      "learning_rate": 4.411895954922763e-05,
+      "loss": 0.09159615635871887,
+      "mean_token_accuracy": 0.9686878834451947,
+      "num_tokens": 7688608.0,
+      "step": 3320
+    },
+    {
+      "epoch": 8.0,
+      "eval_entropy": 0.2590438057197614,
+      "eval_loss": 0.9262065887451172,
+      "eval_mean_token_accuracy": 0.8277404546067956,
+      "eval_num_tokens": 7688608.0,
+      "eval_runtime": 87.7557,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 3320
+    },
+    {
+      "entropy": 0.09969092034734786,
+      "epoch": 8.048280024140013,
+      "grad_norm": 0.2561911642551422,
+      "learning_rate": 4.2102951888817644e-05,
+      "loss": 0.07228735685348511,
+      "mean_token_accuracy": 0.9743556551635265,
+      "num_tokens": 7732856.0,
+      "step": 3340
+    },
+    {
+      "epoch": 8.048280024140013,
+      "eval_entropy": 0.2462534454933713,
+      "eval_loss": 0.9755983352661133,
+      "eval_mean_token_accuracy": 0.827510209565752,
+      "eval_num_tokens": 7732856.0,
+      "eval_runtime": 88.0297,
+      "eval_samples_per_second": 16.131,
+      "eval_steps_per_second": 2.022,
+      "step": 3340
+    },
+    {
+      "entropy": 0.09477624730207027,
+      "epoch": 8.096560048280024,
+      "grad_norm": 0.3670356571674347,
+      "learning_rate": 4.012827193862633e-05,
+      "loss": 0.0686036467552185,
+      "mean_token_accuracy": 0.9743670664727688,
+      "num_tokens": 7778585.0,
+      "step": 3360
+    },
+    {
+      "epoch": 8.096560048280024,
+      "eval_entropy": 0.2440882951356052,
+      "eval_loss": 1.0118423700332642,
+      "eval_mean_token_accuracy": 0.8271241070849172,
+      "eval_num_tokens": 7778585.0,
+      "eval_runtime": 88.1364,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 3360
+    },
+    {
+      "entropy": 0.09417980071157217,
+      "epoch": 8.144840072420036,
+      "grad_norm": 0.5785111784934998,
+      "learning_rate": 3.819547850972133e-05,
+      "loss": 0.07100698947906495,
+      "mean_token_accuracy": 0.9737804099917412,
+      "num_tokens": 7824287.0,
+      "step": 3380
+    },
+    {
+      "epoch": 8.144840072420036,
+      "eval_entropy": 0.24603172860453637,
+      "eval_loss": 1.0021991729736328,
+      "eval_mean_token_accuracy": 0.827162122793412,
+      "eval_num_tokens": 7824287.0,
+      "eval_runtime": 88.4534,
+      "eval_samples_per_second": 16.054,
+      "eval_steps_per_second": 2.012,
+      "step": 3380
+    },
+    {
+      "entropy": 0.087222247896716,
+      "epoch": 8.193120096560047,
+      "grad_norm": 0.44848504662513733,
+      "learning_rate": 3.6305118559780393e-05,
+      "loss": 0.06325234174728393,
+      "mean_token_accuracy": 0.9769335232675076,
+      "num_tokens": 7874784.0,
+      "step": 3400
+    },
+    {
+      "epoch": 8.193120096560047,
+      "eval_entropy": 0.24544863905129807,
+      "eval_loss": 1.005008339881897,
+      "eval_mean_token_accuracy": 0.8273035240976998,
+      "eval_num_tokens": 7874784.0,
+      "eval_runtime": 88.6737,
+      "eval_samples_per_second": 16.014,
+      "eval_steps_per_second": 2.007,
+      "step": 3400
+    },
+    {
+      "entropy": 0.09037275062873959,
+      "epoch": 8.24140012070006,
+      "grad_norm": 0.23529155552387238,
+      "learning_rate": 3.4457727038308435e-05,
+      "loss": 0.06589569449424744,
+      "mean_token_accuracy": 0.9751470290124417,
+      "num_tokens": 7925240.0,
+      "step": 3420
+    },
+    {
+      "epoch": 8.24140012070006,
+      "eval_entropy": 0.24203423494368456,
+      "eval_loss": 1.0052824020385742,
+      "eval_mean_token_accuracy": 0.8275575503874361,
+      "eval_num_tokens": 7925240.0,
+      "eval_runtime": 88.3486,
+      "eval_samples_per_second": 16.073,
+      "eval_steps_per_second": 2.015,
+      "step": 3420
+    },
+    {
+      "entropy": 0.10373208015225828,
+      "epoch": 8.289680144840073,
+      "grad_norm": 0.24796408414840698,
+      "learning_rate": 3.2653826735253594e-05,
+      "loss": 0.07714771628379821,
+      "mean_token_accuracy": 0.9722891598939896,
+      "num_tokens": 7966685.0,
+      "step": 3440
+    },
+    {
+      "epoch": 8.289680144840073,
+      "eval_entropy": 0.23461157050025597,
+      "eval_loss": 1.012108325958252,
+      "eval_mean_token_accuracy": 0.82825900564033,
+      "eval_num_tokens": 7966685.0,
+      "eval_runtime": 88.1018,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 3440
+    },
+    {
+      "entropy": 0.08967954274266958,
+      "epoch": 8.337960168980084,
+      "grad_norm": 0.5314674973487854,
+      "learning_rate": 3.08939281330635e-05,
+      "loss": 0.06746266484260559,
+      "mean_token_accuracy": 0.9753133699297905,
+      "num_tokens": 8013656.0,
+      "step": 3460
+    },
+    {
+      "epoch": 8.337960168980084,
+      "eval_entropy": 0.23781998589467468,
+      "eval_loss": 1.0119913816452026,
+      "eval_mean_token_accuracy": 0.8273287157664139,
+      "eval_num_tokens": 8013656.0,
+      "eval_runtime": 88.6038,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3460
+    },
+    {
+      "entropy": 0.09280927330255509,
+      "epoch": 8.386240193120097,
+      "grad_norm": 0.30658936500549316,
+      "learning_rate": 2.9178529262224946e-05,
+      "loss": 0.06936421990394592,
+      "mean_token_accuracy": 0.9742105580866337,
+      "num_tokens": 8061108.0,
+      "step": 3480
+    },
+    {
+      "epoch": 8.386240193120097,
+      "eval_entropy": 0.23672426734747512,
+      "eval_loss": 1.0162041187286377,
+      "eval_mean_token_accuracy": 0.8278394281194451,
+      "eval_num_tokens": 8061108.0,
+      "eval_runtime": 88.6015,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 3480
+    },
+    {
+      "entropy": 0.09195122267119586,
+      "epoch": 8.43452021726011,
+      "grad_norm": 0.24512454867362976,
+      "learning_rate": 2.7508115560327256e-05,
+      "loss": 0.06996064782142639,
+      "mean_token_accuracy": 0.974104854464531,
+      "num_tokens": 8106924.0,
+      "step": 3500
+    },
+    {
+      "epoch": 8.43452021726011,
+      "eval_entropy": 0.23758245828781235,
+      "eval_loss": 1.0157747268676758,
+      "eval_mean_token_accuracy": 0.8278501579600773,
+      "eval_num_tokens": 8106924.0,
+      "eval_runtime": 88.5678,
+      "eval_samples_per_second": 16.033,
+      "eval_steps_per_second": 2.01,
+      "step": 3500
+    },
+    {
+      "entropy": 0.09518477590754629,
+      "epoch": 8.48280024140012,
+      "grad_norm": 0.2373228371143341,
+      "learning_rate": 2.5883159734688868e-05,
+      "loss": 0.06832797527313232,
+      "mean_token_accuracy": 0.9743806682527065,
+      "num_tokens": 8153070.0,
+      "step": 3520
+    },
+    {
+      "epoch": 8.48280024140012,
+      "eval_entropy": 0.2379308074210467,
+      "eval_loss": 1.0180530548095703,
+      "eval_mean_token_accuracy": 0.8282477052024241,
+      "eval_num_tokens": 8153070.0,
+      "eval_runtime": 88.4651,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 3520
+    },
+    {
+      "entropy": 0.0906269280705601,
+      "epoch": 8.531080265540133,
+      "grad_norm": 0.37602657079696655,
+      "learning_rate": 2.4304121628587173e-05,
+      "loss": 0.06921588778495788,
+      "mean_token_accuracy": 0.9745571479201317,
+      "num_tokens": 8200066.0,
+      "step": 3540
+    },
+    {
+      "epoch": 8.531080265540133,
+      "eval_entropy": 0.2357490769263064,
+      "eval_loss": 1.017139196395874,
+      "eval_mean_token_accuracy": 0.828713178299786,
+      "eval_num_tokens": 8200066.0,
+      "eval_runtime": 88.6424,
+      "eval_samples_per_second": 16.019,
+      "eval_steps_per_second": 2.008,
+      "step": 3540
+    },
+    {
+      "entropy": 0.09273334196768701,
+      "epoch": 8.579360289680144,
+      "grad_norm": 0.2978838086128235,
+      "learning_rate": 2.277144809112794e-05,
+      "loss": 0.06970887780189514,
+      "mean_token_accuracy": 0.9751374177634716,
+      "num_tokens": 8246157.0,
+      "step": 3560
+    },
+    {
+      "epoch": 8.579360289680144,
+      "eval_entropy": 0.23590933138065123,
+      "eval_loss": 1.019231915473938,
+      "eval_mean_token_accuracy": 0.8279163412163767,
+      "eval_num_tokens": 8246157.0,
+      "eval_runtime": 88.2691,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 3560
+    },
+    {
+      "entropy": 0.09854201921261847,
+      "epoch": 8.627640313820157,
+      "grad_norm": 0.5381243228912354,
+      "learning_rate": 2.1285572850792413e-05,
+      "loss": 0.07123501300811767,
+      "mean_token_accuracy": 0.9723988175392151,
+      "num_tokens": 8291752.0,
+      "step": 3580
+    },
+    {
+      "epoch": 8.627640313820157,
+      "eval_entropy": 0.23592024949494372,
+      "eval_loss": 1.020250678062439,
+      "eval_mean_token_accuracy": 0.8282466226079491,
+      "eval_num_tokens": 8291752.0,
+      "eval_runtime": 87.8939,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 3580
+    },
+    {
+      "entropy": 0.09092000788077712,
+      "epoch": 8.67592033796017,
+      "grad_norm": 0.29089561104774475,
+      "learning_rate": 1.9846916392697493e-05,
+      "loss": 0.06982720494270325,
+      "mean_token_accuracy": 0.9749995484948158,
+      "num_tokens": 8337651.0,
+      "step": 3600
+    },
+    {
+      "epoch": 8.67592033796017,
+      "eval_entropy": 0.23658913382318583,
+      "eval_loss": 1.019862413406372,
+      "eval_mean_token_accuracy": 0.8280127842104836,
+      "eval_num_tokens": 8337651.0,
+      "eval_runtime": 88.2855,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 3600
+    },
+    {
+      "entropy": 0.08708424470387399,
+      "epoch": 8.72420036210018,
+      "grad_norm": 0.29904839396476746,
+      "learning_rate": 1.8455885839603106e-05,
+      "loss": 0.06801370978355407,
+      "mean_token_accuracy": 0.9755801670253277,
+      "num_tokens": 8385330.0,
+      "step": 3620
+    },
+    {
+      "epoch": 8.72420036210018,
+      "eval_entropy": 0.23627035771862845,
+      "eval_loss": 1.0204569101333618,
+      "eval_mean_token_accuracy": 0.8279967894045155,
+      "eval_num_tokens": 8385330.0,
+      "eval_runtime": 88.1753,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 3620
+    },
+    {
+      "entropy": 0.0916255566291511,
+      "epoch": 8.772480386240193,
+      "grad_norm": 0.20571033656597137,
+      "learning_rate": 1.7112874836701502e-05,
+      "loss": 0.0692249894142151,
+      "mean_token_accuracy": 0.9745148062705994,
+      "num_tokens": 8432134.0,
+      "step": 3640
+    },
+    {
+      "epoch": 8.772480386240193,
+      "eval_entropy": 0.2340821772813797,
+      "eval_loss": 1.0212821960449219,
+      "eval_mean_token_accuracy": 0.828373622358515,
+      "eval_num_tokens": 8432134.0,
+      "eval_runtime": 87.2518,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 3640
+    },
+    {
+      "entropy": 0.0926868049427867,
+      "epoch": 8.820760410380204,
+      "grad_norm": 0.2618265450000763,
+      "learning_rate": 1.5818263440220203e-05,
+      "loss": 0.07338647246360779,
+      "mean_token_accuracy": 0.9724708944559097,
+      "num_tokens": 8476758.0,
+      "step": 3660
+    },
+    {
+      "epoch": 8.820760410380204,
+      "eval_entropy": 0.23637074575330433,
+      "eval_loss": 1.0186725854873657,
+      "eval_mean_token_accuracy": 0.8277707605549459,
+      "eval_num_tokens": 8476758.0,
+      "eval_runtime": 88.4695,
+      "eval_samples_per_second": 16.051,
+      "eval_steps_per_second": 2.012,
+      "step": 3660
+    },
+    {
+      "entropy": 0.09175782897509635,
+      "epoch": 8.869040434520217,
+      "grad_norm": 0.2889852821826935,
+      "learning_rate": 1.4572418009870604e-05,
+      "loss": 0.07066041231155396,
+      "mean_token_accuracy": 0.9725673109292984,
+      "num_tokens": 8523294.0,
+      "step": 3680
+    },
+    {
+      "epoch": 8.869040434520217,
+      "eval_entropy": 0.23945164404223473,
+      "eval_loss": 1.0145231485366821,
+      "eval_mean_token_accuracy": 0.8281202523895864,
+      "eval_num_tokens": 8523294.0,
+      "eval_runtime": 88.368,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 3680
+    },
+    {
+      "entropy": 0.09265611860901117,
+      "epoch": 8.91732045866023,
+      "grad_norm": 0.41092950105667114,
+      "learning_rate": 1.3375691105172625e-05,
+      "loss": 0.06975027322769164,
+      "mean_token_accuracy": 0.974636297672987,
+      "num_tokens": 8570358.0,
+      "step": 3700
+    },
+    {
+      "epoch": 8.91732045866023,
+      "eval_entropy": 0.239588919901446,
+      "eval_loss": 1.014059066772461,
+      "eval_mean_token_accuracy": 0.828028041660116,
+      "eval_num_tokens": 8570358.0,
+      "eval_runtime": 88.1039,
+      "eval_samples_per_second": 16.117,
+      "eval_steps_per_second": 2.02,
+      "step": 3700
+    },
+    {
+      "entropy": 0.0896465003490448,
+      "epoch": 8.965600482800241,
+      "grad_norm": 0.24000254273414612,
+      "learning_rate": 1.2228421385684305e-05,
+      "loss": 0.06743953824043274,
+      "mean_token_accuracy": 0.974006575345993,
+      "num_tokens": 8618246.0,
+      "step": 3720
+    },
+    {
+      "epoch": 8.965600482800241,
+      "eval_entropy": 0.2390461369846644,
+      "eval_loss": 1.011997938156128,
+      "eval_mean_token_accuracy": 0.8284533211354459,
+      "eval_num_tokens": 8618246.0,
+      "eval_runtime": 88.6202,
+      "eval_samples_per_second": 16.023,
+      "eval_steps_per_second": 2.009,
+      "step": 3720
+    },
+    {
+      "entropy": 0.09575109251520851,
+      "epoch": 9.012070006035003,
+      "grad_norm": 0.18088576197624207,
+      "learning_rate": 1.1130933515165429e-05,
+      "loss": 0.07295305132865906,
+      "mean_token_accuracy": 0.9739722857227573,
+      "num_tokens": 8661177.0,
+      "step": 3740
+    },
+    {
+      "epoch": 9.012070006035003,
+      "eval_entropy": 0.23728882387447892,
+      "eval_loss": 1.0134038925170898,
+      "eval_mean_token_accuracy": 0.8286508951294288,
+      "eval_num_tokens": 8661177.0,
+      "eval_runtime": 88.6434,
+      "eval_samples_per_second": 16.019,
+      "eval_steps_per_second": 2.008,
+      "step": 3740
+    },
+    {
+      "entropy": 0.08883201805874705,
+      "epoch": 9.060350030175016,
+      "grad_norm": 0.23120898008346558,
+      "learning_rate": 1.008353806970132e-05,
+      "loss": 0.06326142549514771,
+      "mean_token_accuracy": 0.9780020996928215,
+      "num_tokens": 8707602.0,
+      "step": 3760
+    },
+    {
+      "epoch": 9.060350030175016,
+      "eval_entropy": 0.23416905331142832,
+      "eval_loss": 1.0284100770950317,
+      "eval_mean_token_accuracy": 0.8284083213029283,
+      "eval_num_tokens": 8707602.0,
+      "eval_runtime": 88.2933,
+      "eval_samples_per_second": 16.083,
+      "eval_steps_per_second": 2.016,
+      "step": 3760
+    },
+    {
+      "entropy": 0.08861468276008963,
+      "epoch": 9.108630054315027,
+      "grad_norm": 0.17277656495571136,
+      "learning_rate": 9.086531449813489e-06,
+      "loss": 0.06727416515350342,
+      "mean_token_accuracy": 0.9760390736162663,
+      "num_tokens": 8751287.0,
+      "step": 3780
+    },
+    {
+      "epoch": 9.108630054315027,
+      "eval_entropy": 0.23370317386442355,
+      "eval_loss": 1.0381091833114624,
+      "eval_mean_token_accuracy": 0.8282668178670862,
+      "eval_num_tokens": 8751287.0,
+      "eval_runtime": 88.5738,
+      "eval_samples_per_second": 16.032,
+      "eval_steps_per_second": 2.01,
+      "step": 3780
+    },
+    {
+      "entropy": 0.09032954541034996,
+      "epoch": 9.15691007845504,
+      "grad_norm": 0.19481085240840912,
+      "learning_rate": 8.140195796581981e-06,
+      "loss": 0.06117786169052124,
+      "mean_token_accuracy": 0.9764454193413258,
+      "num_tokens": 8799982.0,
+      "step": 3800
+    },
+    {
+      "epoch": 9.15691007845504,
+      "eval_entropy": 0.23215936609868254,
+      "eval_loss": 1.0467767715454102,
+      "eval_mean_token_accuracy": 0.828405868806196,
+      "eval_num_tokens": 8799982.0,
+      "eval_runtime": 87.9551,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 3800
+    },
+    {
+      "entropy": 0.08824449330568314,
+      "epoch": 9.20519010259505,
+      "grad_norm": 0.20382793247699738,
+      "learning_rate": 7.2447989118025455e-06,
+      "loss": 0.06422553658485412,
+      "mean_token_accuracy": 0.9763121999800205,
+      "num_tokens": 8845536.0,
+      "step": 3820
+    },
+    {
+      "epoch": 9.20519010259505,
+      "eval_entropy": 0.23169661472352704,
+      "eval_loss": 1.0529043674468994,
+      "eval_mean_token_accuracy": 0.8282025610463003,
+      "eval_num_tokens": 8845536.0,
+      "eval_runtime": 88.5548,
+      "eval_samples_per_second": 16.035,
+      "eval_steps_per_second": 2.01,
+      "step": 3820
+    },
+    {
+      "entropy": 0.07768499720841646,
+      "epoch": 9.253470126735063,
+      "grad_norm": 0.11727919429540634,
+      "learning_rate": 6.400594182202084e-06,
+      "loss": 0.057279336452484134,
+      "mean_token_accuracy": 0.9795166745781898,
+      "num_tokens": 8896053.0,
+      "step": 3840
+    },
+    {
+      "epoch": 9.253470126735063,
+      "eval_entropy": 0.2301936373114586,
+      "eval_loss": 1.0576049089431763,
+      "eval_mean_token_accuracy": 0.8283417489421502,
+      "eval_num_tokens": 8896053.0,
+      "eval_runtime": 88.7854,
+      "eval_samples_per_second": 15.994,
+      "eval_steps_per_second": 2.005,
+      "step": 3840
+    },
+    {
+      "entropy": 0.09015200780704617,
+      "epoch": 9.301750150875076,
+      "grad_norm": 0.2089058756828308,
+      "learning_rate": 5.6078205077330705e-06,
+      "loss": 0.06626194715499878,
+      "mean_token_accuracy": 0.9747506938874722,
+      "num_tokens": 8940591.0,
+      "step": 3860
+    },
+    {
+      "epoch": 9.301750150875076,
+      "eval_entropy": 0.23024900676159377,
+      "eval_loss": 1.0599786043167114,
+      "eval_mean_token_accuracy": 0.8283051915383071,
+      "eval_num_tokens": 8940591.0,
+      "eval_runtime": 88.2963,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3860
+    },
+    {
+      "entropy": 0.08484094133600592,
+      "epoch": 9.350030175015087,
+      "grad_norm": 0.21646778285503387,
+      "learning_rate": 4.866702233967785e-06,
+      "loss": 0.05998039245605469,
+      "mean_token_accuracy": 0.9771868668496608,
+      "num_tokens": 8989035.0,
+      "step": 3880
+    },
+    {
+      "epoch": 9.350030175015087,
+      "eval_entropy": 0.2300723305578982,
+      "eval_loss": 1.0633587837219238,
+      "eval_mean_token_accuracy": 0.828371500701047,
+      "eval_num_tokens": 8989035.0,
+      "eval_runtime": 88.0233,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 3880
+    },
+    {
+      "entropy": 0.08905019680969417,
+      "epoch": 9.3983101991551,
+      "grad_norm": 0.21842406690120697,
+      "learning_rate": 4.177449088611095e-06,
+      "loss": 0.06666865348815917,
+      "mean_token_accuracy": 0.9753445640206337,
+      "num_tokens": 9033383.0,
+      "step": 3900
+    },
+    {
+      "epoch": 9.3983101991551,
+      "eval_entropy": 0.2295504841911659,
+      "eval_loss": 1.0662461519241333,
+      "eval_mean_token_accuracy": 0.8284122297603093,
+      "eval_num_tokens": 9033383.0,
+      "eval_runtime": 88.5525,
+      "eval_samples_per_second": 16.036,
+      "eval_steps_per_second": 2.01,
+      "step": 3900
+    },
+    {
+      "entropy": 0.08582521872594953,
+      "epoch": 9.446590223295111,
+      "grad_norm": 0.2000240683555603,
+      "learning_rate": 3.540256122150088e-06,
+      "loss": 0.06327944993972778,
+      "mean_token_accuracy": 0.9765432856976985,
+      "num_tokens": 9079672.0,
+      "step": 3920
+    },
+    {
+      "epoch": 9.446590223295111,
+      "eval_entropy": 0.2297678625818049,
+      "eval_loss": 1.0676169395446777,
+      "eval_mean_token_accuracy": 0.8283703879693921,
+      "eval_num_tokens": 9079672.0,
+      "eval_runtime": 87.6897,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 3920
+    },
+    {
+      "entropy": 0.09261773955076932,
+      "epoch": 9.494870247435124,
+      "grad_norm": 0.19104281067848206,
+      "learning_rate": 2.955303652656991e-06,
+      "loss": 0.06111966967582703,
+      "mean_token_accuracy": 0.9738784030079841,
+      "num_tokens": 9127647.0,
+      "step": 3940
+    },
+    {
+      "epoch": 9.494870247435124,
+      "eval_entropy": 0.22986975881490815,
+      "eval_loss": 1.0687925815582275,
+      "eval_mean_token_accuracy": 0.8282996858103892,
+      "eval_num_tokens": 9127647.0,
+      "eval_runtime": 88.2224,
+      "eval_samples_per_second": 16.096,
+      "eval_steps_per_second": 2.018,
+      "step": 3940
+    },
+    {
+      "entropy": 0.08347254800610245,
+      "epoch": 9.543150271575136,
+      "grad_norm": 0.18671369552612305,
+      "learning_rate": 2.422757214761356e-06,
+      "loss": 0.06384263038635254,
+      "mean_token_accuracy": 0.9770128890872002,
+      "num_tokens": 9172587.0,
+      "step": 3960
+    },
+    {
+      "epoch": 9.543150271575136,
+      "eval_entropy": 0.22975517851248217,
+      "eval_loss": 1.0701013803482056,
+      "eval_mean_token_accuracy": 0.8282244285840666,
+      "eval_num_tokens": 9172587.0,
+      "eval_runtime": 88.5794,
+      "eval_samples_per_second": 16.031,
+      "eval_steps_per_second": 2.009,
+      "step": 3960
+    },
+    {
+      "entropy": 0.08136763009242713,
+      "epoch": 9.591430295715147,
+      "grad_norm": 0.24811454117298126,
+      "learning_rate": 1.942767512805673e-06,
+      "loss": 0.06044899821281433,
+      "mean_token_accuracy": 0.9776685133576393,
+      "num_tokens": 9219278.0,
+      "step": 3980
+    },
+    {
+      "epoch": 9.591430295715147,
+      "eval_entropy": 0.22938149347064202,
+      "eval_loss": 1.0711463689804077,
+      "eval_mean_token_accuracy": 0.8282721410976367,
+      "eval_num_tokens": 9219278.0,
+      "eval_runtime": 88.4978,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 3980
+    },
+    {
+      "entropy": 0.09432941288687288,
+      "epoch": 9.63971031985516,
+      "grad_norm": 0.30733999609947205,
+      "learning_rate": 1.515470378197889e-06,
+      "loss": 0.0672213613986969,
+      "mean_token_accuracy": 0.9728612303733826,
+      "num_tokens": 9262346.0,
+      "step": 4000
+    },
+    {
+      "epoch": 9.63971031985516,
+      "eval_entropy": 0.22936245958121976,
+      "eval_loss": 1.071343183517456,
+      "eval_mean_token_accuracy": 0.8282827025049189,
+      "eval_num_tokens": 9262346.0,
+      "eval_runtime": 88.4428,
+      "eval_samples_per_second": 16.056,
+      "eval_steps_per_second": 2.013,
+      "step": 4000
+    },
+    {
+      "entropy": 0.08287148931995034,
+      "epoch": 9.687990343995171,
+      "grad_norm": 0.1976361870765686,
+      "learning_rate": 1.1409867309726256e-06,
+      "loss": 0.06318551301956177,
+      "mean_token_accuracy": 0.9770804084837437,
+      "num_tokens": 9309622.0,
+      "step": 4020
+    },
+    {
+      "epoch": 9.687990343995171,
+      "eval_entropy": 0.22929051920269314,
+      "eval_loss": 1.0711324214935303,
+      "eval_mean_token_accuracy": 0.8281597297513084,
+      "eval_num_tokens": 9309622.0,
+      "eval_runtime": 88.5243,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 4020
+    },
+    {
+      "entropy": 0.0907150611979887,
+      "epoch": 9.736270368135184,
+      "grad_norm": 0.16192017495632172,
+      "learning_rate": 8.194225455723828e-07,
+      "loss": 0.06589602828025817,
+      "mean_token_accuracy": 0.9738550461828709,
+      "num_tokens": 9354549.0,
+      "step": 4040
+    },
+    {
+      "epoch": 9.736270368135184,
+      "eval_entropy": 0.22913936391640244,
+      "eval_loss": 1.0715173482894897,
+      "eval_mean_token_accuracy": 0.8281301544623428,
+      "eval_num_tokens": 9354549.0,
+      "eval_runtime": 88.875,
+      "eval_samples_per_second": 15.978,
+      "eval_steps_per_second": 2.003,
+      "step": 4040
+    },
+    {
+      "entropy": 0.09003224167972804,
+      "epoch": 9.784550392275197,
+      "grad_norm": 0.2092711478471756,
+      "learning_rate": 5.508688208578523e-07,
+      "loss": 0.06655729413032532,
+      "mean_token_accuracy": 0.9750426560640335,
+      "num_tokens": 9399394.0,
+      "step": 4060
+    },
+    {
+      "epoch": 9.784550392275197,
+      "eval_entropy": 0.2291239777140403,
+      "eval_loss": 1.0715563297271729,
+      "eval_mean_token_accuracy": 0.8281808595978812,
+      "eval_num_tokens": 9399394.0,
+      "eval_runtime": 88.4328,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 4060
+    },
+    {
+      "entropy": 0.08719871481880545,
+      "epoch": 9.832830416415208,
+      "grad_norm": 0.17966455221176147,
+      "learning_rate": 3.3540155435645076e-07,
+      "loss": 0.05917760729789734,
+      "mean_token_accuracy": 0.9752349570393563,
+      "num_tokens": 9449483.0,
+      "step": 4080
+    },
+    {
+      "epoch": 9.832830416415208,
+      "eval_entropy": 0.2291418021816886,
+      "eval_loss": 1.0714553594589233,
+      "eval_mean_token_accuracy": 0.8281870403986299,
+      "eval_num_tokens": 9449483.0,
+      "eval_runtime": 88.2612,
+      "eval_samples_per_second": 16.089,
+      "eval_steps_per_second": 2.017,
+      "step": 4080
+    },
+    {
+      "entropy": 0.07887464743107557,
+      "epoch": 9.88111044055522,
+      "grad_norm": 0.4397202134132385,
+      "learning_rate": 1.7308172075587924e-07,
+      "loss": 0.060457843542099,
+      "mean_token_accuracy": 0.9779011741280556,
+      "num_tokens": 9497731.0,
+      "step": 4100
+    },
+    {
+      "epoch": 9.88111044055522,
+      "eval_entropy": 0.2292038949855258,
+      "eval_loss": 1.0717124938964844,
+      "eval_mean_token_accuracy": 0.8282261353530241,
+      "eval_num_tokens": 9497731.0,
+      "eval_runtime": 88.5793,
+      "eval_samples_per_second": 16.031,
+      "eval_steps_per_second": 2.009,
+      "step": 4100
+    },
+    {
+      "entropy": 0.07985925199463964,
+      "epoch": 9.929390464695233,
+      "grad_norm": 0.2129203975200653,
+      "learning_rate": 6.395525464895752e-08,
+      "loss": 0.05666370391845703,
+      "mean_token_accuracy": 0.9778168581426143,
+      "num_tokens": 9548917.0,
+      "step": 4120
+    },
+    {
+      "epoch": 9.929390464695233,
+      "eval_entropy": 0.22915430601393239,
+      "eval_loss": 1.0715631246566772,
+      "eval_mean_token_accuracy": 0.8282323436120923,
+      "eval_num_tokens": 9548917.0,
+      "eval_runtime": 88.4991,
+      "eval_samples_per_second": 16.045,
+      "eval_steps_per_second": 2.011,
+      "step": 4120
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 4.039151005944054e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4140/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4140/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4140/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4140/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4140/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4140/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4140/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4140/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4140/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4140/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4140/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4140/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4140/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4140/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..6a191bdc36f524b0aa2d0718b19c2fc6408384ea
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4140/trainer_state.json
@@ -0,0 +1,4381 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 9.977670488835244,
+  "eval_steps": 20,
+  "global_step": 4140,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    },
+    {
+      "entropy": 0.1326121488586068,
+      "epoch": 7.181050090525045,
+      "grad_norm": 0.4843117296695709,
+      "learning_rate": 8.411870386507724e-05,
+      "loss": 0.09537842869758606,
+      "mean_token_accuracy": 0.9645151920616627,
+      "num_tokens": 6898444.0,
+      "step": 2980
+    },
+    {
+      "epoch": 7.181050090525045,
+      "eval_entropy": 0.27999477783280813,
+      "eval_loss": 0.9031037092208862,
+      "eval_mean_token_accuracy": 0.8261882204018282,
+      "eval_num_tokens": 6898444.0,
+      "eval_runtime": 88.5602,
+      "eval_samples_per_second": 16.034,
+      "eval_steps_per_second": 2.01,
+      "step": 2980
+    },
+    {
+      "entropy": 0.1156751093454659,
+      "epoch": 7.229330114665057,
+      "grad_norm": 0.5297922492027283,
+      "learning_rate": 8.149629351430326e-05,
+      "loss": 0.09203402996063233,
+      "mean_token_accuracy": 0.9689733698964119,
+      "num_tokens": 6947106.0,
+      "step": 3000
+    },
+    {
+      "epoch": 7.229330114665057,
+      "eval_entropy": 0.2740225845508361,
+      "eval_loss": 0.8876440525054932,
+      "eval_mean_token_accuracy": 0.8265116857678703,
+      "eval_num_tokens": 6947106.0,
+      "eval_runtime": 87.7971,
+      "eval_samples_per_second": 16.174,
+      "eval_steps_per_second": 2.027,
+      "step": 3000
+    },
+    {
+      "entropy": 0.12116114404052496,
+      "epoch": 7.27761013880507,
+      "grad_norm": 0.5467398166656494,
+      "learning_rate": 7.890406302404618e-05,
+      "loss": 0.0933056652545929,
+      "mean_token_accuracy": 0.9676455594599247,
+      "num_tokens": 6992784.0,
+      "step": 3020
+    },
+    {
+      "epoch": 7.27761013880507,
+      "eval_entropy": 0.27238342793804876,
+      "eval_loss": 0.9166117310523987,
+      "eval_mean_token_accuracy": 0.825499015242866,
+      "eval_num_tokens": 6992784.0,
+      "eval_runtime": 88.3276,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 3020
+    },
+    {
+      "entropy": 0.11715119928121567,
+      "epoch": 7.325890162945082,
+      "grad_norm": 0.4976702332496643,
+      "learning_rate": 7.634274596487176e-05,
+      "loss": 0.09449659585952759,
+      "mean_token_accuracy": 0.966506478190422,
+      "num_tokens": 7040482.0,
+      "step": 3040
+    },
+    {
+      "epoch": 7.325890162945082,
+      "eval_entropy": 0.2788975556077582,
+      "eval_loss": 0.8833065629005432,
+      "eval_mean_token_accuracy": 0.8269920596915684,
+      "eval_num_tokens": 7040482.0,
+      "eval_runtime": 88.9016,
+      "eval_samples_per_second": 15.973,
+      "eval_steps_per_second": 2.002,
+      "step": 3040
+    },
+    {
+      "entropy": 0.12565911412239075,
+      "epoch": 7.3741701870850935,
+      "grad_norm": 0.5728679895401001,
+      "learning_rate": 7.381306715921025e-05,
+      "loss": 0.09626244902610778,
+      "mean_token_accuracy": 0.9661444149911403,
+      "num_tokens": 7086261.0,
+      "step": 3060
+    },
+    {
+      "epoch": 7.3741701870850935,
+      "eval_entropy": 0.2762492011939542,
+      "eval_loss": 0.8961160778999329,
+      "eval_mean_token_accuracy": 0.826369910092836,
+      "eval_num_tokens": 7086261.0,
+      "eval_runtime": 88.5204,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 3060
+    },
+    {
+      "entropy": 0.12258567921817302,
+      "epoch": 7.422450211225105,
+      "grad_norm": 0.468985915184021,
+      "learning_rate": 7.131574247624031e-05,
+      "loss": 0.09120266437530518,
+      "mean_token_accuracy": 0.9674938447773457,
+      "num_tokens": 7133247.0,
+      "step": 3080
+    },
+    {
+      "epoch": 7.422450211225105,
+      "eval_entropy": 0.2688512086366,
+      "eval_loss": 0.9048654437065125,
+      "eval_mean_token_accuracy": 0.8265582302982888,
+      "eval_num_tokens": 7133247.0,
+      "eval_runtime": 88.164,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 3080
+    },
+    {
+      "entropy": 0.12070536343380808,
+      "epoch": 7.470730235365117,
+      "grad_norm": 0.5310841798782349,
+      "learning_rate": 6.885147862930688e-05,
+      "loss": 0.09386247992515565,
+      "mean_token_accuracy": 0.96664222702384,
+      "num_tokens": 7177810.0,
+      "step": 3100
+    },
+    {
+      "epoch": 7.470730235365117,
+      "eval_entropy": 0.26289632874593305,
+      "eval_loss": 0.902837336063385,
+      "eval_mean_token_accuracy": 0.828052184554968,
+      "eval_num_tokens": 7177810.0,
+      "eval_runtime": 88.1201,
+      "eval_samples_per_second": 16.114,
+      "eval_steps_per_second": 2.02,
+      "step": 3100
+    },
+    {
+      "entropy": 0.11322665181942285,
+      "epoch": 7.51901025950513,
+      "grad_norm": 0.6846067905426025,
+      "learning_rate": 6.642097297592908e-05,
+      "loss": 0.08772538304328918,
+      "mean_token_accuracy": 0.9691362135112286,
+      "num_tokens": 7225302.0,
+      "step": 3120
+    },
+    {
+      "epoch": 7.51901025950513,
+      "eval_entropy": 0.2670605613776807,
+      "eval_loss": 0.9130566120147705,
+      "eval_mean_token_accuracy": 0.826366903741708,
+      "eval_num_tokens": 7225302.0,
+      "eval_runtime": 88.6035,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3120
+    },
+    {
+      "entropy": 0.10985926580615342,
+      "epoch": 7.567290283645142,
+      "grad_norm": 0.4799748361110687,
+      "learning_rate": 6.402491332045683e-05,
+      "loss": 0.08672296404838561,
+      "mean_token_accuracy": 0.9698824048042297,
+      "num_tokens": 7274881.0,
+      "step": 3140
+    },
+    {
+      "epoch": 7.567290283645142,
+      "eval_entropy": 0.26657010866015146,
+      "eval_loss": 0.9056380987167358,
+      "eval_mean_token_accuracy": 0.8272326461384806,
+      "eval_num_tokens": 7274881.0,
+      "eval_runtime": 88.2967,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3140
+    },
+    {
+      "entropy": 0.11890473761595785,
+      "epoch": 7.615570307785154,
+      "grad_norm": 0.4081912934780121,
+      "learning_rate": 6.166397771942987e-05,
+      "loss": 0.09396827220916748,
+      "mean_token_accuracy": 0.967512384802103,
+      "num_tokens": 7321163.0,
+      "step": 3160
+    },
+    {
+      "epoch": 7.615570307785154,
+      "eval_entropy": 0.26364776575833226,
+      "eval_loss": 0.9204733967781067,
+      "eval_mean_token_accuracy": 0.827419729714983,
+      "eval_num_tokens": 7321163.0,
+      "eval_runtime": 88.3441,
+      "eval_samples_per_second": 16.074,
+      "eval_steps_per_second": 2.015,
+      "step": 3160
+    },
+    {
+      "entropy": 0.12209134344011545,
+      "epoch": 7.663850331925166,
+      "grad_norm": 0.5159774422645569,
+      "learning_rate": 5.93388342896962e-05,
+      "loss": 0.09292978644371033,
+      "mean_token_accuracy": 0.9658421002328396,
+      "num_tokens": 7366217.0,
+      "step": 3180
+    },
+    {
+      "epoch": 7.663850331925166,
+      "eval_entropy": 0.2714110068725736,
+      "eval_loss": 0.9143092036247253,
+      "eval_mean_token_accuracy": 0.8262251099843657,
+      "eval_num_tokens": 7366217.0,
+      "eval_runtime": 87.9635,
+      "eval_samples_per_second": 16.143,
+      "eval_steps_per_second": 2.024,
+      "step": 3180
+    },
+    {
+      "entropy": 0.11955609126016498,
+      "epoch": 7.712130356065178,
+      "grad_norm": 0.7005976438522339,
+      "learning_rate": 5.705014101934252e-05,
+      "loss": 0.09319776892662049,
+      "mean_token_accuracy": 0.9674513377249241,
+      "num_tokens": 7410887.0,
+      "step": 3200
+    },
+    {
+      "epoch": 7.712130356065178,
+      "eval_entropy": 0.2626291513442993,
+      "eval_loss": 0.91884446144104,
+      "eval_mean_token_accuracy": 0.827563955207889,
+      "eval_num_tokens": 7410887.0,
+      "eval_runtime": 87.8849,
+      "eval_samples_per_second": 16.157,
+      "eval_steps_per_second": 2.025,
+      "step": 3200
+    },
+    {
+      "entropy": 0.11767532899975777,
+      "epoch": 7.76041038020519,
+      "grad_norm": 0.4675362706184387,
+      "learning_rate": 5.479854558149187e-05,
+      "loss": 0.09038087725639343,
+      "mean_token_accuracy": 0.9669680051505566,
+      "num_tokens": 7457578.0,
+      "step": 3220
+    },
+    {
+      "epoch": 7.76041038020519,
+      "eval_entropy": 0.26581037203582486,
+      "eval_loss": 0.9146883487701416,
+      "eval_mean_token_accuracy": 0.8270345812433222,
+      "eval_num_tokens": 7457578.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 3220
+    },
+    {
+      "entropy": 0.11383061073720455,
+      "epoch": 7.808690404345202,
+      "grad_norm": 0.541247546672821,
+      "learning_rate": 5.258468515101918e-05,
+      "loss": 0.08942405581474304,
+      "mean_token_accuracy": 0.968398705124855,
+      "num_tokens": 7505761.0,
+      "step": 3240
+    },
+    {
+      "epoch": 7.808690404345202,
+      "eval_entropy": 0.26687998960861997,
+      "eval_loss": 0.9037085175514221,
+      "eval_mean_token_accuracy": 0.8274977863504646,
+      "eval_num_tokens": 7505761.0,
+      "eval_runtime": 88.5163,
+      "eval_samples_per_second": 16.042,
+      "eval_steps_per_second": 2.011,
+      "step": 3240
+    },
+    {
+      "entropy": 0.12004058873280883,
+      "epoch": 7.856970428485214,
+      "grad_norm": 0.5289552211761475,
+      "learning_rate": 5.040918622423899e-05,
+      "loss": 0.09064805507659912,
+      "mean_token_accuracy": 0.9673330299556255,
+      "num_tokens": 7552557.0,
+      "step": 3260
+    },
+    {
+      "epoch": 7.856970428485214,
+      "eval_entropy": 0.26288379251622085,
+      "eval_loss": 0.9146178364753723,
+      "eval_mean_token_accuracy": 0.8276662448149049,
+      "eval_num_tokens": 7552557.0,
+      "eval_runtime": 88.0723,
+      "eval_samples_per_second": 16.123,
+      "eval_steps_per_second": 2.021,
+      "step": 3260
+    },
+    {
+      "entropy": 0.1180689456872642,
+      "epoch": 7.905250452625227,
+      "grad_norm": 0.4754861891269684,
+      "learning_rate": 4.827266444161377e-05,
+      "loss": 0.09402655363082886,
+      "mean_token_accuracy": 0.9663361787796021,
+      "num_tokens": 7596998.0,
+      "step": 3280
+    },
+    {
+      "epoch": 7.905250452625227,
+      "eval_entropy": 0.261690996922134,
+      "eval_loss": 0.9120186567306519,
+      "eval_mean_token_accuracy": 0.8278995015647974,
+      "eval_num_tokens": 7596998.0,
+      "eval_runtime": 88.6573,
+      "eval_samples_per_second": 16.017,
+      "eval_steps_per_second": 2.008,
+      "step": 3280
+    },
+    {
+      "entropy": 0.1162164019420743,
+      "epoch": 7.953530476765239,
+      "grad_norm": 0.6203398108482361,
+      "learning_rate": 4.61757244135355e-05,
+      "loss": 0.08832768201828003,
+      "mean_token_accuracy": 0.9694220684468746,
+      "num_tokens": 7643288.0,
+      "step": 3300
+    },
+    {
+      "epoch": 7.953530476765239,
+      "eval_entropy": 0.26158678129817664,
+      "eval_loss": 0.9301361441612244,
+      "eval_mean_token_accuracy": 0.8269603885291668,
+      "eval_num_tokens": 7643288.0,
+      "eval_runtime": 88.7064,
+      "eval_samples_per_second": 16.008,
+      "eval_steps_per_second": 2.007,
+      "step": 3300
+    },
+    {
+      "entropy": 0.10877202204489089,
+      "epoch": 8.0,
+      "grad_norm": 3.058673143386841,
+      "learning_rate": 4.411895954922763e-05,
+      "loss": 0.09159615635871887,
+      "mean_token_accuracy": 0.9686878834451947,
+      "num_tokens": 7688608.0,
+      "step": 3320
+    },
+    {
+      "epoch": 8.0,
+      "eval_entropy": 0.2590438057197614,
+      "eval_loss": 0.9262065887451172,
+      "eval_mean_token_accuracy": 0.8277404546067956,
+      "eval_num_tokens": 7688608.0,
+      "eval_runtime": 87.7557,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 3320
+    },
+    {
+      "entropy": 0.09969092034734786,
+      "epoch": 8.048280024140013,
+      "grad_norm": 0.2561911642551422,
+      "learning_rate": 4.2102951888817644e-05,
+      "loss": 0.07228735685348511,
+      "mean_token_accuracy": 0.9743556551635265,
+      "num_tokens": 7732856.0,
+      "step": 3340
+    },
+    {
+      "epoch": 8.048280024140013,
+      "eval_entropy": 0.2462534454933713,
+      "eval_loss": 0.9755983352661133,
+      "eval_mean_token_accuracy": 0.827510209565752,
+      "eval_num_tokens": 7732856.0,
+      "eval_runtime": 88.0297,
+      "eval_samples_per_second": 16.131,
+      "eval_steps_per_second": 2.022,
+      "step": 3340
+    },
+    {
+      "entropy": 0.09477624730207027,
+      "epoch": 8.096560048280024,
+      "grad_norm": 0.3670356571674347,
+      "learning_rate": 4.012827193862633e-05,
+      "loss": 0.0686036467552185,
+      "mean_token_accuracy": 0.9743670664727688,
+      "num_tokens": 7778585.0,
+      "step": 3360
+    },
+    {
+      "epoch": 8.096560048280024,
+      "eval_entropy": 0.2440882951356052,
+      "eval_loss": 1.0118423700332642,
+      "eval_mean_token_accuracy": 0.8271241070849172,
+      "eval_num_tokens": 7778585.0,
+      "eval_runtime": 88.1364,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 3360
+    },
+    {
+      "entropy": 0.09417980071157217,
+      "epoch": 8.144840072420036,
+      "grad_norm": 0.5785111784934998,
+      "learning_rate": 3.819547850972133e-05,
+      "loss": 0.07100698947906495,
+      "mean_token_accuracy": 0.9737804099917412,
+      "num_tokens": 7824287.0,
+      "step": 3380
+    },
+    {
+      "epoch": 8.144840072420036,
+      "eval_entropy": 0.24603172860453637,
+      "eval_loss": 1.0021991729736328,
+      "eval_mean_token_accuracy": 0.827162122793412,
+      "eval_num_tokens": 7824287.0,
+      "eval_runtime": 88.4534,
+      "eval_samples_per_second": 16.054,
+      "eval_steps_per_second": 2.012,
+      "step": 3380
+    },
+    {
+      "entropy": 0.087222247896716,
+      "epoch": 8.193120096560047,
+      "grad_norm": 0.44848504662513733,
+      "learning_rate": 3.6305118559780393e-05,
+      "loss": 0.06325234174728393,
+      "mean_token_accuracy": 0.9769335232675076,
+      "num_tokens": 7874784.0,
+      "step": 3400
+    },
+    {
+      "epoch": 8.193120096560047,
+      "eval_entropy": 0.24544863905129807,
+      "eval_loss": 1.005008339881897,
+      "eval_mean_token_accuracy": 0.8273035240976998,
+      "eval_num_tokens": 7874784.0,
+      "eval_runtime": 88.6737,
+      "eval_samples_per_second": 16.014,
+      "eval_steps_per_second": 2.007,
+      "step": 3400
+    },
+    {
+      "entropy": 0.09037275062873959,
+      "epoch": 8.24140012070006,
+      "grad_norm": 0.23529155552387238,
+      "learning_rate": 3.4457727038308435e-05,
+      "loss": 0.06589569449424744,
+      "mean_token_accuracy": 0.9751470290124417,
+      "num_tokens": 7925240.0,
+      "step": 3420
+    },
+    {
+      "epoch": 8.24140012070006,
+      "eval_entropy": 0.24203423494368456,
+      "eval_loss": 1.0052824020385742,
+      "eval_mean_token_accuracy": 0.8275575503874361,
+      "eval_num_tokens": 7925240.0,
+      "eval_runtime": 88.3486,
+      "eval_samples_per_second": 16.073,
+      "eval_steps_per_second": 2.015,
+      "step": 3420
+    },
+    {
+      "entropy": 0.10373208015225828,
+      "epoch": 8.289680144840073,
+      "grad_norm": 0.24796408414840698,
+      "learning_rate": 3.2653826735253594e-05,
+      "loss": 0.07714771628379821,
+      "mean_token_accuracy": 0.9722891598939896,
+      "num_tokens": 7966685.0,
+      "step": 3440
+    },
+    {
+      "epoch": 8.289680144840073,
+      "eval_entropy": 0.23461157050025597,
+      "eval_loss": 1.012108325958252,
+      "eval_mean_token_accuracy": 0.82825900564033,
+      "eval_num_tokens": 7966685.0,
+      "eval_runtime": 88.1018,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 3440
+    },
+    {
+      "entropy": 0.08967954274266958,
+      "epoch": 8.337960168980084,
+      "grad_norm": 0.5314674973487854,
+      "learning_rate": 3.08939281330635e-05,
+      "loss": 0.06746266484260559,
+      "mean_token_accuracy": 0.9753133699297905,
+      "num_tokens": 8013656.0,
+      "step": 3460
+    },
+    {
+      "epoch": 8.337960168980084,
+      "eval_entropy": 0.23781998589467468,
+      "eval_loss": 1.0119913816452026,
+      "eval_mean_token_accuracy": 0.8273287157664139,
+      "eval_num_tokens": 8013656.0,
+      "eval_runtime": 88.6038,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3460
+    },
+    {
+      "entropy": 0.09280927330255509,
+      "epoch": 8.386240193120097,
+      "grad_norm": 0.30658936500549316,
+      "learning_rate": 2.9178529262224946e-05,
+      "loss": 0.06936421990394592,
+      "mean_token_accuracy": 0.9742105580866337,
+      "num_tokens": 8061108.0,
+      "step": 3480
+    },
+    {
+      "epoch": 8.386240193120097,
+      "eval_entropy": 0.23672426734747512,
+      "eval_loss": 1.0162041187286377,
+      "eval_mean_token_accuracy": 0.8278394281194451,
+      "eval_num_tokens": 8061108.0,
+      "eval_runtime": 88.6015,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 3480
+    },
+    {
+      "entropy": 0.09195122267119586,
+      "epoch": 8.43452021726011,
+      "grad_norm": 0.24512454867362976,
+      "learning_rate": 2.7508115560327256e-05,
+      "loss": 0.06996064782142639,
+      "mean_token_accuracy": 0.974104854464531,
+      "num_tokens": 8106924.0,
+      "step": 3500
+    },
+    {
+      "epoch": 8.43452021726011,
+      "eval_entropy": 0.23758245828781235,
+      "eval_loss": 1.0157747268676758,
+      "eval_mean_token_accuracy": 0.8278501579600773,
+      "eval_num_tokens": 8106924.0,
+      "eval_runtime": 88.5678,
+      "eval_samples_per_second": 16.033,
+      "eval_steps_per_second": 2.01,
+      "step": 3500
+    },
+    {
+      "entropy": 0.09518477590754629,
+      "epoch": 8.48280024140012,
+      "grad_norm": 0.2373228371143341,
+      "learning_rate": 2.5883159734688868e-05,
+      "loss": 0.06832797527313232,
+      "mean_token_accuracy": 0.9743806682527065,
+      "num_tokens": 8153070.0,
+      "step": 3520
+    },
+    {
+      "epoch": 8.48280024140012,
+      "eval_entropy": 0.2379308074210467,
+      "eval_loss": 1.0180530548095703,
+      "eval_mean_token_accuracy": 0.8282477052024241,
+      "eval_num_tokens": 8153070.0,
+      "eval_runtime": 88.4651,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 3520
+    },
+    {
+      "entropy": 0.0906269280705601,
+      "epoch": 8.531080265540133,
+      "grad_norm": 0.37602657079696655,
+      "learning_rate": 2.4304121628587173e-05,
+      "loss": 0.06921588778495788,
+      "mean_token_accuracy": 0.9745571479201317,
+      "num_tokens": 8200066.0,
+      "step": 3540
+    },
+    {
+      "epoch": 8.531080265540133,
+      "eval_entropy": 0.2357490769263064,
+      "eval_loss": 1.017139196395874,
+      "eval_mean_token_accuracy": 0.828713178299786,
+      "eval_num_tokens": 8200066.0,
+      "eval_runtime": 88.6424,
+      "eval_samples_per_second": 16.019,
+      "eval_steps_per_second": 2.008,
+      "step": 3540
+    },
+    {
+      "entropy": 0.09273334196768701,
+      "epoch": 8.579360289680144,
+      "grad_norm": 0.2978838086128235,
+      "learning_rate": 2.277144809112794e-05,
+      "loss": 0.06970887780189514,
+      "mean_token_accuracy": 0.9751374177634716,
+      "num_tokens": 8246157.0,
+      "step": 3560
+    },
+    {
+      "epoch": 8.579360289680144,
+      "eval_entropy": 0.23590933138065123,
+      "eval_loss": 1.019231915473938,
+      "eval_mean_token_accuracy": 0.8279163412163767,
+      "eval_num_tokens": 8246157.0,
+      "eval_runtime": 88.2691,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 3560
+    },
+    {
+      "entropy": 0.09854201921261847,
+      "epoch": 8.627640313820157,
+      "grad_norm": 0.5381243228912354,
+      "learning_rate": 2.1285572850792413e-05,
+      "loss": 0.07123501300811767,
+      "mean_token_accuracy": 0.9723988175392151,
+      "num_tokens": 8291752.0,
+      "step": 3580
+    },
+    {
+      "epoch": 8.627640313820157,
+      "eval_entropy": 0.23592024949494372,
+      "eval_loss": 1.020250678062439,
+      "eval_mean_token_accuracy": 0.8282466226079491,
+      "eval_num_tokens": 8291752.0,
+      "eval_runtime": 87.8939,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 3580
+    },
+    {
+      "entropy": 0.09092000788077712,
+      "epoch": 8.67592033796017,
+      "grad_norm": 0.29089561104774475,
+      "learning_rate": 1.9846916392697493e-05,
+      "loss": 0.06982720494270325,
+      "mean_token_accuracy": 0.9749995484948158,
+      "num_tokens": 8337651.0,
+      "step": 3600
+    },
+    {
+      "epoch": 8.67592033796017,
+      "eval_entropy": 0.23658913382318583,
+      "eval_loss": 1.019862413406372,
+      "eval_mean_token_accuracy": 0.8280127842104836,
+      "eval_num_tokens": 8337651.0,
+      "eval_runtime": 88.2855,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 3600
+    },
+    {
+      "entropy": 0.08708424470387399,
+      "epoch": 8.72420036210018,
+      "grad_norm": 0.29904839396476746,
+      "learning_rate": 1.8455885839603106e-05,
+      "loss": 0.06801370978355407,
+      "mean_token_accuracy": 0.9755801670253277,
+      "num_tokens": 8385330.0,
+      "step": 3620
+    },
+    {
+      "epoch": 8.72420036210018,
+      "eval_entropy": 0.23627035771862845,
+      "eval_loss": 1.0204569101333618,
+      "eval_mean_token_accuracy": 0.8279967894045155,
+      "eval_num_tokens": 8385330.0,
+      "eval_runtime": 88.1753,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 3620
+    },
+    {
+      "entropy": 0.0916255566291511,
+      "epoch": 8.772480386240193,
+      "grad_norm": 0.20571033656597137,
+      "learning_rate": 1.7112874836701502e-05,
+      "loss": 0.0692249894142151,
+      "mean_token_accuracy": 0.9745148062705994,
+      "num_tokens": 8432134.0,
+      "step": 3640
+    },
+    {
+      "epoch": 8.772480386240193,
+      "eval_entropy": 0.2340821772813797,
+      "eval_loss": 1.0212821960449219,
+      "eval_mean_token_accuracy": 0.828373622358515,
+      "eval_num_tokens": 8432134.0,
+      "eval_runtime": 87.2518,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 3640
+    },
+    {
+      "entropy": 0.0926868049427867,
+      "epoch": 8.820760410380204,
+      "grad_norm": 0.2618265450000763,
+      "learning_rate": 1.5818263440220203e-05,
+      "loss": 0.07338647246360779,
+      "mean_token_accuracy": 0.9724708944559097,
+      "num_tokens": 8476758.0,
+      "step": 3660
+    },
+    {
+      "epoch": 8.820760410380204,
+      "eval_entropy": 0.23637074575330433,
+      "eval_loss": 1.0186725854873657,
+      "eval_mean_token_accuracy": 0.8277707605549459,
+      "eval_num_tokens": 8476758.0,
+      "eval_runtime": 88.4695,
+      "eval_samples_per_second": 16.051,
+      "eval_steps_per_second": 2.012,
+      "step": 3660
+    },
+    {
+      "entropy": 0.09175782897509635,
+      "epoch": 8.869040434520217,
+      "grad_norm": 0.2889852821826935,
+      "learning_rate": 1.4572418009870604e-05,
+      "loss": 0.07066041231155396,
+      "mean_token_accuracy": 0.9725673109292984,
+      "num_tokens": 8523294.0,
+      "step": 3680
+    },
+    {
+      "epoch": 8.869040434520217,
+      "eval_entropy": 0.23945164404223473,
+      "eval_loss": 1.0145231485366821,
+      "eval_mean_token_accuracy": 0.8281202523895864,
+      "eval_num_tokens": 8523294.0,
+      "eval_runtime": 88.368,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 3680
+    },
+    {
+      "entropy": 0.09265611860901117,
+      "epoch": 8.91732045866023,
+      "grad_norm": 0.41092950105667114,
+      "learning_rate": 1.3375691105172625e-05,
+      "loss": 0.06975027322769164,
+      "mean_token_accuracy": 0.974636297672987,
+      "num_tokens": 8570358.0,
+      "step": 3700
+    },
+    {
+      "epoch": 8.91732045866023,
+      "eval_entropy": 0.239588919901446,
+      "eval_loss": 1.014059066772461,
+      "eval_mean_token_accuracy": 0.828028041660116,
+      "eval_num_tokens": 8570358.0,
+      "eval_runtime": 88.1039,
+      "eval_samples_per_second": 16.117,
+      "eval_steps_per_second": 2.02,
+      "step": 3700
+    },
+    {
+      "entropy": 0.0896465003490448,
+      "epoch": 8.965600482800241,
+      "grad_norm": 0.24000254273414612,
+      "learning_rate": 1.2228421385684305e-05,
+      "loss": 0.06743953824043274,
+      "mean_token_accuracy": 0.974006575345993,
+      "num_tokens": 8618246.0,
+      "step": 3720
+    },
+    {
+      "epoch": 8.965600482800241,
+      "eval_entropy": 0.2390461369846644,
+      "eval_loss": 1.011997938156128,
+      "eval_mean_token_accuracy": 0.8284533211354459,
+      "eval_num_tokens": 8618246.0,
+      "eval_runtime": 88.6202,
+      "eval_samples_per_second": 16.023,
+      "eval_steps_per_second": 2.009,
+      "step": 3720
+    },
+    {
+      "entropy": 0.09575109251520851,
+      "epoch": 9.012070006035003,
+      "grad_norm": 0.18088576197624207,
+      "learning_rate": 1.1130933515165429e-05,
+      "loss": 0.07295305132865906,
+      "mean_token_accuracy": 0.9739722857227573,
+      "num_tokens": 8661177.0,
+      "step": 3740
+    },
+    {
+      "epoch": 9.012070006035003,
+      "eval_entropy": 0.23728882387447892,
+      "eval_loss": 1.0134038925170898,
+      "eval_mean_token_accuracy": 0.8286508951294288,
+      "eval_num_tokens": 8661177.0,
+      "eval_runtime": 88.6434,
+      "eval_samples_per_second": 16.019,
+      "eval_steps_per_second": 2.008,
+      "step": 3740
+    },
+    {
+      "entropy": 0.08883201805874705,
+      "epoch": 9.060350030175016,
+      "grad_norm": 0.23120898008346558,
+      "learning_rate": 1.008353806970132e-05,
+      "loss": 0.06326142549514771,
+      "mean_token_accuracy": 0.9780020996928215,
+      "num_tokens": 8707602.0,
+      "step": 3760
+    },
+    {
+      "epoch": 9.060350030175016,
+      "eval_entropy": 0.23416905331142832,
+      "eval_loss": 1.0284100770950317,
+      "eval_mean_token_accuracy": 0.8284083213029283,
+      "eval_num_tokens": 8707602.0,
+      "eval_runtime": 88.2933,
+      "eval_samples_per_second": 16.083,
+      "eval_steps_per_second": 2.016,
+      "step": 3760
+    },
+    {
+      "entropy": 0.08861468276008963,
+      "epoch": 9.108630054315027,
+      "grad_norm": 0.17277656495571136,
+      "learning_rate": 9.086531449813489e-06,
+      "loss": 0.06727416515350342,
+      "mean_token_accuracy": 0.9760390736162663,
+      "num_tokens": 8751287.0,
+      "step": 3780
+    },
+    {
+      "epoch": 9.108630054315027,
+      "eval_entropy": 0.23370317386442355,
+      "eval_loss": 1.0381091833114624,
+      "eval_mean_token_accuracy": 0.8282668178670862,
+      "eval_num_tokens": 8751287.0,
+      "eval_runtime": 88.5738,
+      "eval_samples_per_second": 16.032,
+      "eval_steps_per_second": 2.01,
+      "step": 3780
+    },
+    {
+      "entropy": 0.09032954541034996,
+      "epoch": 9.15691007845504,
+      "grad_norm": 0.19481085240840912,
+      "learning_rate": 8.140195796581981e-06,
+      "loss": 0.06117786169052124,
+      "mean_token_accuracy": 0.9764454193413258,
+      "num_tokens": 8799982.0,
+      "step": 3800
+    },
+    {
+      "epoch": 9.15691007845504,
+      "eval_entropy": 0.23215936609868254,
+      "eval_loss": 1.0467767715454102,
+      "eval_mean_token_accuracy": 0.828405868806196,
+      "eval_num_tokens": 8799982.0,
+      "eval_runtime": 87.9551,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 3800
+    },
+    {
+      "entropy": 0.08824449330568314,
+      "epoch": 9.20519010259505,
+      "grad_norm": 0.20382793247699738,
+      "learning_rate": 7.2447989118025455e-06,
+      "loss": 0.06422553658485412,
+      "mean_token_accuracy": 0.9763121999800205,
+      "num_tokens": 8845536.0,
+      "step": 3820
+    },
+    {
+      "epoch": 9.20519010259505,
+      "eval_entropy": 0.23169661472352704,
+      "eval_loss": 1.0529043674468994,
+      "eval_mean_token_accuracy": 0.8282025610463003,
+      "eval_num_tokens": 8845536.0,
+      "eval_runtime": 88.5548,
+      "eval_samples_per_second": 16.035,
+      "eval_steps_per_second": 2.01,
+      "step": 3820
+    },
+    {
+      "entropy": 0.07768499720841646,
+      "epoch": 9.253470126735063,
+      "grad_norm": 0.11727919429540634,
+      "learning_rate": 6.400594182202084e-06,
+      "loss": 0.057279336452484134,
+      "mean_token_accuracy": 0.9795166745781898,
+      "num_tokens": 8896053.0,
+      "step": 3840
+    },
+    {
+      "epoch": 9.253470126735063,
+      "eval_entropy": 0.2301936373114586,
+      "eval_loss": 1.0576049089431763,
+      "eval_mean_token_accuracy": 0.8283417489421502,
+      "eval_num_tokens": 8896053.0,
+      "eval_runtime": 88.7854,
+      "eval_samples_per_second": 15.994,
+      "eval_steps_per_second": 2.005,
+      "step": 3840
+    },
+    {
+      "entropy": 0.09015200780704617,
+      "epoch": 9.301750150875076,
+      "grad_norm": 0.2089058756828308,
+      "learning_rate": 5.6078205077330705e-06,
+      "loss": 0.06626194715499878,
+      "mean_token_accuracy": 0.9747506938874722,
+      "num_tokens": 8940591.0,
+      "step": 3860
+    },
+    {
+      "epoch": 9.301750150875076,
+      "eval_entropy": 0.23024900676159377,
+      "eval_loss": 1.0599786043167114,
+      "eval_mean_token_accuracy": 0.8283051915383071,
+      "eval_num_tokens": 8940591.0,
+      "eval_runtime": 88.2963,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3860
+    },
+    {
+      "entropy": 0.08484094133600592,
+      "epoch": 9.350030175015087,
+      "grad_norm": 0.21646778285503387,
+      "learning_rate": 4.866702233967785e-06,
+      "loss": 0.05998039245605469,
+      "mean_token_accuracy": 0.9771868668496608,
+      "num_tokens": 8989035.0,
+      "step": 3880
+    },
+    {
+      "epoch": 9.350030175015087,
+      "eval_entropy": 0.2300723305578982,
+      "eval_loss": 1.0633587837219238,
+      "eval_mean_token_accuracy": 0.828371500701047,
+      "eval_num_tokens": 8989035.0,
+      "eval_runtime": 88.0233,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 3880
+    },
+    {
+      "entropy": 0.08905019680969417,
+      "epoch": 9.3983101991551,
+      "grad_norm": 0.21842406690120697,
+      "learning_rate": 4.177449088611095e-06,
+      "loss": 0.06666865348815917,
+      "mean_token_accuracy": 0.9753445640206337,
+      "num_tokens": 9033383.0,
+      "step": 3900
+    },
+    {
+      "epoch": 9.3983101991551,
+      "eval_entropy": 0.2295504841911659,
+      "eval_loss": 1.0662461519241333,
+      "eval_mean_token_accuracy": 0.8284122297603093,
+      "eval_num_tokens": 9033383.0,
+      "eval_runtime": 88.5525,
+      "eval_samples_per_second": 16.036,
+      "eval_steps_per_second": 2.01,
+      "step": 3900
+    },
+    {
+      "entropy": 0.08582521872594953,
+      "epoch": 9.446590223295111,
+      "grad_norm": 0.2000240683555603,
+      "learning_rate": 3.540256122150088e-06,
+      "loss": 0.06327944993972778,
+      "mean_token_accuracy": 0.9765432856976985,
+      "num_tokens": 9079672.0,
+      "step": 3920
+    },
+    {
+      "epoch": 9.446590223295111,
+      "eval_entropy": 0.2297678625818049,
+      "eval_loss": 1.0676169395446777,
+      "eval_mean_token_accuracy": 0.8283703879693921,
+      "eval_num_tokens": 9079672.0,
+      "eval_runtime": 87.6897,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 3920
+    },
+    {
+      "entropy": 0.09261773955076932,
+      "epoch": 9.494870247435124,
+      "grad_norm": 0.19104281067848206,
+      "learning_rate": 2.955303652656991e-06,
+      "loss": 0.06111966967582703,
+      "mean_token_accuracy": 0.9738784030079841,
+      "num_tokens": 9127647.0,
+      "step": 3940
+    },
+    {
+      "epoch": 9.494870247435124,
+      "eval_entropy": 0.22986975881490815,
+      "eval_loss": 1.0687925815582275,
+      "eval_mean_token_accuracy": 0.8282996858103892,
+      "eval_num_tokens": 9127647.0,
+      "eval_runtime": 88.2224,
+      "eval_samples_per_second": 16.096,
+      "eval_steps_per_second": 2.018,
+      "step": 3940
+    },
+    {
+      "entropy": 0.08347254800610245,
+      "epoch": 9.543150271575136,
+      "grad_norm": 0.18671369552612305,
+      "learning_rate": 2.422757214761356e-06,
+      "loss": 0.06384263038635254,
+      "mean_token_accuracy": 0.9770128890872002,
+      "num_tokens": 9172587.0,
+      "step": 3960
+    },
+    {
+      "epoch": 9.543150271575136,
+      "eval_entropy": 0.22975517851248217,
+      "eval_loss": 1.0701013803482056,
+      "eval_mean_token_accuracy": 0.8282244285840666,
+      "eval_num_tokens": 9172587.0,
+      "eval_runtime": 88.5794,
+      "eval_samples_per_second": 16.031,
+      "eval_steps_per_second": 2.009,
+      "step": 3960
+    },
+    {
+      "entropy": 0.08136763009242713,
+      "epoch": 9.591430295715147,
+      "grad_norm": 0.24811454117298126,
+      "learning_rate": 1.942767512805673e-06,
+      "loss": 0.06044899821281433,
+      "mean_token_accuracy": 0.9776685133576393,
+      "num_tokens": 9219278.0,
+      "step": 3980
+    },
+    {
+      "epoch": 9.591430295715147,
+      "eval_entropy": 0.22938149347064202,
+      "eval_loss": 1.0711463689804077,
+      "eval_mean_token_accuracy": 0.8282721410976367,
+      "eval_num_tokens": 9219278.0,
+      "eval_runtime": 88.4978,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 3980
+    },
+    {
+      "entropy": 0.09432941288687288,
+      "epoch": 9.63971031985516,
+      "grad_norm": 0.30733999609947205,
+      "learning_rate": 1.515470378197889e-06,
+      "loss": 0.0672213613986969,
+      "mean_token_accuracy": 0.9728612303733826,
+      "num_tokens": 9262346.0,
+      "step": 4000
+    },
+    {
+      "epoch": 9.63971031985516,
+      "eval_entropy": 0.22936245958121976,
+      "eval_loss": 1.071343183517456,
+      "eval_mean_token_accuracy": 0.8282827025049189,
+      "eval_num_tokens": 9262346.0,
+      "eval_runtime": 88.4428,
+      "eval_samples_per_second": 16.056,
+      "eval_steps_per_second": 2.013,
+      "step": 4000
+    },
+    {
+      "entropy": 0.08287148931995034,
+      "epoch": 9.687990343995171,
+      "grad_norm": 0.1976361870765686,
+      "learning_rate": 1.1409867309726256e-06,
+      "loss": 0.06318551301956177,
+      "mean_token_accuracy": 0.9770804084837437,
+      "num_tokens": 9309622.0,
+      "step": 4020
+    },
+    {
+      "epoch": 9.687990343995171,
+      "eval_entropy": 0.22929051920269314,
+      "eval_loss": 1.0711324214935303,
+      "eval_mean_token_accuracy": 0.8281597297513084,
+      "eval_num_tokens": 9309622.0,
+      "eval_runtime": 88.5243,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 4020
+    },
+    {
+      "entropy": 0.0907150611979887,
+      "epoch": 9.736270368135184,
+      "grad_norm": 0.16192017495632172,
+      "learning_rate": 8.194225455723828e-07,
+      "loss": 0.06589602828025817,
+      "mean_token_accuracy": 0.9738550461828709,
+      "num_tokens": 9354549.0,
+      "step": 4040
+    },
+    {
+      "epoch": 9.736270368135184,
+      "eval_entropy": 0.22913936391640244,
+      "eval_loss": 1.0715173482894897,
+      "eval_mean_token_accuracy": 0.8281301544623428,
+      "eval_num_tokens": 9354549.0,
+      "eval_runtime": 88.875,
+      "eval_samples_per_second": 15.978,
+      "eval_steps_per_second": 2.003,
+      "step": 4040
+    },
+    {
+      "entropy": 0.09003224167972804,
+      "epoch": 9.784550392275197,
+      "grad_norm": 0.2092711478471756,
+      "learning_rate": 5.508688208578523e-07,
+      "loss": 0.06655729413032532,
+      "mean_token_accuracy": 0.9750426560640335,
+      "num_tokens": 9399394.0,
+      "step": 4060
+    },
+    {
+      "epoch": 9.784550392275197,
+      "eval_entropy": 0.2291239777140403,
+      "eval_loss": 1.0715563297271729,
+      "eval_mean_token_accuracy": 0.8281808595978812,
+      "eval_num_tokens": 9399394.0,
+      "eval_runtime": 88.4328,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 4060
+    },
+    {
+      "entropy": 0.08719871481880545,
+      "epoch": 9.832830416415208,
+      "grad_norm": 0.17966455221176147,
+      "learning_rate": 3.3540155435645076e-07,
+      "loss": 0.05917760729789734,
+      "mean_token_accuracy": 0.9752349570393563,
+      "num_tokens": 9449483.0,
+      "step": 4080
+    },
+    {
+      "epoch": 9.832830416415208,
+      "eval_entropy": 0.2291418021816886,
+      "eval_loss": 1.0714553594589233,
+      "eval_mean_token_accuracy": 0.8281870403986299,
+      "eval_num_tokens": 9449483.0,
+      "eval_runtime": 88.2612,
+      "eval_samples_per_second": 16.089,
+      "eval_steps_per_second": 2.017,
+      "step": 4080
+    },
+    {
+      "entropy": 0.07887464743107557,
+      "epoch": 9.88111044055522,
+      "grad_norm": 0.4397202134132385,
+      "learning_rate": 1.7308172075587924e-07,
+      "loss": 0.060457843542099,
+      "mean_token_accuracy": 0.9779011741280556,
+      "num_tokens": 9497731.0,
+      "step": 4100
+    },
+    {
+      "epoch": 9.88111044055522,
+      "eval_entropy": 0.2292038949855258,
+      "eval_loss": 1.0717124938964844,
+      "eval_mean_token_accuracy": 0.8282261353530241,
+      "eval_num_tokens": 9497731.0,
+      "eval_runtime": 88.5793,
+      "eval_samples_per_second": 16.031,
+      "eval_steps_per_second": 2.009,
+      "step": 4100
+    },
+    {
+      "entropy": 0.07985925199463964,
+      "epoch": 9.929390464695233,
+      "grad_norm": 0.2129203975200653,
+      "learning_rate": 6.395525464895752e-08,
+      "loss": 0.05666370391845703,
+      "mean_token_accuracy": 0.9778168581426143,
+      "num_tokens": 9548917.0,
+      "step": 4120
+    },
+    {
+      "epoch": 9.929390464695233,
+      "eval_entropy": 0.22915430601393239,
+      "eval_loss": 1.0715631246566772,
+      "eval_mean_token_accuracy": 0.8282323436120923,
+      "eval_num_tokens": 9548917.0,
+      "eval_runtime": 88.4991,
+      "eval_samples_per_second": 16.045,
+      "eval_steps_per_second": 2.011,
+      "step": 4120
+    },
+    {
+      "entropy": 0.08802500045858323,
+      "epoch": 9.977670488835244,
+      "grad_norm": 0.18367069959640503,
+      "learning_rate": 8.053037534806845e-09,
+      "loss": 0.06633861660957337,
+      "mean_token_accuracy": 0.9749571368098259,
+      "num_tokens": 9592427.0,
+      "step": 4140
+    },
+    {
+      "epoch": 9.977670488835244,
+      "eval_entropy": 0.22921692956699413,
+      "eval_loss": 1.0715234279632568,
+      "eval_mean_token_accuracy": 0.828055245822735,
+      "eval_num_tokens": 9592427.0,
+      "eval_runtime": 88.1869,
+      "eval_samples_per_second": 16.102,
+      "eval_steps_per_second": 2.018,
+      "step": 4140
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 4.0573616119372186e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4150/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4150/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4150/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4150/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4150/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4150/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4150/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4150/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4150/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4150/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4150/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4150/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4150/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4150/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..d86404192cf4dbc4a16df525ed84826e7a1d7dd1
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-4150/trainer_state.json
@@ -0,0 +1,4392 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 10.0,
+  "eval_steps": 20,
+  "global_step": 4150,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    },
+    {
+      "entropy": 0.6050681818276644,
+      "epoch": 2.4103802051901027,
+      "grad_norm": 0.9151566028594971,
+      "learning_rate": 0.00035404068900628076,
+      "loss": 0.559452486038208,
+      "mean_token_accuracy": 0.8310242861509323,
+      "num_tokens": 2319305.0,
+      "step": 1000
+    },
+    {
+      "epoch": 2.4103802051901027,
+      "eval_entropy": 0.6634890969549672,
+      "eval_loss": 0.6692465543746948,
+      "eval_mean_token_accuracy": 0.8134353673190213,
+      "eval_num_tokens": 2319305.0,
+      "eval_runtime": 87.7711,
+      "eval_samples_per_second": 16.178,
+      "eval_steps_per_second": 2.028,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6169968567788601,
+      "epoch": 2.4586602293301145,
+      "grad_norm": 0.7081006765365601,
+      "learning_rate": 0.0003525243462115402,
+      "loss": 0.5589694023132324,
+      "mean_token_accuracy": 0.8357502184808254,
+      "num_tokens": 2368592.0,
+      "step": 1020
+    },
+    {
+      "epoch": 2.4586602293301145,
+      "eval_entropy": 0.6126393578695447,
+      "eval_loss": 0.6706712245941162,
+      "eval_mean_token_accuracy": 0.8155577500884452,
+      "eval_num_tokens": 2368592.0,
+      "eval_runtime": 88.1407,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.019,
+      "step": 1020
+    },
+    {
+      "entropy": 0.6072248566895724,
+      "epoch": 2.506940253470127,
+      "grad_norm": 0.6999370455741882,
+      "learning_rate": 0.0003509614855443561,
+      "loss": 0.5500593185424805,
+      "mean_token_accuracy": 0.8337264291942119,
+      "num_tokens": 2412859.0,
+      "step": 1040
+    },
+    {
+      "epoch": 2.506940253470127,
+      "eval_entropy": 0.6417417074187418,
+      "eval_loss": 0.6651163697242737,
+      "eval_mean_token_accuracy": 0.8159113908081912,
+      "eval_num_tokens": 2412859.0,
+      "eval_runtime": 87.4854,
+      "eval_samples_per_second": 16.231,
+      "eval_steps_per_second": 2.035,
+      "step": 1040
+    },
+    {
+      "entropy": 0.6301592070609331,
+      "epoch": 2.5552202776101387,
+      "grad_norm": 0.8768910765647888,
+      "learning_rate": 0.00034935254927581064,
+      "loss": 0.5613903999328613,
+      "mean_token_accuracy": 0.8298896946012974,
+      "num_tokens": 2458201.0,
+      "step": 1060
+    },
+    {
+      "epoch": 2.5552202776101387,
+      "eval_entropy": 0.6569794751285167,
+      "eval_loss": 0.6668341159820557,
+      "eval_mean_token_accuracy": 0.8126791795987761,
+      "eval_num_tokens": 2458201.0,
+      "eval_runtime": 88.0474,
+      "eval_samples_per_second": 16.128,
+      "eval_steps_per_second": 2.022,
+      "step": 1060
+    },
+    {
+      "entropy": 0.6069021724164486,
+      "epoch": 2.603500301750151,
+      "grad_norm": 0.9033508896827698,
+      "learning_rate": 0.0003476979927158357,
+      "loss": 0.5590654373168945,
+      "mean_token_accuracy": 0.831520090252161,
+      "num_tokens": 2505382.0,
+      "step": 1080
+    },
+    {
+      "epoch": 2.603500301750151,
+      "eval_entropy": 0.6636380777600106,
+      "eval_loss": 0.6596238017082214,
+      "eval_mean_token_accuracy": 0.8177202826135614,
+      "eval_num_tokens": 2505382.0,
+      "eval_runtime": 88.08,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1080
+    },
+    {
+      "entropy": 0.6138021353632211,
+      "epoch": 2.651780325890163,
+      "grad_norm": 0.7345808148384094,
+      "learning_rate": 0.0003459982840843664,
+      "loss": 0.5586410522460937,
+      "mean_token_accuracy": 0.8295876093208789,
+      "num_tokens": 2552927.0,
+      "step": 1100
+    },
+    {
+      "epoch": 2.651780325890163,
+      "eval_entropy": 0.5845904127600487,
+      "eval_loss": 0.663780689239502,
+      "eval_mean_token_accuracy": 0.8142095107710763,
+      "eval_num_tokens": 2552927.0,
+      "eval_runtime": 88.0083,
+      "eval_samples_per_second": 16.135,
+      "eval_steps_per_second": 2.023,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6251844003796577,
+      "epoch": 2.700060350030175,
+      "grad_norm": 0.7587347626686096,
+      "learning_rate": 0.00034425390437883976,
+      "loss": 0.5720802307128906,
+      "mean_token_accuracy": 0.8285771444439888,
+      "num_tokens": 2597426.0,
+      "step": 1120
+    },
+    {
+      "epoch": 2.700060350030175,
+      "eval_entropy": 0.6251207221759839,
+      "eval_loss": 0.6556326746940613,
+      "eval_mean_token_accuracy": 0.8181774036937886,
+      "eval_num_tokens": 2597426.0,
+      "eval_runtime": 87.8517,
+      "eval_samples_per_second": 16.164,
+      "eval_steps_per_second": 2.026,
+      "step": 1120
+    },
+    {
+      "entropy": 0.6227757651358843,
+      "epoch": 2.748340374170187,
+      "grad_norm": 0.9254975914955139,
+      "learning_rate": 0.00034246534723807843,
+      "loss": 0.5767871856689453,
+      "mean_token_accuracy": 0.8292960874736309,
+      "num_tokens": 2640514.0,
+      "step": 1140
+    },
+    {
+      "epoch": 2.748340374170187,
+      "eval_entropy": 0.6580858331048087,
+      "eval_loss": 0.6579018831253052,
+      "eval_mean_token_accuracy": 0.8178701273510965,
+      "eval_num_tokens": 2640514.0,
+      "eval_runtime": 87.9523,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 1140
+    },
+    {
+      "entropy": 0.6353523321449757,
+      "epoch": 2.796620398310199,
+      "grad_norm": 1.4437663555145264,
+      "learning_rate": 0.00034063311880259663,
+      "loss": 0.566702127456665,
+      "mean_token_accuracy": 0.831210870295763,
+      "num_tokens": 2687730.0,
+      "step": 1160
+    },
+    {
+      "epoch": 2.796620398310199,
+      "eval_entropy": 0.6347215423423253,
+      "eval_loss": 0.6531020402908325,
+      "eval_mean_token_accuracy": 0.8189373260803436,
+      "eval_num_tokens": 2687730.0,
+      "eval_runtime": 88.0207,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1160
+    },
+    {
+      "entropy": 0.6233414955437183,
+      "epoch": 2.8449004224502112,
+      "grad_norm": 0.753302812576294,
+      "learning_rate": 0.0003387577375713684,
+      "loss": 0.5740012645721435,
+      "mean_token_accuracy": 0.8270302519202233,
+      "num_tokens": 2732930.0,
+      "step": 1180
+    },
+    {
+      "epoch": 2.8449004224502112,
+      "eval_entropy": 0.6291426423560368,
+      "eval_loss": 0.6549608707427979,
+      "eval_mean_token_accuracy": 0.8188753693961026,
+      "eval_num_tokens": 2732930.0,
+      "eval_runtime": 88.2306,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1180
+    },
+    {
+      "entropy": 0.617981968075037,
+      "epoch": 2.8931804465902236,
+      "grad_norm": 0.6777492165565491,
+      "learning_rate": 0.00033683973425509874,
+      "loss": 0.5558822631835938,
+      "mean_token_accuracy": 0.8315935507416725,
+      "num_tokens": 2779450.0,
+      "step": 1200
+    },
+    {
+      "epoch": 2.8931804465902236,
+      "eval_entropy": 0.6340839514571629,
+      "eval_loss": 0.6490646600723267,
+      "eval_mean_token_accuracy": 0.8200173856837026,
+      "eval_num_tokens": 2779450.0,
+      "eval_runtime": 88.3076,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5989151727408171,
+      "epoch": 2.9414604707302354,
+      "grad_norm": 1.0503766536712646,
+      "learning_rate": 0.0003348796516260381,
+      "loss": 0.550087308883667,
+      "mean_token_accuracy": 0.8357349015772343,
+      "num_tokens": 2827819.0,
+      "step": 1220
+    },
+    {
+      "epoch": 2.9414604707302354,
+      "eval_entropy": 0.6683844726407127,
+      "eval_loss": 0.6455520987510681,
+      "eval_mean_token_accuracy": 0.8199967946229356,
+      "eval_num_tokens": 2827819.0,
+      "eval_runtime": 87.7825,
+      "eval_samples_per_second": 16.176,
+      "eval_steps_per_second": 2.028,
+      "step": 1220
+    },
+    {
+      "entropy": 0.6396764755249024,
+      "epoch": 2.9897404948702473,
+      "grad_norm": 0.7088311910629272,
+      "learning_rate": 0.00033287804436438515,
+      "loss": 0.5731344223022461,
+      "mean_token_accuracy": 0.8298504173755645,
+      "num_tokens": 2872493.0,
+      "step": 1240
+    },
+    {
+      "epoch": 2.9897404948702473,
+      "eval_entropy": 0.6510144954317072,
+      "eval_loss": 0.6509669423103333,
+      "eval_mean_token_accuracy": 0.8188618431600292,
+      "eval_num_tokens": 2872493.0,
+      "eval_runtime": 87.8954,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 1240
+    },
+    {
+      "entropy": 0.5436535887903982,
+      "epoch": 3.036210018105009,
+      "grad_norm": 0.8652629852294922,
+      "learning_rate": 0.00033083547890131815,
+      "loss": 0.4816119194030762,
+      "mean_token_accuracy": 0.8523989869402601,
+      "num_tokens": 2918249.0,
+      "step": 1260
+    },
+    {
+      "epoch": 3.036210018105009,
+      "eval_entropy": 0.5676146091035242,
+      "eval_loss": 0.6723794341087341,
+      "eval_mean_token_accuracy": 0.8197960103495737,
+      "eval_num_tokens": 2918249.0,
+      "eval_runtime": 88.1303,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1260
+    },
+    {
+      "entropy": 0.503798121958971,
+      "epoch": 3.084490042245021,
+      "grad_norm": 0.9403247237205505,
+      "learning_rate": 0.0003287525332587019,
+      "loss": 0.43859500885009767,
+      "mean_token_accuracy": 0.8604862734675407,
+      "num_tokens": 2968353.0,
+      "step": 1280
+    },
+    {
+      "epoch": 3.084490042245021,
+      "eval_entropy": 0.5981672088081917,
+      "eval_loss": 0.671315610408783,
+      "eval_mean_token_accuracy": 0.8175233046660263,
+      "eval_num_tokens": 2968353.0,
+      "eval_runtime": 88.3097,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1280
+    },
+    {
+      "entropy": 0.5218576818704606,
+      "epoch": 3.1327700663850333,
+      "grad_norm": 1.0271061658859253,
+      "learning_rate": 0.0003266297968855144,
+      "loss": 0.45096940994262696,
+      "mean_token_accuracy": 0.8555397011339665,
+      "num_tokens": 3018291.0,
+      "step": 1300
+    },
+    {
+      "epoch": 3.1327700663850333,
+      "eval_entropy": 0.5708769342202819,
+      "eval_loss": 0.6631077527999878,
+      "eval_mean_token_accuracy": 0.8207426647121987,
+      "eval_num_tokens": 3018291.0,
+      "eval_runtime": 87.7272,
+      "eval_samples_per_second": 16.187,
+      "eval_steps_per_second": 2.029,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5191332377493382,
+      "epoch": 3.181050090525045,
+      "grad_norm": 0.958055317401886,
+      "learning_rate": 0.0003244678704910397,
+      "loss": 0.45679512023925783,
+      "mean_token_accuracy": 0.8554921194911003,
+      "num_tokens": 3070712.0,
+      "step": 1320
+    },
+    {
+      "epoch": 3.181050090525045,
+      "eval_entropy": 0.5703962919082535,
+      "eval_loss": 0.6597784161567688,
+      "eval_mean_token_accuracy": 0.821146364962117,
+      "eval_num_tokens": 3070712.0,
+      "eval_runtime": 88.4154,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 1320
+    },
+    {
+      "entropy": 0.5207117382436991,
+      "epoch": 3.2293301146650575,
+      "grad_norm": 0.7903274297714233,
+      "learning_rate": 0.00032226736587487434,
+      "loss": 0.44762430191040037,
+      "mean_token_accuracy": 0.8578196220099926,
+      "num_tokens": 3116564.0,
+      "step": 1340
+    },
+    {
+      "epoch": 3.2293301146650575,
+      "eval_entropy": 0.5453257858753204,
+      "eval_loss": 0.6688455939292908,
+      "eval_mean_token_accuracy": 0.8215647212574991,
+      "eval_num_tokens": 3116564.0,
+      "eval_runtime": 88.0214,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 1340
+    },
+    {
+      "entropy": 0.5263226825743914,
+      "epoch": 3.2776101388050694,
+      "grad_norm": 0.8782205581665039,
+      "learning_rate": 0.00032002890575379524,
+      "loss": 0.46592864990234373,
+      "mean_token_accuracy": 0.8551752805709839,
+      "num_tokens": 3164054.0,
+      "step": 1360
+    },
+    {
+      "epoch": 3.2776101388050694,
+      "eval_entropy": 0.5805480952343244,
+      "eval_loss": 0.6707106232643127,
+      "eval_mean_token_accuracy": 0.8191288441754458,
+      "eval_num_tokens": 3164054.0,
+      "eval_runtime": 88.2293,
+      "eval_samples_per_second": 16.094,
+      "eval_steps_per_second": 2.017,
+      "step": 1360
+    },
+    {
+      "entropy": 0.5454599127173424,
+      "epoch": 3.3258901629450817,
+      "grad_norm": 0.9048764705657959,
+      "learning_rate": 0.0003177531235855382,
+      "loss": 0.4763782501220703,
+      "mean_token_accuracy": 0.8505888104438781,
+      "num_tokens": 3207419.0,
+      "step": 1380
+    },
+    {
+      "epoch": 3.3258901629450817,
+      "eval_entropy": 0.5952914317002457,
+      "eval_loss": 0.662697970867157,
+      "eval_mean_token_accuracy": 0.819491108146946,
+      "eval_num_tokens": 3207419.0,
+      "eval_runtime": 87.9869,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 1380
+    },
+    {
+      "entropy": 0.5299010306596756,
+      "epoch": 3.3741701870850935,
+      "grad_norm": 0.8653950691223145,
+      "learning_rate": 0.000315440663389537,
+      "loss": 0.4581890106201172,
+      "mean_token_accuracy": 0.8536585591733455,
+      "num_tokens": 3252679.0,
+      "step": 1400
+    },
+    {
+      "epoch": 3.3741701870850935,
+      "eval_entropy": 0.5326506612675913,
+      "eval_loss": 0.6679039001464844,
+      "eval_mean_token_accuracy": 0.8209870109397374,
+      "eval_num_tokens": 3252679.0,
+      "eval_runtime": 88.3059,
+      "eval_samples_per_second": 16.08,
+      "eval_steps_per_second": 2.016,
+      "step": 1400
+    },
+    {
+      "entropy": 0.5074611511081457,
+      "epoch": 3.4224502112251054,
+      "grad_norm": 0.7786624431610107,
+      "learning_rate": 0.00031309217956467325,
+      "loss": 0.45597033500671386,
+      "mean_token_accuracy": 0.856324628740549,
+      "num_tokens": 3297309.0,
+      "step": 1420
+    },
+    {
+      "epoch": 3.4224502112251054,
+      "eval_entropy": 0.5424539073464576,
+      "eval_loss": 0.6710956692695618,
+      "eval_mean_token_accuracy": 0.8209191105338964,
+      "eval_num_tokens": 3297309.0,
+      "eval_runtime": 87.2334,
+      "eval_samples_per_second": 16.278,
+      "eval_steps_per_second": 2.041,
+      "step": 1420
+    },
+    {
+      "entropy": 0.5192026473581791,
+      "epoch": 3.4707302353651177,
+      "grad_norm": 0.7682847380638123,
+      "learning_rate": 0.0003107083367040895,
+      "loss": 0.46319098472595216,
+      "mean_token_accuracy": 0.8517456069588661,
+      "num_tokens": 3341825.0,
+      "step": 1440
+    },
+    {
+      "epoch": 3.4707302353651177,
+      "eval_entropy": 0.569345246372598,
+      "eval_loss": 0.6640089154243469,
+      "eval_mean_token_accuracy": 0.8204013628236363,
+      "eval_num_tokens": 3341825.0,
+      "eval_runtime": 87.668,
+      "eval_samples_per_second": 16.197,
+      "eval_steps_per_second": 2.03,
+      "step": 1440
+    },
+    {
+      "entropy": 0.5302285175770521,
+      "epoch": 3.51901025950513,
+      "grad_norm": 0.9500018954277039,
+      "learning_rate": 0.00030828980940711723,
+      "loss": 0.4712235927581787,
+      "mean_token_accuracy": 0.8510453663766384,
+      "num_tokens": 3388463.0,
+      "step": 1460
+    },
+    {
+      "epoch": 3.51901025950513,
+      "eval_entropy": 0.5648383499866121,
+      "eval_loss": 0.6629963517189026,
+      "eval_mean_token_accuracy": 0.8197558281126986,
+      "eval_num_tokens": 3388463.0,
+      "eval_runtime": 87.9214,
+      "eval_samples_per_second": 16.151,
+      "eval_steps_per_second": 2.025,
+      "step": 1460
+    },
+    {
+      "entropy": 0.5347354769706726,
+      "epoch": 3.567290283645142,
+      "grad_norm": 0.9935700297355652,
+      "learning_rate": 0.00030583728208837305,
+      "loss": 0.47901196479797364,
+      "mean_token_accuracy": 0.8491650126874447,
+      "num_tokens": 3433731.0,
+      "step": 1480
+    },
+    {
+      "epoch": 3.567290283645142,
+      "eval_entropy": 0.5595698517360045,
+      "eval_loss": 0.6594937443733215,
+      "eval_mean_token_accuracy": 0.8217685289597243,
+      "eval_num_tokens": 3433731.0,
+      "eval_runtime": 88.1348,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 1480
+    },
+    {
+      "entropy": 0.5250448524951935,
+      "epoch": 3.6155703077851538,
+      "grad_norm": 0.8686122894287109,
+      "learning_rate": 0.0003033514487840775,
+      "loss": 0.48284616470336916,
+      "mean_token_accuracy": 0.8505332618951797,
+      "num_tokens": 3478600.0,
+      "step": 1500
+    },
+    {
+      "epoch": 3.6155703077851538,
+      "eval_entropy": 0.5399963330017047,
+      "eval_loss": 0.6682186722755432,
+      "eval_mean_token_accuracy": 0.8201540199558387,
+      "eval_num_tokens": 3478600.0,
+      "eval_runtime": 87.9383,
+      "eval_samples_per_second": 16.148,
+      "eval_steps_per_second": 2.024,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5496356297284365,
+      "epoch": 3.663850331925166,
+      "grad_norm": 0.9891234636306763,
+      "learning_rate": 0.00030083301295565117,
+      "loss": 0.4842066287994385,
+      "mean_token_accuracy": 0.8493696190416813,
+      "num_tokens": 3521787.0,
+      "step": 1520
+    },
+    {
+      "epoch": 3.663850331925166,
+      "eval_entropy": 0.5889502502559276,
+      "eval_loss": 0.6597762703895569,
+      "eval_mean_token_accuracy": 0.819690072469497,
+      "eval_num_tokens": 3521787.0,
+      "eval_runtime": 88.0165,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1520
+    },
+    {
+      "entropy": 0.5289484921842813,
+      "epoch": 3.712130356065178,
+      "grad_norm": 0.7241885662078857,
+      "learning_rate": 0.00029828268729064326,
+      "loss": 0.48619561195373534,
+      "mean_token_accuracy": 0.8518108949065208,
+      "num_tokens": 3569036.0,
+      "step": 1540
+    },
+    {
+      "epoch": 3.712130356065178,
+      "eval_entropy": 0.5958652837892596,
+      "eval_loss": 0.6565147042274475,
+      "eval_mean_token_accuracy": 0.8217123010185328,
+      "eval_num_tokens": 3569036.0,
+      "eval_runtime": 88.0126,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1540
+    },
+    {
+      "entropy": 0.5450621031224727,
+      "epoch": 3.7604103802051903,
+      "grad_norm": 0.7989155650138855,
+      "learning_rate": 0.0002957011935010495,
+      "loss": 0.48642635345458984,
+      "mean_token_accuracy": 0.8462958924472332,
+      "num_tokens": 3614313.0,
+      "step": 1560
+    },
+    {
+      "epoch": 3.7604103802051903,
+      "eval_entropy": 0.5554354205895006,
+      "eval_loss": 0.6484317183494568,
+      "eval_mean_token_accuracy": 0.823845865016573,
+      "eval_num_tokens": 3614313.0,
+      "eval_runtime": 88.2711,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1560
+    },
+    {
+      "entropy": 0.5318546604365111,
+      "epoch": 3.808690404345202,
+      "grad_norm": 1.0375983715057373,
+      "learning_rate": 0.00029308926211907624,
+      "loss": 0.4860549449920654,
+      "mean_token_accuracy": 0.8488477177917957,
+      "num_tokens": 3659786.0,
+      "step": 1580
+    },
+    {
+      "epoch": 3.808690404345202,
+      "eval_entropy": 0.5575427207049359,
+      "eval_loss": 0.659203827381134,
+      "eval_mean_token_accuracy": 0.8210098288032446,
+      "eval_num_tokens": 3659786.0,
+      "eval_runtime": 88.3024,
+      "eval_samples_per_second": 16.081,
+      "eval_steps_per_second": 2.016,
+      "step": 1580
+    },
+    {
+      "entropy": 0.5312924966216087,
+      "epoch": 3.856970428485214,
+      "grad_norm": 0.9233641624450684,
+      "learning_rate": 0.00029044763229040845,
+      "loss": 0.48835930824279783,
+      "mean_token_accuracy": 0.8447530090808868,
+      "num_tokens": 3706362.0,
+      "step": 1600
+    },
+    {
+      "epoch": 3.856970428485214,
+      "eval_entropy": 0.5433843780434533,
+      "eval_loss": 0.6514624953269958,
+      "eval_mean_token_accuracy": 0.8230627580974879,
+      "eval_num_tokens": 3706362.0,
+      "eval_runtime": 87.9727,
+      "eval_samples_per_second": 16.141,
+      "eval_steps_per_second": 2.023,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5400687273591757,
+      "epoch": 3.9052504526252263,
+      "grad_norm": 0.7173060178756714,
+      "learning_rate": 0.00028777705156504007,
+      "loss": 0.486147403717041,
+      "mean_token_accuracy": 0.8478169530630112,
+      "num_tokens": 3753013.0,
+      "step": 1620
+    },
+    {
+      "epoch": 3.9052504526252263,
+      "eval_entropy": 0.5856067625324378,
+      "eval_loss": 0.6469284892082214,
+      "eval_mean_token_accuracy": 0.8219244604030352,
+      "eval_num_tokens": 3753013.0,
+      "eval_runtime": 88.0776,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1620
+    },
+    {
+      "entropy": 0.5115721028298139,
+      "epoch": 3.9535304767652386,
+      "grad_norm": 1.0881255865097046,
+      "learning_rate": 0.00028507827568572655,
+      "loss": 0.47065014839172364,
+      "mean_token_accuracy": 0.8535217240452766,
+      "num_tokens": 3800024.0,
+      "step": 1640
+    },
+    {
+      "epoch": 3.9535304767652386,
+      "eval_entropy": 0.565960791003838,
+      "eval_loss": 0.6459378004074097,
+      "eval_mean_token_accuracy": 0.8246003888296277,
+      "eval_num_tokens": 3800024.0,
+      "eval_runtime": 88.4974,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 1640
+    },
+    {
+      "entropy": 0.5376694152107486,
+      "epoch": 4.0,
+      "grad_norm": 3.4068243503570557,
+      "learning_rate": 0.0002823520683741183,
+      "loss": 0.4839733600616455,
+      "mean_token_accuracy": 0.8507238850965129,
+      "num_tokens": 3844304.0,
+      "step": 1660
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5631327726198047,
+      "eval_loss": 0.6423600316047668,
+      "eval_mean_token_accuracy": 0.82502972175566,
+      "eval_num_tokens": 3844304.0,
+      "eval_runtime": 88.1524,
+      "eval_samples_per_second": 16.108,
+      "eval_steps_per_second": 2.019,
+      "step": 1660
+    },
+    {
+      "entropy": 0.4188325941562653,
+      "epoch": 4.048280024140012,
+      "grad_norm": 0.6339647173881531,
+      "learning_rate": 0.00027959920111463656,
+      "loss": 0.3500279426574707,
+      "mean_token_accuracy": 0.8840543255209923,
+      "num_tokens": 3890738.0,
+      "step": 1680
+    },
+    {
+      "epoch": 4.048280024140012,
+      "eval_entropy": 0.5159014272555876,
+      "eval_loss": 0.6822672486305237,
+      "eval_mean_token_accuracy": 0.8223584498582261,
+      "eval_num_tokens": 3890738.0,
+      "eval_runtime": 87.8011,
+      "eval_samples_per_second": 16.173,
+      "eval_steps_per_second": 2.027,
+      "step": 1680
+    },
+    {
+      "entropy": 0.42600946351885793,
+      "epoch": 4.096560048280024,
+      "grad_norm": 0.6938858032226562,
+      "learning_rate": 0.0002768204529361524,
+      "loss": 0.3462975263595581,
+      "mean_token_accuracy": 0.8850771695375442,
+      "num_tokens": 3938205.0,
+      "step": 1700
+    },
+    {
+      "epoch": 4.096560048280024,
+      "eval_entropy": 0.4818309486916896,
+      "eval_loss": 0.697238028049469,
+      "eval_mean_token_accuracy": 0.822570590825563,
+      "eval_num_tokens": 3938205.0,
+      "eval_runtime": 87.6294,
+      "eval_samples_per_second": 16.205,
+      "eval_steps_per_second": 2.031,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4128956265747547,
+      "epoch": 4.1448400724200365,
+      "grad_norm": 0.7023837566375732,
+      "learning_rate": 0.000274016610191531,
+      "loss": 0.35343332290649415,
+      "mean_token_accuracy": 0.8837565891444683,
+      "num_tokens": 3986350.0,
+      "step": 1720
+    },
+    {
+      "epoch": 4.1448400724200365,
+      "eval_entropy": 0.4736787666765492,
+      "eval_loss": 0.686392605304718,
+      "eval_mean_token_accuracy": 0.8246323891570059,
+      "eval_num_tokens": 3986350.0,
+      "eval_runtime": 87.755,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 1720
+    },
+    {
+      "entropy": 0.4070820797234774,
+      "epoch": 4.193120096560048,
+      "grad_norm": 0.7664099335670471,
+      "learning_rate": 0.00027118846633510276,
+      "loss": 0.35689282417297363,
+      "mean_token_accuracy": 0.8821764968335628,
+      "num_tokens": 4033541.0,
+      "step": 1740
+    },
+    {
+      "epoch": 4.193120096560048,
+      "eval_entropy": 0.48123268063148755,
+      "eval_loss": 0.6800221800804138,
+      "eval_mean_token_accuracy": 0.8234515280536051,
+      "eval_num_tokens": 4033541.0,
+      "eval_runtime": 87.613,
+      "eval_samples_per_second": 16.208,
+      "eval_steps_per_second": 2.032,
+      "step": 1740
+    },
+    {
+      "entropy": 0.4199690703302622,
+      "epoch": 4.24140012070006,
+      "grad_norm": 0.7411720752716064,
+      "learning_rate": 0.00026833682169812544,
+      "loss": 0.35881006717681885,
+      "mean_token_accuracy": 0.8812312394380569,
+      "num_tokens": 4079253.0,
+      "step": 1760
+    },
+    {
+      "epoch": 4.24140012070006,
+      "eval_entropy": 0.47287314242861245,
+      "eval_loss": 0.691737949848175,
+      "eval_mean_token_accuracy": 0.8229516232281588,
+      "eval_num_tokens": 4079253.0,
+      "eval_runtime": 87.9596,
+      "eval_samples_per_second": 16.144,
+      "eval_steps_per_second": 2.024,
+      "step": 1760
+    },
+    {
+      "entropy": 0.4195733394473791,
+      "epoch": 4.289680144840072,
+      "grad_norm": 0.9975268244743347,
+      "learning_rate": 0.00026546248326229954,
+      "loss": 0.36998801231384276,
+      "mean_token_accuracy": 0.8777976937592029,
+      "num_tokens": 4125495.0,
+      "step": 1780
+    },
+    {
+      "epoch": 4.289680144840072,
+      "eval_entropy": 0.48877454825331657,
+      "eval_loss": 0.6843481063842773,
+      "eval_mean_token_accuracy": 0.8226028995567494,
+      "eval_num_tokens": 4125495.0,
+      "eval_runtime": 88.169,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 1780
+    },
+    {
+      "entropy": 0.4087799321860075,
+      "epoch": 4.337960168980085,
+      "grad_norm": 0.8101420402526855,
+      "learning_rate": 0.00026256626443140217,
+      "loss": 0.3555623531341553,
+      "mean_token_accuracy": 0.8815289497375488,
+      "num_tokens": 4173422.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.337960168980085,
+      "eval_entropy": 0.4845838766084628,
+      "eval_loss": 0.6782916188240051,
+      "eval_mean_token_accuracy": 0.8235659428526846,
+      "eval_num_tokens": 4173422.0,
+      "eval_runtime": 88.0763,
+      "eval_samples_per_second": 16.122,
+      "eval_steps_per_second": 2.021,
+      "step": 1800
+    },
+    {
+      "entropy": 0.4086425334215164,
+      "epoch": 4.386240193120097,
+      "grad_norm": 0.8936498761177063,
+      "learning_rate": 0.0002596489848011036,
+      "loss": 0.3710860013961792,
+      "mean_token_accuracy": 0.8773891530930996,
+      "num_tokens": 4216243.0,
+      "step": 1820
+    },
+    {
+      "epoch": 4.386240193120097,
+      "eval_entropy": 0.49018637685293565,
+      "eval_loss": 0.6762362718582153,
+      "eval_mean_token_accuracy": 0.8242661363623115,
+      "eval_num_tokens": 4216243.0,
+      "eval_runtime": 88.3684,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 1820
+    },
+    {
+      "entropy": 0.41956189945340155,
+      "epoch": 4.434520217260109,
+      "grad_norm": 0.6450079083442688,
+      "learning_rate": 0.00025671146992703106,
+      "loss": 0.3715682029724121,
+      "mean_token_accuracy": 0.878043319284916,
+      "num_tokens": 4262463.0,
+      "step": 1840
+    },
+    {
+      "epoch": 4.434520217260109,
+      "eval_entropy": 0.4821142643020394,
+      "eval_loss": 0.6749269962310791,
+      "eval_mean_token_accuracy": 0.8252766048640348,
+      "eval_num_tokens": 4262463.0,
+      "eval_runtime": 87.6946,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 1840
+    },
+    {
+      "entropy": 0.40638086050748823,
+      "epoch": 4.4828002414001205,
+      "grad_norm": 0.8167530298233032,
+      "learning_rate": 0.0002537545510911465,
+      "loss": 0.3652146100997925,
+      "mean_token_accuracy": 0.8761600315570831,
+      "num_tokens": 4311451.0,
+      "step": 1860
+    },
+    {
+      "epoch": 4.4828002414001205,
+      "eval_entropy": 0.5017492558514134,
+      "eval_loss": 0.6696512699127197,
+      "eval_mean_token_accuracy": 0.8252955695216575,
+      "eval_num_tokens": 4311451.0,
+      "eval_runtime": 88.02,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 1860
+    },
+    {
+      "entropy": 0.41664722077548505,
+      "epoch": 4.531080265540133,
+      "grad_norm": 0.757738471031189,
+      "learning_rate": 0.0002507790650665036,
+      "loss": 0.368887734413147,
+      "mean_token_accuracy": 0.8786893397569656,
+      "num_tokens": 4357724.0,
+      "step": 1880
+    },
+    {
+      "epoch": 4.531080265540133,
+      "eval_entropy": 0.4798511176296834,
+      "eval_loss": 0.6738249659538269,
+      "eval_mean_token_accuracy": 0.8250340461061242,
+      "eval_num_tokens": 4357724.0,
+      "eval_runtime": 88.2713,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 1880
+    },
+    {
+      "entropy": 0.42732664197683334,
+      "epoch": 4.579360289680145,
+      "grad_norm": 0.7890474200248718,
+      "learning_rate": 0.00024778585388045044,
+      "loss": 0.38672595024108886,
+      "mean_token_accuracy": 0.8742976360023021,
+      "num_tokens": 4401934.0,
+      "step": 1900
+    },
+    {
+      "epoch": 4.579360289680145,
+      "eval_entropy": 0.49321879295820603,
+      "eval_loss": 0.6752269268035889,
+      "eval_mean_token_accuracy": 0.824115453811174,
+      "eval_num_tokens": 4401934.0,
+      "eval_runtime": 87.3378,
+      "eval_samples_per_second": 16.259,
+      "eval_steps_per_second": 2.038,
+      "step": 1900
+    },
+    {
+      "entropy": 0.41300065964460375,
+      "epoch": 4.627640313820157,
+      "grad_norm": 0.866765022277832,
+      "learning_rate": 0.00024477576457634665,
+      "loss": 0.364247989654541,
+      "mean_token_accuracy": 0.8774793080985546,
+      "num_tokens": 4449745.0,
+      "step": 1920
+    },
+    {
+      "epoch": 4.627640313820157,
+      "eval_entropy": 0.49179744352115673,
+      "eval_loss": 0.6731056571006775,
+      "eval_mean_token_accuracy": 0.8245729244826885,
+      "eval_num_tokens": 4449745.0,
+      "eval_runtime": 88.2458,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 1920
+    },
+    {
+      "entropy": 0.42646181732416155,
+      "epoch": 4.675920337960169,
+      "grad_norm": 0.7921772003173828,
+      "learning_rate": 0.00024174964897385928,
+      "loss": 0.3822937488555908,
+      "mean_token_accuracy": 0.8755100265145301,
+      "num_tokens": 4495074.0,
+      "step": 1940
+    },
+    {
+      "epoch": 4.675920337960169,
+      "eval_entropy": 0.5194089940089858,
+      "eval_loss": 0.6604889631271362,
+      "eval_mean_token_accuracy": 0.8260471847619903,
+      "eval_num_tokens": 4495074.0,
+      "eval_runtime": 88.0139,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 1940
+    },
+    {
+      "entropy": 0.4235709119588137,
+      "epoch": 4.724200362100181,
+      "grad_norm": 0.8816011548042297,
+      "learning_rate": 0.00023870836342790837,
+      "loss": 0.3772272109985352,
+      "mean_token_accuracy": 0.8731484808027744,
+      "num_tokens": 4541053.0,
+      "step": 1960
+    },
+    {
+      "epoch": 4.724200362100181,
+      "eval_entropy": 0.47387497545628066,
+      "eval_loss": 0.6644074320793152,
+      "eval_mean_token_accuracy": 0.8271099258674665,
+      "eval_num_tokens": 4541053.0,
+      "eval_runtime": 88.393,
+      "eval_samples_per_second": 16.065,
+      "eval_steps_per_second": 2.014,
+      "step": 1960
+    },
+    {
+      "entropy": 0.42933564484119413,
+      "epoch": 4.772480386240193,
+      "grad_norm": 0.9746852517127991,
+      "learning_rate": 0.0002356527685863284,
+      "loss": 0.3843918561935425,
+      "mean_token_accuracy": 0.873461090028286,
+      "num_tokens": 4584210.0,
+      "step": 1980
+    },
+    {
+      "epoch": 4.772480386240193,
+      "eval_entropy": 0.47977291734031074,
+      "eval_loss": 0.6671409010887146,
+      "eval_mean_token_accuracy": 0.8273240890395775,
+      "eval_num_tokens": 4584210.0,
+      "eval_runtime": 88.6022,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 1980
+    },
+    {
+      "entropy": 0.44694459773600104,
+      "epoch": 4.820760410380205,
+      "grad_norm": 0.7372738122940063,
+      "learning_rate": 0.00023258372914631542,
+      "loss": 0.38873090744018557,
+      "mean_token_accuracy": 0.871940117329359,
+      "num_tokens": 4630254.0,
+      "step": 2000
+    },
+    {
+      "epoch": 4.820760410380205,
+      "eval_entropy": 0.49700497023844986,
+      "eval_loss": 0.6683036088943481,
+      "eval_mean_token_accuracy": 0.8265726067376941,
+      "eval_num_tokens": 4630254.0,
+      "eval_runtime": 88.1382,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 2000
+    },
+    {
+      "entropy": 0.42922686524689196,
+      "epoch": 4.869040434520217,
+      "grad_norm": 0.6791961789131165,
+      "learning_rate": 0.0002295021136097277,
+      "loss": 0.3735378265380859,
+      "mean_token_accuracy": 0.8772910334169864,
+      "num_tokens": 4679265.0,
+      "step": 2020
+    },
+    {
+      "epoch": 4.869040434520217,
+      "eval_entropy": 0.49750003386079594,
+      "eval_loss": 0.66725093126297,
+      "eval_mean_token_accuracy": 0.8255943565556173,
+      "eval_num_tokens": 4679265.0,
+      "eval_runtime": 86.9873,
+      "eval_samples_per_second": 16.324,
+      "eval_steps_per_second": 2.046,
+      "step": 2020
+    },
+    {
+      "entropy": 0.4518359176814556,
+      "epoch": 4.917320458660229,
+      "grad_norm": 0.7970417141914368,
+      "learning_rate": 0.00022640879403731002,
+      "loss": 0.38829789161682127,
+      "mean_token_accuracy": 0.8700004398822785,
+      "num_tokens": 4723592.0,
+      "step": 2040
+    },
+    {
+      "epoch": 4.917320458660229,
+      "eval_entropy": 0.4902817869789145,
+      "eval_loss": 0.6645081639289856,
+      "eval_mean_token_accuracy": 0.8280317391572374,
+      "eval_num_tokens": 4723592.0,
+      "eval_runtime": 87.4334,
+      "eval_samples_per_second": 16.241,
+      "eval_steps_per_second": 2.036,
+      "step": 2040
+    },
+    {
+      "entropy": 0.4405197504907846,
+      "epoch": 4.965600482800241,
+      "grad_norm": 1.0449241399765015,
+      "learning_rate": 0.00022330464580191104,
+      "loss": 0.3832916975021362,
+      "mean_token_accuracy": 0.8744640938937664,
+      "num_tokens": 4767253.0,
+      "step": 2060
+    },
+    {
+      "epoch": 4.965600482800241,
+      "eval_entropy": 0.4897575577658214,
+      "eval_loss": 0.6547137498855591,
+      "eval_mean_token_accuracy": 0.8286678559994429,
+      "eval_num_tokens": 4767253.0,
+      "eval_runtime": 88.2716,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 2060
+    },
+    {
+      "entropy": 0.38312033089724457,
+      "epoch": 5.012070006035003,
+      "grad_norm": 0.6516358256340027,
+      "learning_rate": 0.0002201905473407628,
+      "loss": 0.3343928098678589,
+      "mean_token_accuracy": 0.8923151601444591,
+      "num_tokens": 4816587.0,
+      "step": 2080
+    },
+    {
+      "epoch": 5.012070006035003,
+      "eval_entropy": 0.4274808027771082,
+      "eval_loss": 0.6996631026268005,
+      "eval_mean_token_accuracy": 0.8259488166718001,
+      "eval_num_tokens": 4816587.0,
+      "eval_runtime": 88.3757,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 2080
+    },
+    {
+      "entropy": 0.3049557346850634,
+      "epoch": 5.060350030175015,
+      "grad_norm": 0.6272661685943604,
+      "learning_rate": 0.00021706737990689358,
+      "loss": 0.2521080255508423,
+      "mean_token_accuracy": 0.912626949697733,
+      "num_tokens": 4861071.0,
+      "step": 2100
+    },
+    {
+      "epoch": 5.060350030175015,
+      "eval_entropy": 0.41769456176945335,
+      "eval_loss": 0.7133082151412964,
+      "eval_mean_token_accuracy": 0.8254162324278542,
+      "eval_num_tokens": 4861071.0,
+      "eval_runtime": 87.9672,
+      "eval_samples_per_second": 16.142,
+      "eval_steps_per_second": 2.023,
+      "step": 2100
+    },
+    {
+      "entropy": 0.31178536619991065,
+      "epoch": 5.108630054315027,
+      "grad_norm": 0.9577748775482178,
+      "learning_rate": 0.0002139360273197436,
+      "loss": 0.24547300338745118,
+      "mean_token_accuracy": 0.9145238555967807,
+      "num_tokens": 4904742.0,
+      "step": 2120
+    },
+    {
+      "epoch": 5.108630054315027,
+      "eval_entropy": 0.42631214170643456,
+      "eval_loss": 0.7339685559272766,
+      "eval_mean_token_accuracy": 0.8219655203015617,
+      "eval_num_tokens": 4904742.0,
+      "eval_runtime": 87.9257,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 2120
+    },
+    {
+      "entropy": 0.3008375624194741,
+      "epoch": 5.15691007845504,
+      "grad_norm": 0.797623336315155,
+      "learning_rate": 0.00021079737571505472,
+      "loss": 0.24592554569244385,
+      "mean_token_accuracy": 0.9159514293074608,
+      "num_tokens": 4950795.0,
+      "step": 2140
+    },
+    {
+      "epoch": 5.15691007845504,
+      "eval_entropy": 0.4180516125111098,
+      "eval_loss": 0.7242252826690674,
+      "eval_mean_token_accuracy": 0.8247456322895007,
+      "eval_num_tokens": 4950795.0,
+      "eval_runtime": 87.832,
+      "eval_samples_per_second": 16.167,
+      "eval_steps_per_second": 2.027,
+      "step": 2140
+    },
+    {
+      "entropy": 0.32739414311945436,
+      "epoch": 5.2051901025950515,
+      "grad_norm": 0.6331654191017151,
+      "learning_rate": 0.00020765231329410443,
+      "loss": 0.2587909460067749,
+      "mean_token_accuracy": 0.9127019934356213,
+      "num_tokens": 4994886.0,
+      "step": 2160
+    },
+    {
+      "epoch": 5.2051901025950515,
+      "eval_entropy": 0.43612574191575637,
+      "eval_loss": 0.7163131833076477,
+      "eval_mean_token_accuracy": 0.8249044545580831,
+      "eval_num_tokens": 4994886.0,
+      "eval_runtime": 88.184,
+      "eval_samples_per_second": 16.103,
+      "eval_steps_per_second": 2.019,
+      "step": 2160
+    },
+    {
+      "entropy": 0.31282360590994357,
+      "epoch": 5.253470126735063,
+      "grad_norm": 0.8501704931259155,
+      "learning_rate": 0.00020450173007235538,
+      "loss": 0.2543140649795532,
+      "mean_token_accuracy": 0.9109282083809376,
+      "num_tokens": 5043663.0,
+      "step": 2180
+    },
+    {
+      "epoch": 5.253470126735063,
+      "eval_entropy": 0.42737145202883176,
+      "eval_loss": 0.715088963508606,
+      "eval_mean_token_accuracy": 0.8260066037097674,
+      "eval_num_tokens": 5043663.0,
+      "eval_runtime": 88.1598,
+      "eval_samples_per_second": 16.107,
+      "eval_steps_per_second": 2.019,
+      "step": 2180
+    },
+    {
+      "entropy": 0.31767325326800344,
+      "epoch": 5.301750150875075,
+      "grad_norm": 0.808242917060852,
+      "learning_rate": 0.0002013465176275914,
+      "loss": 0.25598506927490233,
+      "mean_token_accuracy": 0.9112607099115848,
+      "num_tokens": 5087959.0,
+      "step": 2200
+    },
+    {
+      "epoch": 5.301750150875075,
+      "eval_entropy": 0.420320589723212,
+      "eval_loss": 0.7172654867172241,
+      "eval_mean_token_accuracy": 0.8260447238938192,
+      "eval_num_tokens": 5087959.0,
+      "eval_runtime": 88.1646,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 2200
+    },
+    {
+      "entropy": 0.30062707886099815,
+      "epoch": 5.350030175015087,
+      "grad_norm": 0.7561016082763672,
+      "learning_rate": 0.0001981875688476119,
+      "loss": 0.24859883785247802,
+      "mean_token_accuracy": 0.9129510529339313,
+      "num_tokens": 5136121.0,
+      "step": 2220
+    },
+    {
+      "epoch": 5.350030175015087,
+      "eval_entropy": 0.4228216555346264,
+      "eval_loss": 0.722082257270813,
+      "eval_mean_token_accuracy": 0.8257333281334867,
+      "eval_num_tokens": 5136121.0,
+      "eval_runtime": 88.2664,
+      "eval_samples_per_second": 16.088,
+      "eval_steps_per_second": 2.017,
+      "step": 2220
+    },
+    {
+      "entropy": 0.3080605179071426,
+      "epoch": 5.3983101991551,
+      "grad_norm": 0.8511217832565308,
+      "learning_rate": 0.00019502577767755477,
+      "loss": 0.2575610876083374,
+      "mean_token_accuracy": 0.9120334684848785,
+      "num_tokens": 5182144.0,
+      "step": 2240
+    },
+    {
+      "epoch": 5.3983101991551,
+      "eval_entropy": 0.407860318931301,
+      "eval_loss": 0.7172144651412964,
+      "eval_mean_token_accuracy": 0.8263489242350117,
+      "eval_num_tokens": 5182144.0,
+      "eval_runtime": 88.1308,
+      "eval_samples_per_second": 16.112,
+      "eval_steps_per_second": 2.02,
+      "step": 2240
+    },
+    {
+      "entropy": 0.30764442328363656,
+      "epoch": 5.446590223295112,
+      "grad_norm": 0.928459107875824,
+      "learning_rate": 0.0001918620388669208,
+      "loss": 0.2558996915817261,
+      "mean_token_accuracy": 0.9127804256975651,
+      "num_tokens": 5226845.0,
+      "step": 2260
+    },
+    {
+      "epoch": 5.446590223295112,
+      "eval_entropy": 0.40719465191444654,
+      "eval_loss": 0.7216960191726685,
+      "eval_mean_token_accuracy": 0.8262304382377796,
+      "eval_num_tokens": 5226845.0,
+      "eval_runtime": 88.4134,
+      "eval_samples_per_second": 16.061,
+      "eval_steps_per_second": 2.013,
+      "step": 2260
+    },
+    {
+      "entropy": 0.30657214783132075,
+      "epoch": 5.494870247435124,
+      "grad_norm": 0.90412437915802,
+      "learning_rate": 0.00018869724771637,
+      "loss": 0.259705924987793,
+      "mean_token_accuracy": 0.9097815379500389,
+      "num_tokens": 5272403.0,
+      "step": 2280
+    },
+    {
+      "epoch": 5.494870247435124,
+      "eval_entropy": 0.39945232700765804,
+      "eval_loss": 0.7241353392601013,
+      "eval_mean_token_accuracy": 0.8252729776869999,
+      "eval_num_tokens": 5272403.0,
+      "eval_runtime": 88.2952,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 2280
+    },
+    {
+      "entropy": 0.2879099613055587,
+      "epoch": 5.5431502715751355,
+      "grad_norm": 0.8688379526138306,
+      "learning_rate": 0.00018553229982436206,
+      "loss": 0.24692924022674562,
+      "mean_token_accuracy": 0.913682048022747,
+      "num_tokens": 5320882.0,
+      "step": 2300
+    },
+    {
+      "epoch": 5.5431502715751355,
+      "eval_entropy": 0.39947180751334416,
+      "eval_loss": 0.7101960182189941,
+      "eval_mean_token_accuracy": 0.82783506827408,
+      "eval_num_tokens": 5320882.0,
+      "eval_runtime": 88.4907,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2300
+    },
+    {
+      "entropy": 0.3070108935236931,
+      "epoch": 5.591430295715147,
+      "grad_norm": 0.7736193537712097,
+      "learning_rate": 0.00018236809083371273,
+      "loss": 0.2597432851791382,
+      "mean_token_accuracy": 0.90810831412673,
+      "num_tokens": 5370943.0,
+      "step": 2320
+    },
+    {
+      "epoch": 5.591430295715147,
+      "eval_entropy": 0.4090571740035261,
+      "eval_loss": 0.7136856913566589,
+      "eval_mean_token_accuracy": 0.8271592694721864,
+      "eval_num_tokens": 5370943.0,
+      "eval_runtime": 88.0111,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 2320
+    },
+    {
+      "entropy": 0.2969514708966017,
+      "epoch": 5.63971031985516,
+      "grad_norm": 0.8571580052375793,
+      "learning_rate": 0.00017920551617813733,
+      "loss": 0.25244779586791993,
+      "mean_token_accuracy": 0.9115345612168312,
+      "num_tokens": 5420284.0,
+      "step": 2340
+    },
+    {
+      "epoch": 5.63971031985516,
+      "eval_entropy": 0.4035519739885009,
+      "eval_loss": 0.7175078988075256,
+      "eval_mean_token_accuracy": 0.828284356031525,
+      "eval_num_tokens": 5420284.0,
+      "eval_runtime": 87.9478,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2340
+    },
+    {
+      "entropy": 0.2961398035287857,
+      "epoch": 5.687990343995172,
+      "grad_norm": 0.933102011680603,
+      "learning_rate": 0.0001760454708288538,
+      "loss": 0.2576280117034912,
+      "mean_token_accuracy": 0.9119551591575146,
+      "num_tokens": 5469507.0,
+      "step": 2360
+    },
+    {
+      "epoch": 5.687990343995172,
+      "eval_entropy": 0.41212199142809663,
+      "eval_loss": 0.7149607539176941,
+      "eval_mean_token_accuracy": 0.8266791905579942,
+      "eval_num_tokens": 5469507.0,
+      "eval_runtime": 87.7632,
+      "eval_samples_per_second": 16.18,
+      "eval_steps_per_second": 2.028,
+      "step": 2360
+    },
+    {
+      "entropy": 0.3072124743834138,
+      "epoch": 5.736270368135184,
+      "grad_norm": 0.7064250707626343,
+      "learning_rate": 0.00017288884904131615,
+      "loss": 0.258055591583252,
+      "mean_token_accuracy": 0.9088842839002609,
+      "num_tokens": 5514274.0,
+      "step": 2380
+    },
+    {
+      "epoch": 5.736270368135184,
+      "eval_entropy": 0.39512370325876084,
+      "eval_loss": 0.7194926738739014,
+      "eval_mean_token_accuracy": 0.8285625402847033,
+      "eval_num_tokens": 5514274.0,
+      "eval_runtime": 87.9474,
+      "eval_samples_per_second": 16.146,
+      "eval_steps_per_second": 2.024,
+      "step": 2380
+    },
+    {
+      "entropy": 0.31159722357988356,
+      "epoch": 5.784550392275197,
+      "grad_norm": 0.6565853953361511,
+      "learning_rate": 0.00016973654410215114,
+      "loss": 0.2594040632247925,
+      "mean_token_accuracy": 0.9096784308552742,
+      "num_tokens": 5561349.0,
+      "step": 2400
+    },
+    {
+      "epoch": 5.784550392275197,
+      "eval_entropy": 0.4137382763490248,
+      "eval_loss": 0.7111276984214783,
+      "eval_mean_token_accuracy": 0.8272288163726249,
+      "eval_num_tokens": 5561349.0,
+      "eval_runtime": 88.0204,
+      "eval_samples_per_second": 16.133,
+      "eval_steps_per_second": 2.022,
+      "step": 2400
+    },
+    {
+      "entropy": 0.3013706898316741,
+      "epoch": 5.8328304164152085,
+      "grad_norm": 0.7962143421173096,
+      "learning_rate": 0.00016658944807636825,
+      "loss": 0.25681519508361816,
+      "mean_token_accuracy": 0.9116937138140202,
+      "num_tokens": 5608215.0,
+      "step": 2420
+    },
+    {
+      "epoch": 5.8328304164152085,
+      "eval_entropy": 0.4103235746367594,
+      "eval_loss": 0.7075955271720886,
+      "eval_mean_token_accuracy": 0.8287353076961603,
+      "eval_num_tokens": 5608215.0,
+      "eval_runtime": 88.4363,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 2420
+    },
+    {
+      "entropy": 0.30746946930885316,
+      "epoch": 5.88111044055522,
+      "grad_norm": 0.8445194363594055,
+      "learning_rate": 0.0001634484515549164,
+      "loss": 0.25779383182525634,
+      "mean_token_accuracy": 0.9099907353520393,
+      "num_tokens": 5653216.0,
+      "step": 2440
+    },
+    {
+      "epoch": 5.88111044055522,
+      "eval_entropy": 0.39043665166651265,
+      "eval_loss": 0.7137237787246704,
+      "eval_mean_token_accuracy": 0.8278630744205432,
+      "eval_num_tokens": 5653216.0,
+      "eval_runtime": 88.2017,
+      "eval_samples_per_second": 16.099,
+      "eval_steps_per_second": 2.018,
+      "step": 2440
+    },
+    {
+      "entropy": 0.3028501259163022,
+      "epoch": 5.929390464695232,
+      "grad_norm": 0.7994105815887451,
+      "learning_rate": 0.0001603144434026566,
+      "loss": 0.2614556074142456,
+      "mean_token_accuracy": 0.9096644438803196,
+      "num_tokens": 5698417.0,
+      "step": 2460
+    },
+    {
+      "epoch": 5.929390464695232,
+      "eval_entropy": 0.41160977790864667,
+      "eval_loss": 0.7089618444442749,
+      "eval_mean_token_accuracy": 0.8270713757263141,
+      "eval_num_tokens": 5698417.0,
+      "eval_runtime": 88.1176,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2460
+    },
+    {
+      "entropy": 0.3106454351916909,
+      "epoch": 5.977670488835244,
+      "grad_norm": 0.7986742854118347,
+      "learning_rate": 0.0001571883105068242,
+      "loss": 0.26073107719421384,
+      "mean_token_accuracy": 0.9098567090928554,
+      "num_tokens": 5744944.0,
+      "step": 2480
+    },
+    {
+      "epoch": 5.977670488835244,
+      "eval_entropy": 0.39703783945421156,
+      "eval_loss": 0.7098209857940674,
+      "eval_mean_token_accuracy": 0.8296985512369135,
+      "eval_num_tokens": 5744944.0,
+      "eval_runtime": 88.0813,
+      "eval_samples_per_second": 16.121,
+      "eval_steps_per_second": 2.021,
+      "step": 2480
+    },
+    {
+      "entropy": 0.2523191448349457,
+      "epoch": 6.024140012070006,
+      "grad_norm": 0.7271217107772827,
+      "learning_rate": 0.00015407093752605022,
+      "loss": 0.20083463191986084,
+      "mean_token_accuracy": 0.9312780762647653,
+      "num_tokens": 5792908.0,
+      "step": 2500
+    },
+    {
+      "epoch": 6.024140012070006,
+      "eval_entropy": 0.3289327997337566,
+      "eval_loss": 0.8134235143661499,
+      "eval_mean_token_accuracy": 0.8248051414998729,
+      "eval_num_tokens": 5792908.0,
+      "eval_runtime": 89.1335,
+      "eval_samples_per_second": 15.931,
+      "eval_steps_per_second": 1.997,
+      "step": 2500
+    },
+    {
+      "entropy": 0.17906902078539133,
+      "epoch": 6.072420036210018,
+      "grad_norm": 0.6044061183929443,
+      "learning_rate": 0.00015096320664001312,
+      "loss": 0.14825894832611083,
+      "mean_token_accuracy": 0.9478232830762863,
+      "num_tokens": 5836969.0,
+      "step": 2520
+    },
+    {
+      "epoch": 6.072420036210018,
+      "eval_entropy": 0.34401263899347756,
+      "eval_loss": 0.7951938509941101,
+      "eval_mean_token_accuracy": 0.8232996959364816,
+      "eval_num_tokens": 5836969.0,
+      "eval_runtime": 88.4607,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 2520
+    },
+    {
+      "entropy": 0.1896738692186773,
+      "epoch": 6.12070006035003,
+      "grad_norm": 0.7623595595359802,
+      "learning_rate": 0.0001478659972997929,
+      "loss": 0.15062018632888793,
+      "mean_token_accuracy": 0.9476156957447529,
+      "num_tokens": 5880440.0,
+      "step": 2540
+    },
+    {
+      "epoch": 6.12070006035003,
+      "eval_entropy": 0.3361820410309213,
+      "eval_loss": 0.7982146739959717,
+      "eval_mean_token_accuracy": 0.825476693638255,
+      "eval_num_tokens": 5880440.0,
+      "eval_runtime": 89.0158,
+      "eval_samples_per_second": 15.952,
+      "eval_steps_per_second": 2.0,
+      "step": 2540
+    },
+    {
+      "entropy": 0.19159424304962158,
+      "epoch": 6.168980084490042,
+      "grad_norm": 0.7059116959571838,
+      "learning_rate": 0.00014478018597899632,
+      "loss": 0.15041557550430298,
+      "mean_token_accuracy": 0.946315186470747,
+      "num_tokens": 5924665.0,
+      "step": 2560
+    },
+    {
+      "epoch": 6.168980084490042,
+      "eval_entropy": 0.328647531987576,
+      "eval_loss": 0.8098915815353394,
+      "eval_mean_token_accuracy": 0.823520571328281,
+      "eval_num_tokens": 5924665.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 2560
+    },
+    {
+      "entropy": 0.1834640829823911,
+      "epoch": 6.217260108630055,
+      "grad_norm": 0.609506368637085,
+      "learning_rate": 0.00014170664592572544,
+      "loss": 0.1507703423500061,
+      "mean_token_accuracy": 0.947880481928587,
+      "num_tokens": 5971116.0,
+      "step": 2580
+    },
+    {
+      "epoch": 6.217260108630055,
+      "eval_entropy": 0.3346639998843161,
+      "eval_loss": 0.7952263355255127,
+      "eval_mean_token_accuracy": 0.8255967082602255,
+      "eval_num_tokens": 5971116.0,
+      "eval_runtime": 88.588,
+      "eval_samples_per_second": 16.029,
+      "eval_steps_per_second": 2.009,
+      "step": 2580
+    },
+    {
+      "entropy": 0.18807780891656875,
+      "epoch": 6.265540132770067,
+      "grad_norm": 0.782401442527771,
+      "learning_rate": 0.00013864624691545897,
+      "loss": 0.150755774974823,
+      "mean_token_accuracy": 0.9453216694295407,
+      "num_tokens": 6018684.0,
+      "step": 2600
+    },
+    {
+      "epoch": 6.265540132770067,
+      "eval_entropy": 0.321210315137097,
+      "eval_loss": 0.7982205152511597,
+      "eval_mean_token_accuracy": 0.826972935976607,
+      "eval_num_tokens": 6018684.0,
+      "eval_runtime": 88.0326,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 2600
+    },
+    {
+      "entropy": 0.18887900887057185,
+      "epoch": 6.3138201569100785,
+      "grad_norm": 0.7818743586540222,
+      "learning_rate": 0.00013559985500491572,
+      "loss": 0.1514918327331543,
+      "mean_token_accuracy": 0.9457836978137493,
+      "num_tokens": 6066954.0,
+      "step": 2620
+    },
+    {
+      "epoch": 6.3138201569100785,
+      "eval_entropy": 0.33611535657657665,
+      "eval_loss": 0.8096784353256226,
+      "eval_mean_token_accuracy": 0.8244635074326162,
+      "eval_num_tokens": 6066954.0,
+      "eval_runtime": 87.9857,
+      "eval_samples_per_second": 16.139,
+      "eval_steps_per_second": 2.023,
+      "step": 2620
+    },
+    {
+      "entropy": 0.17428901745006442,
+      "epoch": 6.36210018105009,
+      "grad_norm": 0.7160996198654175,
+      "learning_rate": 0.00013256833228697111,
+      "loss": 0.14635720252990722,
+      "mean_token_accuracy": 0.9494572594761849,
+      "num_tokens": 6117479.0,
+      "step": 2640
+    },
+    {
+      "epoch": 6.36210018105009,
+      "eval_entropy": 0.32968130741226537,
+      "eval_loss": 0.795514702796936,
+      "eval_mean_token_accuracy": 0.8256657645943459,
+      "eval_num_tokens": 6117479.0,
+      "eval_runtime": 88.9931,
+      "eval_samples_per_second": 15.956,
+      "eval_steps_per_second": 2.0,
+      "step": 2640
+    },
+    {
+      "entropy": 0.18460844503715634,
+      "epoch": 6.410380205190102,
+      "grad_norm": 0.8100566267967224,
+      "learning_rate": 0.0001295525366466948,
+      "loss": 0.1496778130531311,
+      "mean_token_accuracy": 0.9469681069254875,
+      "num_tokens": 6163103.0,
+      "step": 2660
+    },
+    {
+      "epoch": 6.410380205190102,
+      "eval_entropy": 0.32415712340159364,
+      "eval_loss": 0.8005954623222351,
+      "eval_mean_token_accuracy": 0.8246443308471294,
+      "eval_num_tokens": 6163103.0,
+      "eval_runtime": 88.7372,
+      "eval_samples_per_second": 16.002,
+      "eval_steps_per_second": 2.006,
+      "step": 2660
+    },
+    {
+      "entropy": 0.19344648774713277,
+      "epoch": 6.458660229330115,
+      "grad_norm": 0.6743079423904419,
+      "learning_rate": 0.0001265533215185796,
+      "loss": 0.1577553391456604,
+      "mean_token_accuracy": 0.944937615096569,
+      "num_tokens": 6210504.0,
+      "step": 2680
+    },
+    {
+      "epoch": 6.458660229330115,
+      "eval_entropy": 0.33167651705862433,
+      "eval_loss": 0.7990406155586243,
+      "eval_mean_token_accuracy": 0.824895450573289,
+      "eval_num_tokens": 6210504.0,
+      "eval_runtime": 88.7596,
+      "eval_samples_per_second": 15.998,
+      "eval_steps_per_second": 2.005,
+      "step": 2680
+    },
+    {
+      "entropy": 0.18892329484224318,
+      "epoch": 6.506940253470127,
+      "grad_norm": 0.7439920902252197,
+      "learning_rate": 0.00012357153564502942,
+      "loss": 0.15402607917785643,
+      "mean_token_accuracy": 0.9461932055652141,
+      "num_tokens": 6256168.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.506940253470127,
+      "eval_entropy": 0.3296281209822451,
+      "eval_loss": 0.7959091067314148,
+      "eval_mean_token_accuracy": 0.8255601418152284,
+      "eval_num_tokens": 6256168.0,
+      "eval_runtime": 88.4555,
+      "eval_samples_per_second": 16.053,
+      "eval_steps_per_second": 2.012,
+      "step": 2700
+    },
+    {
+      "entropy": 0.1912553665228188,
+      "epoch": 6.555220277610139,
+      "grad_norm": 0.6863725185394287,
+      "learning_rate": 0.00012060802283617536,
+      "loss": 0.15676106214523317,
+      "mean_token_accuracy": 0.9435188084840774,
+      "num_tokens": 6299642.0,
+      "step": 2720
+    },
+    {
+      "epoch": 6.555220277610139,
+      "eval_entropy": 0.33108900378594236,
+      "eval_loss": 0.7999242544174194,
+      "eval_mean_token_accuracy": 0.8262744124016065,
+      "eval_num_tokens": 6299642.0,
+      "eval_runtime": 87.2499,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 2720
+    },
+    {
+      "entropy": 0.19352400647476314,
+      "epoch": 6.603500301750151,
+      "grad_norm": 0.6504069566726685,
+      "learning_rate": 0.00011766362173108738,
+      "loss": 0.15693014860153198,
+      "mean_token_accuracy": 0.9453470788896083,
+      "num_tokens": 6344919.0,
+      "step": 2740
+    },
+    {
+      "epoch": 6.603500301750151,
+      "eval_entropy": 0.3351304212145591,
+      "eval_loss": 0.7975817918777466,
+      "eval_mean_token_accuracy": 0.8262755204452558,
+      "eval_num_tokens": 6344919.0,
+      "eval_runtime": 88.6027,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 2740
+    },
+    {
+      "entropy": 0.19650057442486285,
+      "epoch": 6.651780325890163,
+      "grad_norm": 0.9232484698295593,
+      "learning_rate": 0.00011473916556044933,
+      "loss": 0.15872747898101808,
+      "mean_token_accuracy": 0.9443749114871025,
+      "num_tokens": 6388338.0,
+      "step": 2760
+    },
+    {
+      "epoch": 6.651780325890163,
+      "eval_entropy": 0.31218746689598215,
+      "eval_loss": 0.8183822631835938,
+      "eval_mean_token_accuracy": 0.8263859531182921,
+      "eval_num_tokens": 6388338.0,
+      "eval_runtime": 89.2038,
+      "eval_samples_per_second": 15.919,
+      "eval_steps_per_second": 1.995,
+      "step": 2760
+    },
+    {
+      "entropy": 0.18859502710402012,
+      "epoch": 6.700060350030175,
+      "grad_norm": 0.8948624730110168,
+      "learning_rate": 0.0001118354819107644,
+      "loss": 0.15504695177078248,
+      "mean_token_accuracy": 0.9447060629725457,
+      "num_tokens": 6433425.0,
+      "step": 2780
+    },
+    {
+      "epoch": 6.700060350030175,
+      "eval_entropy": 0.3277170530865701,
+      "eval_loss": 0.800417423248291,
+      "eval_mean_token_accuracy": 0.826184545340163,
+      "eval_num_tokens": 6433425.0,
+      "eval_runtime": 89.1717,
+      "eval_samples_per_second": 15.924,
+      "eval_steps_per_second": 1.996,
+      "step": 2780
+    },
+    {
+      "entropy": 0.18859981009736657,
+      "epoch": 6.748340374170187,
+      "grad_norm": 0.5598214864730835,
+      "learning_rate": 0.00010895339249015836,
+      "loss": 0.1544385075569153,
+      "mean_token_accuracy": 0.9451094619929791,
+      "num_tokens": 6478249.0,
+      "step": 2800
+    },
+    {
+      "epoch": 6.748340374170187,
+      "eval_entropy": 0.3262680545449257,
+      "eval_loss": 0.7925727367401123,
+      "eval_mean_token_accuracy": 0.8272479052624006,
+      "eval_num_tokens": 6478249.0,
+      "eval_runtime": 88.3958,
+      "eval_samples_per_second": 16.064,
+      "eval_steps_per_second": 2.014,
+      "step": 2800
+    },
+    {
+      "entropy": 0.18286875914782286,
+      "epoch": 6.796620398310199,
+      "grad_norm": 0.7288182973861694,
+      "learning_rate": 0.00010609371289584521,
+      "loss": 0.15483447313308715,
+      "mean_token_accuracy": 0.946397764235735,
+      "num_tokens": 6526597.0,
+      "step": 2820
+    },
+    {
+      "epoch": 6.796620398310199,
+      "eval_entropy": 0.3231485774008076,
+      "eval_loss": 0.8040792942047119,
+      "eval_mean_token_accuracy": 0.8263571630702929,
+      "eval_num_tokens": 6526597.0,
+      "eval_runtime": 88.6613,
+      "eval_samples_per_second": 16.016,
+      "eval_steps_per_second": 2.008,
+      "step": 2820
+    },
+    {
+      "entropy": 0.1826595276594162,
+      "epoch": 6.844900422450211,
+      "grad_norm": 0.7522679567337036,
+      "learning_rate": 0.0001032572523833236,
+      "loss": 0.15428470373153685,
+      "mean_token_accuracy": 0.9460848644375801,
+      "num_tokens": 6573255.0,
+      "step": 2840
+    },
+    {
+      "epoch": 6.844900422450211,
+      "eval_entropy": 0.31538148190868037,
+      "eval_loss": 0.8011861443519592,
+      "eval_mean_token_accuracy": 0.8273241754328267,
+      "eval_num_tokens": 6573255.0,
+      "eval_runtime": 88.9392,
+      "eval_samples_per_second": 15.966,
+      "eval_steps_per_second": 2.001,
+      "step": 2840
+    },
+    {
+      "entropy": 0.18775210417807103,
+      "epoch": 6.8931804465902236,
+      "grad_norm": 0.9014021754264832,
+      "learning_rate": 0.00010044481363736635,
+      "loss": 0.15666497945785524,
+      "mean_token_accuracy": 0.9447238989174366,
+      "num_tokens": 6619167.0,
+      "step": 2860
+    },
+    {
+      "epoch": 6.8931804465902236,
+      "eval_entropy": 0.3193651750851213,
+      "eval_loss": 0.8041541576385498,
+      "eval_mean_token_accuracy": 0.8278855812013819,
+      "eval_num_tokens": 6619167.0,
+      "eval_runtime": 88.4814,
+      "eval_samples_per_second": 16.049,
+      "eval_steps_per_second": 2.012,
+      "step": 2860
+    },
+    {
+      "entropy": 0.17472777999937533,
+      "epoch": 6.941460470730235,
+      "grad_norm": 0.6994450092315674,
+      "learning_rate": 9.765719254487061e-05,
+      "loss": 0.14739950895309448,
+      "mean_token_accuracy": 0.9484775498509407,
+      "num_tokens": 6669111.0,
+      "step": 2880
+    },
+    {
+      "epoch": 6.941460470730235,
+      "eval_entropy": 0.314815216436145,
+      "eval_loss": 0.8106184601783752,
+      "eval_mean_token_accuracy": 0.8279167480683058,
+      "eval_num_tokens": 6669111.0,
+      "eval_runtime": 89.0502,
+      "eval_samples_per_second": 15.946,
+      "eval_steps_per_second": 1.999,
+      "step": 2880
+    },
+    {
+      "entropy": 0.17916967235505582,
+      "epoch": 6.989740494870247,
+      "grad_norm": 0.6007354259490967,
+      "learning_rate": 9.489517796963125e-05,
+      "loss": 0.1507383704185486,
+      "mean_token_accuracy": 0.9476396568119526,
+      "num_tokens": 6716225.0,
+      "step": 2900
+    },
+    {
+      "epoch": 6.989740494870247,
+      "eval_entropy": 0.32720738933997207,
+      "eval_loss": 0.7926715016365051,
+      "eval_mean_token_accuracy": 0.8280609012319801,
+      "eval_num_tokens": 6716225.0,
+      "eval_runtime": 88.0383,
+      "eval_samples_per_second": 16.129,
+      "eval_steps_per_second": 2.022,
+      "step": 2900
+    },
+    {
+      "entropy": 0.15926258372409002,
+      "epoch": 7.036210018105009,
+      "grad_norm": 0.44286999106407166,
+      "learning_rate": 9.21595515291015e-05,
+      "loss": 0.10761069059371949,
+      "mean_token_accuracy": 0.9592437008758644,
+      "num_tokens": 6762449.0,
+      "step": 2920
+    },
+    {
+      "epoch": 7.036210018105009,
+      "eval_entropy": 0.2828580670979586,
+      "eval_loss": 0.9056740999221802,
+      "eval_mean_token_accuracy": 0.8254911166228606,
+      "eval_num_tokens": 6762449.0,
+      "eval_runtime": 88.2719,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.016,
+      "step": 2920
+    },
+    {
+      "entropy": 0.11310737961903214,
+      "epoch": 7.084490042245021,
+      "grad_norm": 0.5048796534538269,
+      "learning_rate": 8.945108737320479e-05,
+      "loss": 0.09092960357666016,
+      "mean_token_accuracy": 0.9674679920077324,
+      "num_tokens": 6808754.0,
+      "step": 2940
+    },
+    {
+      "epoch": 7.084490042245021,
+      "eval_entropy": 0.2810762938153878,
+      "eval_loss": 0.8879074454307556,
+      "eval_mean_token_accuracy": 0.82501031709521,
+      "eval_num_tokens": 6808754.0,
+      "eval_runtime": 88.1129,
+      "eval_samples_per_second": 16.116,
+      "eval_steps_per_second": 2.02,
+      "step": 2940
+    },
+    {
+      "entropy": 0.12247815132141113,
+      "epoch": 7.132770066385033,
+      "grad_norm": 0.5650779604911804,
+      "learning_rate": 8.677055196525914e-05,
+      "loss": 0.09224110841751099,
+      "mean_token_accuracy": 0.9685524098575116,
+      "num_tokens": 6854766.0,
+      "step": 2960
+    },
+    {
+      "epoch": 7.132770066385033,
+      "eval_entropy": 0.27917667369494276,
+      "eval_loss": 0.895638644695282,
+      "eval_mean_token_accuracy": 0.8260931526676992,
+      "eval_num_tokens": 6854766.0,
+      "eval_runtime": 88.1172,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 2960
+    },
+    {
+      "entropy": 0.1326121488586068,
+      "epoch": 7.181050090525045,
+      "grad_norm": 0.4843117296695709,
+      "learning_rate": 8.411870386507724e-05,
+      "loss": 0.09537842869758606,
+      "mean_token_accuracy": 0.9645151920616627,
+      "num_tokens": 6898444.0,
+      "step": 2980
+    },
+    {
+      "epoch": 7.181050090525045,
+      "eval_entropy": 0.27999477783280813,
+      "eval_loss": 0.9031037092208862,
+      "eval_mean_token_accuracy": 0.8261882204018282,
+      "eval_num_tokens": 6898444.0,
+      "eval_runtime": 88.5602,
+      "eval_samples_per_second": 16.034,
+      "eval_steps_per_second": 2.01,
+      "step": 2980
+    },
+    {
+      "entropy": 0.1156751093454659,
+      "epoch": 7.229330114665057,
+      "grad_norm": 0.5297922492027283,
+      "learning_rate": 8.149629351430326e-05,
+      "loss": 0.09203402996063233,
+      "mean_token_accuracy": 0.9689733698964119,
+      "num_tokens": 6947106.0,
+      "step": 3000
+    },
+    {
+      "epoch": 7.229330114665057,
+      "eval_entropy": 0.2740225845508361,
+      "eval_loss": 0.8876440525054932,
+      "eval_mean_token_accuracy": 0.8265116857678703,
+      "eval_num_tokens": 6947106.0,
+      "eval_runtime": 87.7971,
+      "eval_samples_per_second": 16.174,
+      "eval_steps_per_second": 2.027,
+      "step": 3000
+    },
+    {
+      "entropy": 0.12116114404052496,
+      "epoch": 7.27761013880507,
+      "grad_norm": 0.5467398166656494,
+      "learning_rate": 7.890406302404618e-05,
+      "loss": 0.0933056652545929,
+      "mean_token_accuracy": 0.9676455594599247,
+      "num_tokens": 6992784.0,
+      "step": 3020
+    },
+    {
+      "epoch": 7.27761013880507,
+      "eval_entropy": 0.27238342793804876,
+      "eval_loss": 0.9166117310523987,
+      "eval_mean_token_accuracy": 0.825499015242866,
+      "eval_num_tokens": 6992784.0,
+      "eval_runtime": 88.3276,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 3020
+    },
+    {
+      "entropy": 0.11715119928121567,
+      "epoch": 7.325890162945082,
+      "grad_norm": 0.4976702332496643,
+      "learning_rate": 7.634274596487176e-05,
+      "loss": 0.09449659585952759,
+      "mean_token_accuracy": 0.966506478190422,
+      "num_tokens": 7040482.0,
+      "step": 3040
+    },
+    {
+      "epoch": 7.325890162945082,
+      "eval_entropy": 0.2788975556077582,
+      "eval_loss": 0.8833065629005432,
+      "eval_mean_token_accuracy": 0.8269920596915684,
+      "eval_num_tokens": 7040482.0,
+      "eval_runtime": 88.9016,
+      "eval_samples_per_second": 15.973,
+      "eval_steps_per_second": 2.002,
+      "step": 3040
+    },
+    {
+      "entropy": 0.12565911412239075,
+      "epoch": 7.3741701870850935,
+      "grad_norm": 0.5728679895401001,
+      "learning_rate": 7.381306715921025e-05,
+      "loss": 0.09626244902610778,
+      "mean_token_accuracy": 0.9661444149911403,
+      "num_tokens": 7086261.0,
+      "step": 3060
+    },
+    {
+      "epoch": 7.3741701870850935,
+      "eval_entropy": 0.2762492011939542,
+      "eval_loss": 0.8961160778999329,
+      "eval_mean_token_accuracy": 0.826369910092836,
+      "eval_num_tokens": 7086261.0,
+      "eval_runtime": 88.5204,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 3060
+    },
+    {
+      "entropy": 0.12258567921817302,
+      "epoch": 7.422450211225105,
+      "grad_norm": 0.468985915184021,
+      "learning_rate": 7.131574247624031e-05,
+      "loss": 0.09120266437530518,
+      "mean_token_accuracy": 0.9674938447773457,
+      "num_tokens": 7133247.0,
+      "step": 3080
+    },
+    {
+      "epoch": 7.422450211225105,
+      "eval_entropy": 0.2688512086366,
+      "eval_loss": 0.9048654437065125,
+      "eval_mean_token_accuracy": 0.8265582302982888,
+      "eval_num_tokens": 7133247.0,
+      "eval_runtime": 88.164,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 3080
+    },
+    {
+      "entropy": 0.12070536343380808,
+      "epoch": 7.470730235365117,
+      "grad_norm": 0.5310841798782349,
+      "learning_rate": 6.885147862930688e-05,
+      "loss": 0.09386247992515565,
+      "mean_token_accuracy": 0.96664222702384,
+      "num_tokens": 7177810.0,
+      "step": 3100
+    },
+    {
+      "epoch": 7.470730235365117,
+      "eval_entropy": 0.26289632874593305,
+      "eval_loss": 0.902837336063385,
+      "eval_mean_token_accuracy": 0.828052184554968,
+      "eval_num_tokens": 7177810.0,
+      "eval_runtime": 88.1201,
+      "eval_samples_per_second": 16.114,
+      "eval_steps_per_second": 2.02,
+      "step": 3100
+    },
+    {
+      "entropy": 0.11322665181942285,
+      "epoch": 7.51901025950513,
+      "grad_norm": 0.6846067905426025,
+      "learning_rate": 6.642097297592908e-05,
+      "loss": 0.08772538304328918,
+      "mean_token_accuracy": 0.9691362135112286,
+      "num_tokens": 7225302.0,
+      "step": 3120
+    },
+    {
+      "epoch": 7.51901025950513,
+      "eval_entropy": 0.2670605613776807,
+      "eval_loss": 0.9130566120147705,
+      "eval_mean_token_accuracy": 0.826366903741708,
+      "eval_num_tokens": 7225302.0,
+      "eval_runtime": 88.6035,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3120
+    },
+    {
+      "entropy": 0.10985926580615342,
+      "epoch": 7.567290283645142,
+      "grad_norm": 0.4799748361110687,
+      "learning_rate": 6.402491332045683e-05,
+      "loss": 0.08672296404838561,
+      "mean_token_accuracy": 0.9698824048042297,
+      "num_tokens": 7274881.0,
+      "step": 3140
+    },
+    {
+      "epoch": 7.567290283645142,
+      "eval_entropy": 0.26657010866015146,
+      "eval_loss": 0.9056380987167358,
+      "eval_mean_token_accuracy": 0.8272326461384806,
+      "eval_num_tokens": 7274881.0,
+      "eval_runtime": 88.2967,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3140
+    },
+    {
+      "entropy": 0.11890473761595785,
+      "epoch": 7.615570307785154,
+      "grad_norm": 0.4081912934780121,
+      "learning_rate": 6.166397771942987e-05,
+      "loss": 0.09396827220916748,
+      "mean_token_accuracy": 0.967512384802103,
+      "num_tokens": 7321163.0,
+      "step": 3160
+    },
+    {
+      "epoch": 7.615570307785154,
+      "eval_entropy": 0.26364776575833226,
+      "eval_loss": 0.9204733967781067,
+      "eval_mean_token_accuracy": 0.827419729714983,
+      "eval_num_tokens": 7321163.0,
+      "eval_runtime": 88.3441,
+      "eval_samples_per_second": 16.074,
+      "eval_steps_per_second": 2.015,
+      "step": 3160
+    },
+    {
+      "entropy": 0.12209134344011545,
+      "epoch": 7.663850331925166,
+      "grad_norm": 0.5159774422645569,
+      "learning_rate": 5.93388342896962e-05,
+      "loss": 0.09292978644371033,
+      "mean_token_accuracy": 0.9658421002328396,
+      "num_tokens": 7366217.0,
+      "step": 3180
+    },
+    {
+      "epoch": 7.663850331925166,
+      "eval_entropy": 0.2714110068725736,
+      "eval_loss": 0.9143092036247253,
+      "eval_mean_token_accuracy": 0.8262251099843657,
+      "eval_num_tokens": 7366217.0,
+      "eval_runtime": 87.9635,
+      "eval_samples_per_second": 16.143,
+      "eval_steps_per_second": 2.024,
+      "step": 3180
+    },
+    {
+      "entropy": 0.11955609126016498,
+      "epoch": 7.712130356065178,
+      "grad_norm": 0.7005976438522339,
+      "learning_rate": 5.705014101934252e-05,
+      "loss": 0.09319776892662049,
+      "mean_token_accuracy": 0.9674513377249241,
+      "num_tokens": 7410887.0,
+      "step": 3200
+    },
+    {
+      "epoch": 7.712130356065178,
+      "eval_entropy": 0.2626291513442993,
+      "eval_loss": 0.91884446144104,
+      "eval_mean_token_accuracy": 0.827563955207889,
+      "eval_num_tokens": 7410887.0,
+      "eval_runtime": 87.8849,
+      "eval_samples_per_second": 16.157,
+      "eval_steps_per_second": 2.025,
+      "step": 3200
+    },
+    {
+      "entropy": 0.11767532899975777,
+      "epoch": 7.76041038020519,
+      "grad_norm": 0.4675362706184387,
+      "learning_rate": 5.479854558149187e-05,
+      "loss": 0.09038087725639343,
+      "mean_token_accuracy": 0.9669680051505566,
+      "num_tokens": 7457578.0,
+      "step": 3220
+    },
+    {
+      "epoch": 7.76041038020519,
+      "eval_entropy": 0.26581037203582486,
+      "eval_loss": 0.9146883487701416,
+      "eval_mean_token_accuracy": 0.8270345812433222,
+      "eval_num_tokens": 7457578.0,
+      "eval_runtime": 88.4908,
+      "eval_samples_per_second": 16.047,
+      "eval_steps_per_second": 2.012,
+      "step": 3220
+    },
+    {
+      "entropy": 0.11383061073720455,
+      "epoch": 7.808690404345202,
+      "grad_norm": 0.541247546672821,
+      "learning_rate": 5.258468515101918e-05,
+      "loss": 0.08942405581474304,
+      "mean_token_accuracy": 0.968398705124855,
+      "num_tokens": 7505761.0,
+      "step": 3240
+    },
+    {
+      "epoch": 7.808690404345202,
+      "eval_entropy": 0.26687998960861997,
+      "eval_loss": 0.9037085175514221,
+      "eval_mean_token_accuracy": 0.8274977863504646,
+      "eval_num_tokens": 7505761.0,
+      "eval_runtime": 88.5163,
+      "eval_samples_per_second": 16.042,
+      "eval_steps_per_second": 2.011,
+      "step": 3240
+    },
+    {
+      "entropy": 0.12004058873280883,
+      "epoch": 7.856970428485214,
+      "grad_norm": 0.5289552211761475,
+      "learning_rate": 5.040918622423899e-05,
+      "loss": 0.09064805507659912,
+      "mean_token_accuracy": 0.9673330299556255,
+      "num_tokens": 7552557.0,
+      "step": 3260
+    },
+    {
+      "epoch": 7.856970428485214,
+      "eval_entropy": 0.26288379251622085,
+      "eval_loss": 0.9146178364753723,
+      "eval_mean_token_accuracy": 0.8276662448149049,
+      "eval_num_tokens": 7552557.0,
+      "eval_runtime": 88.0723,
+      "eval_samples_per_second": 16.123,
+      "eval_steps_per_second": 2.021,
+      "step": 3260
+    },
+    {
+      "entropy": 0.1180689456872642,
+      "epoch": 7.905250452625227,
+      "grad_norm": 0.4754861891269684,
+      "learning_rate": 4.827266444161377e-05,
+      "loss": 0.09402655363082886,
+      "mean_token_accuracy": 0.9663361787796021,
+      "num_tokens": 7596998.0,
+      "step": 3280
+    },
+    {
+      "epoch": 7.905250452625227,
+      "eval_entropy": 0.261690996922134,
+      "eval_loss": 0.9120186567306519,
+      "eval_mean_token_accuracy": 0.8278995015647974,
+      "eval_num_tokens": 7596998.0,
+      "eval_runtime": 88.6573,
+      "eval_samples_per_second": 16.017,
+      "eval_steps_per_second": 2.008,
+      "step": 3280
+    },
+    {
+      "entropy": 0.1162164019420743,
+      "epoch": 7.953530476765239,
+      "grad_norm": 0.6203398108482361,
+      "learning_rate": 4.61757244135355e-05,
+      "loss": 0.08832768201828003,
+      "mean_token_accuracy": 0.9694220684468746,
+      "num_tokens": 7643288.0,
+      "step": 3300
+    },
+    {
+      "epoch": 7.953530476765239,
+      "eval_entropy": 0.26158678129817664,
+      "eval_loss": 0.9301361441612244,
+      "eval_mean_token_accuracy": 0.8269603885291668,
+      "eval_num_tokens": 7643288.0,
+      "eval_runtime": 88.7064,
+      "eval_samples_per_second": 16.008,
+      "eval_steps_per_second": 2.007,
+      "step": 3300
+    },
+    {
+      "entropy": 0.10877202204489089,
+      "epoch": 8.0,
+      "grad_norm": 3.058673143386841,
+      "learning_rate": 4.411895954922763e-05,
+      "loss": 0.09159615635871887,
+      "mean_token_accuracy": 0.9686878834451947,
+      "num_tokens": 7688608.0,
+      "step": 3320
+    },
+    {
+      "epoch": 8.0,
+      "eval_entropy": 0.2590438057197614,
+      "eval_loss": 0.9262065887451172,
+      "eval_mean_token_accuracy": 0.8277404546067956,
+      "eval_num_tokens": 7688608.0,
+      "eval_runtime": 87.7557,
+      "eval_samples_per_second": 16.181,
+      "eval_steps_per_second": 2.028,
+      "step": 3320
+    },
+    {
+      "entropy": 0.09969092034734786,
+      "epoch": 8.048280024140013,
+      "grad_norm": 0.2561911642551422,
+      "learning_rate": 4.2102951888817644e-05,
+      "loss": 0.07228735685348511,
+      "mean_token_accuracy": 0.9743556551635265,
+      "num_tokens": 7732856.0,
+      "step": 3340
+    },
+    {
+      "epoch": 8.048280024140013,
+      "eval_entropy": 0.2462534454933713,
+      "eval_loss": 0.9755983352661133,
+      "eval_mean_token_accuracy": 0.827510209565752,
+      "eval_num_tokens": 7732856.0,
+      "eval_runtime": 88.0297,
+      "eval_samples_per_second": 16.131,
+      "eval_steps_per_second": 2.022,
+      "step": 3340
+    },
+    {
+      "entropy": 0.09477624730207027,
+      "epoch": 8.096560048280024,
+      "grad_norm": 0.3670356571674347,
+      "learning_rate": 4.012827193862633e-05,
+      "loss": 0.0686036467552185,
+      "mean_token_accuracy": 0.9743670664727688,
+      "num_tokens": 7778585.0,
+      "step": 3360
+    },
+    {
+      "epoch": 8.096560048280024,
+      "eval_entropy": 0.2440882951356052,
+      "eval_loss": 1.0118423700332642,
+      "eval_mean_token_accuracy": 0.8271241070849172,
+      "eval_num_tokens": 7778585.0,
+      "eval_runtime": 88.1364,
+      "eval_samples_per_second": 16.111,
+      "eval_steps_per_second": 2.02,
+      "step": 3360
+    },
+    {
+      "entropy": 0.09417980071157217,
+      "epoch": 8.144840072420036,
+      "grad_norm": 0.5785111784934998,
+      "learning_rate": 3.819547850972133e-05,
+      "loss": 0.07100698947906495,
+      "mean_token_accuracy": 0.9737804099917412,
+      "num_tokens": 7824287.0,
+      "step": 3380
+    },
+    {
+      "epoch": 8.144840072420036,
+      "eval_entropy": 0.24603172860453637,
+      "eval_loss": 1.0021991729736328,
+      "eval_mean_token_accuracy": 0.827162122793412,
+      "eval_num_tokens": 7824287.0,
+      "eval_runtime": 88.4534,
+      "eval_samples_per_second": 16.054,
+      "eval_steps_per_second": 2.012,
+      "step": 3380
+    },
+    {
+      "entropy": 0.087222247896716,
+      "epoch": 8.193120096560047,
+      "grad_norm": 0.44848504662513733,
+      "learning_rate": 3.6305118559780393e-05,
+      "loss": 0.06325234174728393,
+      "mean_token_accuracy": 0.9769335232675076,
+      "num_tokens": 7874784.0,
+      "step": 3400
+    },
+    {
+      "epoch": 8.193120096560047,
+      "eval_entropy": 0.24544863905129807,
+      "eval_loss": 1.005008339881897,
+      "eval_mean_token_accuracy": 0.8273035240976998,
+      "eval_num_tokens": 7874784.0,
+      "eval_runtime": 88.6737,
+      "eval_samples_per_second": 16.014,
+      "eval_steps_per_second": 2.007,
+      "step": 3400
+    },
+    {
+      "entropy": 0.09037275062873959,
+      "epoch": 8.24140012070006,
+      "grad_norm": 0.23529155552387238,
+      "learning_rate": 3.4457727038308435e-05,
+      "loss": 0.06589569449424744,
+      "mean_token_accuracy": 0.9751470290124417,
+      "num_tokens": 7925240.0,
+      "step": 3420
+    },
+    {
+      "epoch": 8.24140012070006,
+      "eval_entropy": 0.24203423494368456,
+      "eval_loss": 1.0052824020385742,
+      "eval_mean_token_accuracy": 0.8275575503874361,
+      "eval_num_tokens": 7925240.0,
+      "eval_runtime": 88.3486,
+      "eval_samples_per_second": 16.073,
+      "eval_steps_per_second": 2.015,
+      "step": 3420
+    },
+    {
+      "entropy": 0.10373208015225828,
+      "epoch": 8.289680144840073,
+      "grad_norm": 0.24796408414840698,
+      "learning_rate": 3.2653826735253594e-05,
+      "loss": 0.07714771628379821,
+      "mean_token_accuracy": 0.9722891598939896,
+      "num_tokens": 7966685.0,
+      "step": 3440
+    },
+    {
+      "epoch": 8.289680144840073,
+      "eval_entropy": 0.23461157050025597,
+      "eval_loss": 1.012108325958252,
+      "eval_mean_token_accuracy": 0.82825900564033,
+      "eval_num_tokens": 7966685.0,
+      "eval_runtime": 88.1018,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 3440
+    },
+    {
+      "entropy": 0.08967954274266958,
+      "epoch": 8.337960168980084,
+      "grad_norm": 0.5314674973487854,
+      "learning_rate": 3.08939281330635e-05,
+      "loss": 0.06746266484260559,
+      "mean_token_accuracy": 0.9753133699297905,
+      "num_tokens": 8013656.0,
+      "step": 3460
+    },
+    {
+      "epoch": 8.337960168980084,
+      "eval_entropy": 0.23781998589467468,
+      "eval_loss": 1.0119913816452026,
+      "eval_mean_token_accuracy": 0.8273287157664139,
+      "eval_num_tokens": 8013656.0,
+      "eval_runtime": 88.6038,
+      "eval_samples_per_second": 16.026,
+      "eval_steps_per_second": 2.009,
+      "step": 3460
+    },
+    {
+      "entropy": 0.09280927330255509,
+      "epoch": 8.386240193120097,
+      "grad_norm": 0.30658936500549316,
+      "learning_rate": 2.9178529262224946e-05,
+      "loss": 0.06936421990394592,
+      "mean_token_accuracy": 0.9742105580866337,
+      "num_tokens": 8061108.0,
+      "step": 3480
+    },
+    {
+      "epoch": 8.386240193120097,
+      "eval_entropy": 0.23672426734747512,
+      "eval_loss": 1.0162041187286377,
+      "eval_mean_token_accuracy": 0.8278394281194451,
+      "eval_num_tokens": 8061108.0,
+      "eval_runtime": 88.6015,
+      "eval_samples_per_second": 16.027,
+      "eval_steps_per_second": 2.009,
+      "step": 3480
+    },
+    {
+      "entropy": 0.09195122267119586,
+      "epoch": 8.43452021726011,
+      "grad_norm": 0.24512454867362976,
+      "learning_rate": 2.7508115560327256e-05,
+      "loss": 0.06996064782142639,
+      "mean_token_accuracy": 0.974104854464531,
+      "num_tokens": 8106924.0,
+      "step": 3500
+    },
+    {
+      "epoch": 8.43452021726011,
+      "eval_entropy": 0.23758245828781235,
+      "eval_loss": 1.0157747268676758,
+      "eval_mean_token_accuracy": 0.8278501579600773,
+      "eval_num_tokens": 8106924.0,
+      "eval_runtime": 88.5678,
+      "eval_samples_per_second": 16.033,
+      "eval_steps_per_second": 2.01,
+      "step": 3500
+    },
+    {
+      "entropy": 0.09518477590754629,
+      "epoch": 8.48280024140012,
+      "grad_norm": 0.2373228371143341,
+      "learning_rate": 2.5883159734688868e-05,
+      "loss": 0.06832797527313232,
+      "mean_token_accuracy": 0.9743806682527065,
+      "num_tokens": 8153070.0,
+      "step": 3520
+    },
+    {
+      "epoch": 8.48280024140012,
+      "eval_entropy": 0.2379308074210467,
+      "eval_loss": 1.0180530548095703,
+      "eval_mean_token_accuracy": 0.8282477052024241,
+      "eval_num_tokens": 8153070.0,
+      "eval_runtime": 88.4651,
+      "eval_samples_per_second": 16.052,
+      "eval_steps_per_second": 2.012,
+      "step": 3520
+    },
+    {
+      "entropy": 0.0906269280705601,
+      "epoch": 8.531080265540133,
+      "grad_norm": 0.37602657079696655,
+      "learning_rate": 2.4304121628587173e-05,
+      "loss": 0.06921588778495788,
+      "mean_token_accuracy": 0.9745571479201317,
+      "num_tokens": 8200066.0,
+      "step": 3540
+    },
+    {
+      "epoch": 8.531080265540133,
+      "eval_entropy": 0.2357490769263064,
+      "eval_loss": 1.017139196395874,
+      "eval_mean_token_accuracy": 0.828713178299786,
+      "eval_num_tokens": 8200066.0,
+      "eval_runtime": 88.6424,
+      "eval_samples_per_second": 16.019,
+      "eval_steps_per_second": 2.008,
+      "step": 3540
+    },
+    {
+      "entropy": 0.09273334196768701,
+      "epoch": 8.579360289680144,
+      "grad_norm": 0.2978838086128235,
+      "learning_rate": 2.277144809112794e-05,
+      "loss": 0.06970887780189514,
+      "mean_token_accuracy": 0.9751374177634716,
+      "num_tokens": 8246157.0,
+      "step": 3560
+    },
+    {
+      "epoch": 8.579360289680144,
+      "eval_entropy": 0.23590933138065123,
+      "eval_loss": 1.019231915473938,
+      "eval_mean_token_accuracy": 0.8279163412163767,
+      "eval_num_tokens": 8246157.0,
+      "eval_runtime": 88.2691,
+      "eval_samples_per_second": 16.087,
+      "eval_steps_per_second": 2.017,
+      "step": 3560
+    },
+    {
+      "entropy": 0.09854201921261847,
+      "epoch": 8.627640313820157,
+      "grad_norm": 0.5381243228912354,
+      "learning_rate": 2.1285572850792413e-05,
+      "loss": 0.07123501300811767,
+      "mean_token_accuracy": 0.9723988175392151,
+      "num_tokens": 8291752.0,
+      "step": 3580
+    },
+    {
+      "epoch": 8.627640313820157,
+      "eval_entropy": 0.23592024949494372,
+      "eval_loss": 1.020250678062439,
+      "eval_mean_token_accuracy": 0.8282466226079491,
+      "eval_num_tokens": 8291752.0,
+      "eval_runtime": 87.8939,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 2.025,
+      "step": 3580
+    },
+    {
+      "entropy": 0.09092000788077712,
+      "epoch": 8.67592033796017,
+      "grad_norm": 0.29089561104774475,
+      "learning_rate": 1.9846916392697493e-05,
+      "loss": 0.06982720494270325,
+      "mean_token_accuracy": 0.9749995484948158,
+      "num_tokens": 8337651.0,
+      "step": 3600
+    },
+    {
+      "epoch": 8.67592033796017,
+      "eval_entropy": 0.23658913382318583,
+      "eval_loss": 1.019862413406372,
+      "eval_mean_token_accuracy": 0.8280127842104836,
+      "eval_num_tokens": 8337651.0,
+      "eval_runtime": 88.2855,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 3600
+    },
+    {
+      "entropy": 0.08708424470387399,
+      "epoch": 8.72420036210018,
+      "grad_norm": 0.29904839396476746,
+      "learning_rate": 1.8455885839603106e-05,
+      "loss": 0.06801370978355407,
+      "mean_token_accuracy": 0.9755801670253277,
+      "num_tokens": 8385330.0,
+      "step": 3620
+    },
+    {
+      "epoch": 8.72420036210018,
+      "eval_entropy": 0.23627035771862845,
+      "eval_loss": 1.0204569101333618,
+      "eval_mean_token_accuracy": 0.8279967894045155,
+      "eval_num_tokens": 8385330.0,
+      "eval_runtime": 88.1753,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 3620
+    },
+    {
+      "entropy": 0.0916255566291511,
+      "epoch": 8.772480386240193,
+      "grad_norm": 0.20571033656597137,
+      "learning_rate": 1.7112874836701502e-05,
+      "loss": 0.0692249894142151,
+      "mean_token_accuracy": 0.9745148062705994,
+      "num_tokens": 8432134.0,
+      "step": 3640
+    },
+    {
+      "epoch": 8.772480386240193,
+      "eval_entropy": 0.2340821772813797,
+      "eval_loss": 1.0212821960449219,
+      "eval_mean_token_accuracy": 0.828373622358515,
+      "eval_num_tokens": 8432134.0,
+      "eval_runtime": 87.2518,
+      "eval_samples_per_second": 16.275,
+      "eval_steps_per_second": 2.04,
+      "step": 3640
+    },
+    {
+      "entropy": 0.0926868049427867,
+      "epoch": 8.820760410380204,
+      "grad_norm": 0.2618265450000763,
+      "learning_rate": 1.5818263440220203e-05,
+      "loss": 0.07338647246360779,
+      "mean_token_accuracy": 0.9724708944559097,
+      "num_tokens": 8476758.0,
+      "step": 3660
+    },
+    {
+      "epoch": 8.820760410380204,
+      "eval_entropy": 0.23637074575330433,
+      "eval_loss": 1.0186725854873657,
+      "eval_mean_token_accuracy": 0.8277707605549459,
+      "eval_num_tokens": 8476758.0,
+      "eval_runtime": 88.4695,
+      "eval_samples_per_second": 16.051,
+      "eval_steps_per_second": 2.012,
+      "step": 3660
+    },
+    {
+      "entropy": 0.09175782897509635,
+      "epoch": 8.869040434520217,
+      "grad_norm": 0.2889852821826935,
+      "learning_rate": 1.4572418009870604e-05,
+      "loss": 0.07066041231155396,
+      "mean_token_accuracy": 0.9725673109292984,
+      "num_tokens": 8523294.0,
+      "step": 3680
+    },
+    {
+      "epoch": 8.869040434520217,
+      "eval_entropy": 0.23945164404223473,
+      "eval_loss": 1.0145231485366821,
+      "eval_mean_token_accuracy": 0.8281202523895864,
+      "eval_num_tokens": 8523294.0,
+      "eval_runtime": 88.368,
+      "eval_samples_per_second": 16.069,
+      "eval_steps_per_second": 2.014,
+      "step": 3680
+    },
+    {
+      "entropy": 0.09265611860901117,
+      "epoch": 8.91732045866023,
+      "grad_norm": 0.41092950105667114,
+      "learning_rate": 1.3375691105172625e-05,
+      "loss": 0.06975027322769164,
+      "mean_token_accuracy": 0.974636297672987,
+      "num_tokens": 8570358.0,
+      "step": 3700
+    },
+    {
+      "epoch": 8.91732045866023,
+      "eval_entropy": 0.239588919901446,
+      "eval_loss": 1.014059066772461,
+      "eval_mean_token_accuracy": 0.828028041660116,
+      "eval_num_tokens": 8570358.0,
+      "eval_runtime": 88.1039,
+      "eval_samples_per_second": 16.117,
+      "eval_steps_per_second": 2.02,
+      "step": 3700
+    },
+    {
+      "entropy": 0.0896465003490448,
+      "epoch": 8.965600482800241,
+      "grad_norm": 0.24000254273414612,
+      "learning_rate": 1.2228421385684305e-05,
+      "loss": 0.06743953824043274,
+      "mean_token_accuracy": 0.974006575345993,
+      "num_tokens": 8618246.0,
+      "step": 3720
+    },
+    {
+      "epoch": 8.965600482800241,
+      "eval_entropy": 0.2390461369846644,
+      "eval_loss": 1.011997938156128,
+      "eval_mean_token_accuracy": 0.8284533211354459,
+      "eval_num_tokens": 8618246.0,
+      "eval_runtime": 88.6202,
+      "eval_samples_per_second": 16.023,
+      "eval_steps_per_second": 2.009,
+      "step": 3720
+    },
+    {
+      "entropy": 0.09575109251520851,
+      "epoch": 9.012070006035003,
+      "grad_norm": 0.18088576197624207,
+      "learning_rate": 1.1130933515165429e-05,
+      "loss": 0.07295305132865906,
+      "mean_token_accuracy": 0.9739722857227573,
+      "num_tokens": 8661177.0,
+      "step": 3740
+    },
+    {
+      "epoch": 9.012070006035003,
+      "eval_entropy": 0.23728882387447892,
+      "eval_loss": 1.0134038925170898,
+      "eval_mean_token_accuracy": 0.8286508951294288,
+      "eval_num_tokens": 8661177.0,
+      "eval_runtime": 88.6434,
+      "eval_samples_per_second": 16.019,
+      "eval_steps_per_second": 2.008,
+      "step": 3740
+    },
+    {
+      "entropy": 0.08883201805874705,
+      "epoch": 9.060350030175016,
+      "grad_norm": 0.23120898008346558,
+      "learning_rate": 1.008353806970132e-05,
+      "loss": 0.06326142549514771,
+      "mean_token_accuracy": 0.9780020996928215,
+      "num_tokens": 8707602.0,
+      "step": 3760
+    },
+    {
+      "epoch": 9.060350030175016,
+      "eval_entropy": 0.23416905331142832,
+      "eval_loss": 1.0284100770950317,
+      "eval_mean_token_accuracy": 0.8284083213029283,
+      "eval_num_tokens": 8707602.0,
+      "eval_runtime": 88.2933,
+      "eval_samples_per_second": 16.083,
+      "eval_steps_per_second": 2.016,
+      "step": 3760
+    },
+    {
+      "entropy": 0.08861468276008963,
+      "epoch": 9.108630054315027,
+      "grad_norm": 0.17277656495571136,
+      "learning_rate": 9.086531449813489e-06,
+      "loss": 0.06727416515350342,
+      "mean_token_accuracy": 0.9760390736162663,
+      "num_tokens": 8751287.0,
+      "step": 3780
+    },
+    {
+      "epoch": 9.108630054315027,
+      "eval_entropy": 0.23370317386442355,
+      "eval_loss": 1.0381091833114624,
+      "eval_mean_token_accuracy": 0.8282668178670862,
+      "eval_num_tokens": 8751287.0,
+      "eval_runtime": 88.5738,
+      "eval_samples_per_second": 16.032,
+      "eval_steps_per_second": 2.01,
+      "step": 3780
+    },
+    {
+      "entropy": 0.09032954541034996,
+      "epoch": 9.15691007845504,
+      "grad_norm": 0.19481085240840912,
+      "learning_rate": 8.140195796581981e-06,
+      "loss": 0.06117786169052124,
+      "mean_token_accuracy": 0.9764454193413258,
+      "num_tokens": 8799982.0,
+      "step": 3800
+    },
+    {
+      "epoch": 9.15691007845504,
+      "eval_entropy": 0.23215936609868254,
+      "eval_loss": 1.0467767715454102,
+      "eval_mean_token_accuracy": 0.828405868806196,
+      "eval_num_tokens": 8799982.0,
+      "eval_runtime": 87.9551,
+      "eval_samples_per_second": 16.145,
+      "eval_steps_per_second": 2.024,
+      "step": 3800
+    },
+    {
+      "entropy": 0.08824449330568314,
+      "epoch": 9.20519010259505,
+      "grad_norm": 0.20382793247699738,
+      "learning_rate": 7.2447989118025455e-06,
+      "loss": 0.06422553658485412,
+      "mean_token_accuracy": 0.9763121999800205,
+      "num_tokens": 8845536.0,
+      "step": 3820
+    },
+    {
+      "epoch": 9.20519010259505,
+      "eval_entropy": 0.23169661472352704,
+      "eval_loss": 1.0529043674468994,
+      "eval_mean_token_accuracy": 0.8282025610463003,
+      "eval_num_tokens": 8845536.0,
+      "eval_runtime": 88.5548,
+      "eval_samples_per_second": 16.035,
+      "eval_steps_per_second": 2.01,
+      "step": 3820
+    },
+    {
+      "entropy": 0.07768499720841646,
+      "epoch": 9.253470126735063,
+      "grad_norm": 0.11727919429540634,
+      "learning_rate": 6.400594182202084e-06,
+      "loss": 0.057279336452484134,
+      "mean_token_accuracy": 0.9795166745781898,
+      "num_tokens": 8896053.0,
+      "step": 3840
+    },
+    {
+      "epoch": 9.253470126735063,
+      "eval_entropy": 0.2301936373114586,
+      "eval_loss": 1.0576049089431763,
+      "eval_mean_token_accuracy": 0.8283417489421502,
+      "eval_num_tokens": 8896053.0,
+      "eval_runtime": 88.7854,
+      "eval_samples_per_second": 15.994,
+      "eval_steps_per_second": 2.005,
+      "step": 3840
+    },
+    {
+      "entropy": 0.09015200780704617,
+      "epoch": 9.301750150875076,
+      "grad_norm": 0.2089058756828308,
+      "learning_rate": 5.6078205077330705e-06,
+      "loss": 0.06626194715499878,
+      "mean_token_accuracy": 0.9747506938874722,
+      "num_tokens": 8940591.0,
+      "step": 3860
+    },
+    {
+      "epoch": 9.301750150875076,
+      "eval_entropy": 0.23024900676159377,
+      "eval_loss": 1.0599786043167114,
+      "eval_mean_token_accuracy": 0.8283051915383071,
+      "eval_num_tokens": 8940591.0,
+      "eval_runtime": 88.2963,
+      "eval_samples_per_second": 16.082,
+      "eval_steps_per_second": 2.016,
+      "step": 3860
+    },
+    {
+      "entropy": 0.08484094133600592,
+      "epoch": 9.350030175015087,
+      "grad_norm": 0.21646778285503387,
+      "learning_rate": 4.866702233967785e-06,
+      "loss": 0.05998039245605469,
+      "mean_token_accuracy": 0.9771868668496608,
+      "num_tokens": 8989035.0,
+      "step": 3880
+    },
+    {
+      "epoch": 9.350030175015087,
+      "eval_entropy": 0.2300723305578982,
+      "eval_loss": 1.0633587837219238,
+      "eval_mean_token_accuracy": 0.828371500701047,
+      "eval_num_tokens": 8989035.0,
+      "eval_runtime": 88.0233,
+      "eval_samples_per_second": 16.132,
+      "eval_steps_per_second": 2.022,
+      "step": 3880
+    },
+    {
+      "entropy": 0.08905019680969417,
+      "epoch": 9.3983101991551,
+      "grad_norm": 0.21842406690120697,
+      "learning_rate": 4.177449088611095e-06,
+      "loss": 0.06666865348815917,
+      "mean_token_accuracy": 0.9753445640206337,
+      "num_tokens": 9033383.0,
+      "step": 3900
+    },
+    {
+      "epoch": 9.3983101991551,
+      "eval_entropy": 0.2295504841911659,
+      "eval_loss": 1.0662461519241333,
+      "eval_mean_token_accuracy": 0.8284122297603093,
+      "eval_num_tokens": 9033383.0,
+      "eval_runtime": 88.5525,
+      "eval_samples_per_second": 16.036,
+      "eval_steps_per_second": 2.01,
+      "step": 3900
+    },
+    {
+      "entropy": 0.08582521872594953,
+      "epoch": 9.446590223295111,
+      "grad_norm": 0.2000240683555603,
+      "learning_rate": 3.540256122150088e-06,
+      "loss": 0.06327944993972778,
+      "mean_token_accuracy": 0.9765432856976985,
+      "num_tokens": 9079672.0,
+      "step": 3920
+    },
+    {
+      "epoch": 9.446590223295111,
+      "eval_entropy": 0.2297678625818049,
+      "eval_loss": 1.0676169395446777,
+      "eval_mean_token_accuracy": 0.8283703879693921,
+      "eval_num_tokens": 9079672.0,
+      "eval_runtime": 87.6897,
+      "eval_samples_per_second": 16.193,
+      "eval_steps_per_second": 2.03,
+      "step": 3920
+    },
+    {
+      "entropy": 0.09261773955076932,
+      "epoch": 9.494870247435124,
+      "grad_norm": 0.19104281067848206,
+      "learning_rate": 2.955303652656991e-06,
+      "loss": 0.06111966967582703,
+      "mean_token_accuracy": 0.9738784030079841,
+      "num_tokens": 9127647.0,
+      "step": 3940
+    },
+    {
+      "epoch": 9.494870247435124,
+      "eval_entropy": 0.22986975881490815,
+      "eval_loss": 1.0687925815582275,
+      "eval_mean_token_accuracy": 0.8282996858103892,
+      "eval_num_tokens": 9127647.0,
+      "eval_runtime": 88.2224,
+      "eval_samples_per_second": 16.096,
+      "eval_steps_per_second": 2.018,
+      "step": 3940
+    },
+    {
+      "entropy": 0.08347254800610245,
+      "epoch": 9.543150271575136,
+      "grad_norm": 0.18671369552612305,
+      "learning_rate": 2.422757214761356e-06,
+      "loss": 0.06384263038635254,
+      "mean_token_accuracy": 0.9770128890872002,
+      "num_tokens": 9172587.0,
+      "step": 3960
+    },
+    {
+      "epoch": 9.543150271575136,
+      "eval_entropy": 0.22975517851248217,
+      "eval_loss": 1.0701013803482056,
+      "eval_mean_token_accuracy": 0.8282244285840666,
+      "eval_num_tokens": 9172587.0,
+      "eval_runtime": 88.5794,
+      "eval_samples_per_second": 16.031,
+      "eval_steps_per_second": 2.009,
+      "step": 3960
+    },
+    {
+      "entropy": 0.08136763009242713,
+      "epoch": 9.591430295715147,
+      "grad_norm": 0.24811454117298126,
+      "learning_rate": 1.942767512805673e-06,
+      "loss": 0.06044899821281433,
+      "mean_token_accuracy": 0.9776685133576393,
+      "num_tokens": 9219278.0,
+      "step": 3980
+    },
+    {
+      "epoch": 9.591430295715147,
+      "eval_entropy": 0.22938149347064202,
+      "eval_loss": 1.0711463689804077,
+      "eval_mean_token_accuracy": 0.8282721410976367,
+      "eval_num_tokens": 9219278.0,
+      "eval_runtime": 88.4978,
+      "eval_samples_per_second": 16.046,
+      "eval_steps_per_second": 2.011,
+      "step": 3980
+    },
+    {
+      "entropy": 0.09432941288687288,
+      "epoch": 9.63971031985516,
+      "grad_norm": 0.30733999609947205,
+      "learning_rate": 1.515470378197889e-06,
+      "loss": 0.0672213613986969,
+      "mean_token_accuracy": 0.9728612303733826,
+      "num_tokens": 9262346.0,
+      "step": 4000
+    },
+    {
+      "epoch": 9.63971031985516,
+      "eval_entropy": 0.22936245958121976,
+      "eval_loss": 1.071343183517456,
+      "eval_mean_token_accuracy": 0.8282827025049189,
+      "eval_num_tokens": 9262346.0,
+      "eval_runtime": 88.4428,
+      "eval_samples_per_second": 16.056,
+      "eval_steps_per_second": 2.013,
+      "step": 4000
+    },
+    {
+      "entropy": 0.08287148931995034,
+      "epoch": 9.687990343995171,
+      "grad_norm": 0.1976361870765686,
+      "learning_rate": 1.1409867309726256e-06,
+      "loss": 0.06318551301956177,
+      "mean_token_accuracy": 0.9770804084837437,
+      "num_tokens": 9309622.0,
+      "step": 4020
+    },
+    {
+      "epoch": 9.687990343995171,
+      "eval_entropy": 0.22929051920269314,
+      "eval_loss": 1.0711324214935303,
+      "eval_mean_token_accuracy": 0.8281597297513084,
+      "eval_num_tokens": 9309622.0,
+      "eval_runtime": 88.5243,
+      "eval_samples_per_second": 16.041,
+      "eval_steps_per_second": 2.011,
+      "step": 4020
+    },
+    {
+      "entropy": 0.0907150611979887,
+      "epoch": 9.736270368135184,
+      "grad_norm": 0.16192017495632172,
+      "learning_rate": 8.194225455723828e-07,
+      "loss": 0.06589602828025817,
+      "mean_token_accuracy": 0.9738550461828709,
+      "num_tokens": 9354549.0,
+      "step": 4040
+    },
+    {
+      "epoch": 9.736270368135184,
+      "eval_entropy": 0.22913936391640244,
+      "eval_loss": 1.0715173482894897,
+      "eval_mean_token_accuracy": 0.8281301544623428,
+      "eval_num_tokens": 9354549.0,
+      "eval_runtime": 88.875,
+      "eval_samples_per_second": 15.978,
+      "eval_steps_per_second": 2.003,
+      "step": 4040
+    },
+    {
+      "entropy": 0.09003224167972804,
+      "epoch": 9.784550392275197,
+      "grad_norm": 0.2092711478471756,
+      "learning_rate": 5.508688208578523e-07,
+      "loss": 0.06655729413032532,
+      "mean_token_accuracy": 0.9750426560640335,
+      "num_tokens": 9399394.0,
+      "step": 4060
+    },
+    {
+      "epoch": 9.784550392275197,
+      "eval_entropy": 0.2291239777140403,
+      "eval_loss": 1.0715563297271729,
+      "eval_mean_token_accuracy": 0.8281808595978812,
+      "eval_num_tokens": 9399394.0,
+      "eval_runtime": 88.4328,
+      "eval_samples_per_second": 16.057,
+      "eval_steps_per_second": 2.013,
+      "step": 4060
+    },
+    {
+      "entropy": 0.08719871481880545,
+      "epoch": 9.832830416415208,
+      "grad_norm": 0.17966455221176147,
+      "learning_rate": 3.3540155435645076e-07,
+      "loss": 0.05917760729789734,
+      "mean_token_accuracy": 0.9752349570393563,
+      "num_tokens": 9449483.0,
+      "step": 4080
+    },
+    {
+      "epoch": 9.832830416415208,
+      "eval_entropy": 0.2291418021816886,
+      "eval_loss": 1.0714553594589233,
+      "eval_mean_token_accuracy": 0.8281870403986299,
+      "eval_num_tokens": 9449483.0,
+      "eval_runtime": 88.2612,
+      "eval_samples_per_second": 16.089,
+      "eval_steps_per_second": 2.017,
+      "step": 4080
+    },
+    {
+      "entropy": 0.07887464743107557,
+      "epoch": 9.88111044055522,
+      "grad_norm": 0.4397202134132385,
+      "learning_rate": 1.7308172075587924e-07,
+      "loss": 0.060457843542099,
+      "mean_token_accuracy": 0.9779011741280556,
+      "num_tokens": 9497731.0,
+      "step": 4100
+    },
+    {
+      "epoch": 9.88111044055522,
+      "eval_entropy": 0.2292038949855258,
+      "eval_loss": 1.0717124938964844,
+      "eval_mean_token_accuracy": 0.8282261353530241,
+      "eval_num_tokens": 9497731.0,
+      "eval_runtime": 88.5793,
+      "eval_samples_per_second": 16.031,
+      "eval_steps_per_second": 2.009,
+      "step": 4100
+    },
+    {
+      "entropy": 0.07985925199463964,
+      "epoch": 9.929390464695233,
+      "grad_norm": 0.2129203975200653,
+      "learning_rate": 6.395525464895752e-08,
+      "loss": 0.05666370391845703,
+      "mean_token_accuracy": 0.9778168581426143,
+      "num_tokens": 9548917.0,
+      "step": 4120
+    },
+    {
+      "epoch": 9.929390464695233,
+      "eval_entropy": 0.22915430601393239,
+      "eval_loss": 1.0715631246566772,
+      "eval_mean_token_accuracy": 0.8282323436120923,
+      "eval_num_tokens": 9548917.0,
+      "eval_runtime": 88.4991,
+      "eval_samples_per_second": 16.045,
+      "eval_steps_per_second": 2.011,
+      "step": 4120
+    },
+    {
+      "entropy": 0.08802500045858323,
+      "epoch": 9.977670488835244,
+      "grad_norm": 0.18367069959640503,
+      "learning_rate": 8.053037534806845e-09,
+      "loss": 0.06633861660957337,
+      "mean_token_accuracy": 0.9749571368098259,
+      "num_tokens": 9592427.0,
+      "step": 4140
+    },
+    {
+      "epoch": 9.977670488835244,
+      "eval_entropy": 0.22921692956699413,
+      "eval_loss": 1.0715234279632568,
+      "eval_mean_token_accuracy": 0.828055245822735,
+      "eval_num_tokens": 9592427.0,
+      "eval_runtime": 88.1869,
+      "eval_samples_per_second": 16.102,
+      "eval_steps_per_second": 2.018,
+      "step": 4140
+    },
+    {
+      "epoch": 10.0,
+      "eval_entropy": 0.229176854986823,
+      "eval_loss": 1.0716373920440674,
+      "eval_mean_token_accuracy": 0.828094540352232,
+      "eval_num_tokens": 9610760.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 4150
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": true
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 4.065114277436621e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-420/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-420/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-420/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-420/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-420/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-420/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-420/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-420/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-420/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-420/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-420/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-420/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-420/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-420/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..d2b2328b64f6d67d1272c35f48b5846a170f2b51
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-420/trainer_state.json
@@ -0,0 +1,475 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.012070006035003,
+  "eval_steps": 20,
+  "global_step": 420,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 4.121406190398874e+16,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-440/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-440/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-440/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-440/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-440/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-440/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-440/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-440/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-440/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-440/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-440/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-440/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-440/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-440/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..6e24a358efbd1aa845c333f7617738640438c8d6
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-440/trainer_state.json
@@ -0,0 +1,496 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.060350030175015,
+  "eval_steps": 20,
+  "global_step": 440,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 4.335126423740621e+16,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-460/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-460/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-460/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-460/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-460/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-460/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-460/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-460/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-460/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-460/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-460/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-460/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-460/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-460/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..7ce3be5f5d5bbae25b20543d0a82f684c4b36966
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-460/trainer_state.json
@@ -0,0 +1,517 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.1086300543150271,
+  "eval_steps": 20,
+  "global_step": 460,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 4.52529392245248e+16,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-480/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-480/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-480/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-480/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-480/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-480/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-480/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-480/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-480/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-480/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-480/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-480/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-480/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-480/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..52acd0befc0359dec4e21771a0ce5d6ad078f6fe
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-480/trainer_state.json
@@ -0,0 +1,538 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.1569100784550392,
+  "eval_steps": 20,
+  "global_step": 480,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 4.726057958755738e+16,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-500/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-500/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-500/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-500/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-500/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-500/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-500/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-500/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-500/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-500/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-500/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-500/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-500/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-500/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..e7240a697ab1749665d8e524b13c36265f6e8247
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-500/trainer_state.json
@@ -0,0 +1,559 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.2051901025950513,
+  "eval_steps": 20,
+  "global_step": 500,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 4.911927193047245e+16,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-520/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-520/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-520/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-520/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-520/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-520/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-520/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-520/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-520/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-520/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-520/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-520/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-520/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-520/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..78b9aa19f90413387eeb00263fa6845c460482f1
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-520/trainer_state.json
@@ -0,0 +1,580 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.2534701267350634,
+  "eval_steps": 20,
+  "global_step": 520,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 5.114147376072499e+16,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-540/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-540/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-540/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-540/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-540/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-540/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-540/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-540/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-540/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-540/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-540/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-540/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-540/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-540/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..ae395697c51afa9ac346eaed8de4fb2f6dbd7c58
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-540/trainer_state.json
@@ -0,0 +1,601 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.3017501508750755,
+  "eval_steps": 20,
+  "global_step": 540,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 5.303253466631578e+16,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-560/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-560/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-560/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-560/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-560/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-560/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-560/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-560/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-560/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-560/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-560/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-560/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-560/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-560/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..2257cf2a5a7def11a8199885ce490e09ed6e37a1
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-560/trainer_state.json
@@ -0,0 +1,622 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.3500301750150876,
+  "eval_steps": 20,
+  "global_step": 560,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 5.499473623404749e+16,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-580/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-580/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-580/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-580/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-580/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-580/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-580/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-580/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-580/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-580/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-580/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-580/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-580/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-580/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..f1fc8c1e63ac62b33482d1bd2a36c4b00c43498b
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-580/trainer_state.json
@@ -0,0 +1,643 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.3983101991550995,
+  "eval_steps": 20,
+  "global_step": 580,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 5.669246296040448e+16,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-60/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-60/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-60/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-60/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-60/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-60/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-60/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-60/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-60/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-60/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-60/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-60/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-60/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-60/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..8c4bdf64a9e0c19d890d83819f8260cdd7fa01ca
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-60/trainer_state.json
@@ -0,0 +1,97 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.14484007242003621,
+  "eval_steps": 20,
+  "global_step": 60,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 5955464653602816.0,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-600/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-600/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-600/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-600/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-600/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-600/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-600/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-600/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-600/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-600/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-600/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-600/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-600/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-600/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..81d34d2880beb4477f908e688186a2c217b1b145
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-600/trainer_state.json
@@ -0,0 +1,664 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.4465902232951118,
+  "eval_steps": 20,
+  "global_step": 600,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 5.868799800731443e+16,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-620/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-620/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-620/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-620/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-620/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-620/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-620/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-620/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-620/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-620/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-620/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-620/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-620/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-620/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..6ec6bd6f87fcb72db191f80dcc638ca183af6e34
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-620/trainer_state.json
@@ -0,0 +1,685 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.4948702474351236,
+  "eval_steps": 20,
+  "global_step": 620,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 6.069125238624461e+16,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-640/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-640/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-640/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-640/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-640/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-640/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-640/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-640/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-640/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-640/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-640/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-640/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-640/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-640/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..a76c23296ecdb29b6853916df5dca6d8e4957315
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-640/trainer_state.json
@@ -0,0 +1,706 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.5431502715751357,
+  "eval_steps": 20,
+  "global_step": 640,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 6.270862963398451e+16,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-660/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-660/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-660/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-660/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-660/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-660/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-660/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-660/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-660/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-660/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-660/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-660/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-660/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-660/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..d586e72c495c90c3fd41e41b80265793bfb6b00a
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-660/trainer_state.json
@@ -0,0 +1,727 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.5914302957151478,
+  "eval_steps": 20,
+  "global_step": 660,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 6.456188335661261e+16,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-680/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-680/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-680/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-680/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-680/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-680/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-680/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-680/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-680/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-680/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-680/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-680/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-680/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-680/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..77b58bbe2b8d8f11b447c057dc02b2b411840a24
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-680/trainer_state.json
@@ -0,0 +1,748 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.63971031985516,
+  "eval_steps": 20,
+  "global_step": 680,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 6.649496198990438e+16,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-700/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-700/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-700/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-700/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-700/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-700/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-700/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-700/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-700/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-700/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-700/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-700/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-700/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-700/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..8965145e7ee8bc55f73478b59ec741e999cdd020
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-700/trainer_state.json
@@ -0,0 +1,769 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.687990343995172,
+  "eval_steps": 20,
+  "global_step": 700,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 6.850146199707034e+16,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-720/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-720/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-720/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-720/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-720/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-720/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-720/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-720/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-720/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-720/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-720/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-720/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-720/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-720/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..5c845d689c7be30745b8742cfdd4b9e4bf9a54c7
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-720/trainer_state.json
@@ -0,0 +1,790 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.736270368135184,
+  "eval_steps": 20,
+  "global_step": 720,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 7.0671910089984e+16,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-740/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-740/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-740/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-740/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-740/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-740/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-740/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-740/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-740/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-740/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-740/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-740/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-740/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-740/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..d9afda6e4a14bc65d58855e3f2398646b3dead2f
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-740/trainer_state.json
@@ -0,0 +1,811 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.7845503922751962,
+  "eval_steps": 20,
+  "global_step": 740,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 7.25684973355438e+16,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-760/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-760/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-760/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-760/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-760/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-760/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-760/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-760/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-760/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-760/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-760/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-760/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-760/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-760/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..9315ac75cbed0212ad995580a7ca62e17c871d38
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-760/trainer_state.json
@@ -0,0 +1,832 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.832830416415208,
+  "eval_steps": 20,
+  "global_step": 760,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 7.438140000442982e+16,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-780/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-780/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-780/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-780/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-780/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-780/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-780/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-780/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-780/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-780/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-780/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-780/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-780/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-780/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..72c630fe0416ab807e17d190d912f456c9a103ea
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-780/trainer_state.json
@@ -0,0 +1,853 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.8811104405552204,
+  "eval_steps": 20,
+  "global_step": 780,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 7.630070664764006e+16,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-80/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-80/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-80/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-80/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-80/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-80/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-80/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-80/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-80/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-80/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-80/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-80/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-80/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-80/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..0d60daf8a117ea26acefbfc13ecf0a374a566efd
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-80/trainer_state.json
@@ -0,0 +1,118 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.19312009656004828,
+  "eval_steps": 20,
+  "global_step": 80,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 7831350054199296.0,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-800/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-800/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-800/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-800/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-800/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-800/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-800/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-800/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-800/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-800/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-800/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-800/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-800/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-800/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..d320f1c12511584678692fff34cc1089f3ef7f21
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-800/trainer_state.json
@@ -0,0 +1,874 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.9293904646952322,
+  "eval_steps": 20,
+  "global_step": 800,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 7.837527712807526e+16,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-820/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-820/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-820/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-820/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-820/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-820/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-820/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-820/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-820/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-820/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-820/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-820/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-820/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-820/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..572182fe39f6aa01232f04fde5f1a5c21b57a3ba
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-820/trainer_state.json
@@ -0,0 +1,895 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.9776704888352445,
+  "eval_steps": 20,
+  "global_step": 820,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 8.023625018272358e+16,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-840/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-840/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-840/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-840/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-840/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-840/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-840/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-840/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-840/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-840/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-840/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-840/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-840/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-840/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..a654c75ed9d5c6f778989ac4936b74d15c689fdf
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-840/trainer_state.json
@@ -0,0 +1,916 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 2.024140012070006,
+  "eval_steps": 20,
+  "global_step": 840,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 8.22923995299287e+16,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-860/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-860/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-860/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-860/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-860/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-860/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-860/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-860/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-860/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-860/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-860/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-860/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-860/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-860/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..8829ebe513db3a620cb5ca7d2468bfd1310f78df
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-860/trainer_state.json
@@ -0,0 +1,937 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 2.0724200362100182,
+  "eval_steps": 20,
+  "global_step": 860,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 8.435240854314394e+16,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-880/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-880/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-880/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-880/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-880/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-880/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-880/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-880/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-880/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-880/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-880/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-880/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-880/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-880/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..46d0346f987385fa1e388a54558a355c1b462ce1
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-880/trainer_state.json
@@ -0,0 +1,958 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 2.12070006035003,
+  "eval_steps": 20,
+  "global_step": 880,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 8.62381185481298e+16,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-900/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-900/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-900/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-900/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-900/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-900/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-900/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-900/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-900/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-900/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-900/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-900/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-900/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-900/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..bb6ce4ec88df402addec1e7f1d8dd1ef4b59a76d
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-900/trainer_state.json
@@ -0,0 +1,979 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 2.1689800844900424,
+  "eval_steps": 20,
+  "global_step": 900,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 8.83278645335593e+16,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-920/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-920/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-920/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-920/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-920/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-920/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-920/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-920/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-920/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-920/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-920/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-920/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-920/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-920/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..26353a6a075892360676c2cbb35c83267422ab47
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-920/trainer_state.json
@@ -0,0 +1,1000 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 2.2172601086300543,
+  "eval_steps": 20,
+  "global_step": 920,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 9.051331269210317e+16,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-940/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-940/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-940/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-940/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-940/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-940/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-940/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-940/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-940/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-940/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-940/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-940/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-940/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-940/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..7307ea22dbcfbdcf38345bc9f1ad13100f94cc51
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-940/trainer_state.json
@@ -0,0 +1,1021 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 2.2655401327700666,
+  "eval_steps": 20,
+  "global_step": 940,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 9.228428535297024e+16,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-960/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-960/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-960/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-960/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-960/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-960/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-960/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-960/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-960/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-960/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-960/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-960/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-960/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-960/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..047f82ea637bfca91aae23266a9bc5e59153a7bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-960/trainer_state.json
@@ -0,0 +1,1042 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 2.3138201569100785,
+  "eval_steps": 20,
+  "global_step": 960,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 9.419350423274496e+16,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-980/README.md b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-980/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41e6c854e77830e9ea767c8c35f8c82a65c1ba35
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-980/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-4B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-4B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-980/adapter_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-980/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0dd029c541eac295dd0ee9608fbb57ff26a95c6c
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-980/adapter_config.json
@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.03828026524568501,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-980/chat_template.jinja b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-980/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..a585dec894e63da457d9440ec6aa7caa16d20860
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-980/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-980/tokenizer_config.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-980/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-980/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-980/trainer_state.json b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-980/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..759f439bb68fcef1f4f7fb246f45e4acee490f6a
--- /dev/null
+++ b/overgeneralisation_original_Estonian/Qwen3.5-4B-Base_overgeneralisation_splits_original_features_train_overgeneralisation_splits_original_features_test2/checkpoint-980/trainer_state.json
@@ -0,0 +1,1063 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 2.3621001810500903,
+  "eval_steps": 20,
+  "global_step": 980,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.172999547421932,
+      "epoch": 0.04828002414001207,
+      "grad_norm": 2.5071208477020264,
+      "learning_rate": 1.7227585487708594e-05,
+      "loss": 1.9347333908081055,
+      "mean_token_accuracy": 0.6086816020309925,
+      "num_tokens": 47778.0,
+      "step": 20
+    },
+    {
+      "epoch": 0.04828002414001207,
+      "eval_entropy": 1.6909754791956269,
+      "eval_loss": 1.4728385210037231,
+      "eval_mean_token_accuracy": 0.6773593797442619,
+      "eval_num_tokens": 47778.0,
+      "eval_runtime": 90.1265,
+      "eval_samples_per_second": 15.756,
+      "eval_steps_per_second": 1.975,
+      "step": 20
+    },
+    {
+      "entropy": 1.2014626950025558,
+      "epoch": 0.09656004828002414,
+      "grad_norm": 1.5074543952941895,
+      "learning_rate": 3.5361886001086065e-05,
+      "loss": 1.0867146492004394,
+      "mean_token_accuracy": 0.7285927519202232,
+      "num_tokens": 95981.0,
+      "step": 40
+    },
+    {
+      "epoch": 0.09656004828002414,
+      "eval_entropy": 1.0304168352250302,
+      "eval_loss": 0.9211422204971313,
+      "eval_mean_token_accuracy": 0.7613122848312507,
+      "eval_num_tokens": 95981.0,
+      "eval_runtime": 88.6338,
+      "eval_samples_per_second": 16.021,
+      "eval_steps_per_second": 2.008,
+      "step": 40
+    },
+    {
+      "entropy": 0.9454198732972146,
+      "epoch": 0.14484007242003621,
+      "grad_norm": 1.7578108310699463,
+      "learning_rate": 5.349618651446354e-05,
+      "loss": 0.8447297096252442,
+      "mean_token_accuracy": 0.7723424732685089,
+      "num_tokens": 142784.0,
+      "step": 60
+    },
+    {
+      "epoch": 0.14484007242003621,
+      "eval_entropy": 0.8647834474451086,
+      "eval_loss": 0.8240737318992615,
+      "eval_mean_token_accuracy": 0.7821227014064789,
+      "eval_num_tokens": 142784.0,
+      "eval_runtime": 88.0135,
+      "eval_samples_per_second": 16.134,
+      "eval_steps_per_second": 2.022,
+      "step": 60
+    },
+    {
+      "entropy": 0.8963550612330436,
+      "epoch": 0.19312009656004828,
+      "grad_norm": 1.1464931964874268,
+      "learning_rate": 7.1630487027841e-05,
+      "loss": 0.7979836463928223,
+      "mean_token_accuracy": 0.7822489373385906,
+      "num_tokens": 187338.0,
+      "step": 80
+    },
+    {
+      "epoch": 0.19312009656004828,
+      "eval_entropy": 0.8663296133614657,
+      "eval_loss": 0.7752988338470459,
+      "eval_mean_token_accuracy": 0.7885957012015782,
+      "eval_num_tokens": 187338.0,
+      "eval_runtime": 88.284,
+      "eval_samples_per_second": 16.084,
+      "eval_steps_per_second": 2.016,
+      "step": 80
+    },
+    {
+      "entropy": 0.8265103206038475,
+      "epoch": 0.24140012070006034,
+      "grad_norm": 1.0106662511825562,
+      "learning_rate": 8.976478754121848e-05,
+      "loss": 0.7297060012817382,
+      "mean_token_accuracy": 0.7955138415098191,
+      "num_tokens": 236038.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.24140012070006034,
+      "eval_entropy": 0.8122169536151244,
+      "eval_loss": 0.754906177520752,
+      "eval_mean_token_accuracy": 0.7917155255092664,
+      "eval_num_tokens": 236038.0,
+      "eval_runtime": 87.5754,
+      "eval_samples_per_second": 16.215,
+      "eval_steps_per_second": 2.033,
+      "step": 100
+    },
+    {
+      "entropy": 0.8098392844200134,
+      "epoch": 0.28968014484007243,
+      "grad_norm": 1.2238041162490845,
+      "learning_rate": 0.00010789908805459595,
+      "loss": 0.7201489448547364,
+      "mean_token_accuracy": 0.7975016921758652,
+      "num_tokens": 282918.0,
+      "step": 120
+    },
+    {
+      "epoch": 0.28968014484007243,
+      "eval_entropy": 0.8418726462326692,
+      "eval_loss": 0.7393178939819336,
+      "eval_mean_token_accuracy": 0.7936771585700217,
+      "eval_num_tokens": 282918.0,
+      "eval_runtime": 87.1597,
+      "eval_samples_per_second": 16.292,
+      "eval_steps_per_second": 2.042,
+      "step": 120
+    },
+    {
+      "entropy": 0.8080591082572937,
+      "epoch": 0.33796016898008446,
+      "grad_norm": 1.011591911315918,
+      "learning_rate": 0.00012603338856797343,
+      "loss": 0.7147697925567627,
+      "mean_token_accuracy": 0.7947175532579422,
+      "num_tokens": 331305.0,
+      "step": 140
+    },
+    {
+      "epoch": 0.33796016898008446,
+      "eval_entropy": 0.7894941431083037,
+      "eval_loss": 0.720219075679779,
+      "eval_mean_token_accuracy": 0.801899236239744,
+      "eval_num_tokens": 331305.0,
+      "eval_runtime": 87.1493,
+      "eval_samples_per_second": 16.294,
+      "eval_steps_per_second": 2.042,
+      "step": 140
+    },
+    {
+      "entropy": 0.7957205109298229,
+      "epoch": 0.38624019312009655,
+      "grad_norm": 0.8481830954551697,
+      "learning_rate": 0.00014416768908135088,
+      "loss": 0.7050027847290039,
+      "mean_token_accuracy": 0.7995368830859662,
+      "num_tokens": 375876.0,
+      "step": 160
+    },
+    {
+      "epoch": 0.38624019312009655,
+      "eval_entropy": 0.7746140591883928,
+      "eval_loss": 0.7175475358963013,
+      "eval_mean_token_accuracy": 0.7994848955213354,
+      "eval_num_tokens": 375876.0,
+      "eval_runtime": 86.655,
+      "eval_samples_per_second": 16.387,
+      "eval_steps_per_second": 2.054,
+      "step": 160
+    },
+    {
+      "entropy": 0.779565304517746,
+      "epoch": 0.43452021726010864,
+      "grad_norm": 1.032624363899231,
+      "learning_rate": 0.00016230198959472835,
+      "loss": 0.6978332042694092,
+      "mean_token_accuracy": 0.8015618294477462,
+      "num_tokens": 423913.0,
+      "step": 180
+    },
+    {
+      "epoch": 0.43452021726010864,
+      "eval_entropy": 0.7455583871080634,
+      "eval_loss": 0.7118850350379944,
+      "eval_mean_token_accuracy": 0.7999682195400923,
+      "eval_num_tokens": 423913.0,
+      "eval_runtime": 86.7842,
+      "eval_samples_per_second": 16.362,
+      "eval_steps_per_second": 2.051,
+      "step": 180
+    },
+    {
+      "entropy": 0.7810172818601131,
+      "epoch": 0.4828002414001207,
+      "grad_norm": 0.9276663064956665,
+      "learning_rate": 0.00018043629010810582,
+      "loss": 0.6937861442565918,
+      "mean_token_accuracy": 0.802893053740263,
+      "num_tokens": 472395.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.4828002414001207,
+      "eval_entropy": 0.7650193290764027,
+      "eval_loss": 0.7020567059516907,
+      "eval_mean_token_accuracy": 0.8045302153973097,
+      "eval_num_tokens": 472395.0,
+      "eval_runtime": 88.326,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 200
+    },
+    {
+      "entropy": 0.7862060204148292,
+      "epoch": 0.5310802655401328,
+      "grad_norm": 0.8195295929908752,
+      "learning_rate": 0.0001985705906214833,
+      "loss": 0.6878782272338867,
+      "mean_token_accuracy": 0.8027168907225132,
+      "num_tokens": 517049.0,
+      "step": 220
+    },
+    {
+      "epoch": 0.5310802655401328,
+      "eval_entropy": 0.7656565583154057,
+      "eval_loss": 0.6973932981491089,
+      "eval_mean_token_accuracy": 0.8047967278555538,
+      "eval_num_tokens": 517049.0,
+      "eval_runtime": 87.7698,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 220
+    },
+    {
+      "entropy": 0.7611534461379051,
+      "epoch": 0.5793602896801449,
+      "grad_norm": 0.7499712109565735,
+      "learning_rate": 0.00021670489113486077,
+      "loss": 0.681937837600708,
+      "mean_token_accuracy": 0.8041978515684605,
+      "num_tokens": 562836.0,
+      "step": 240
+    },
+    {
+      "epoch": 0.5793602896801449,
+      "eval_entropy": 0.7599442380197933,
+      "eval_loss": 0.7002046704292297,
+      "eval_mean_token_accuracy": 0.8051861517884759,
+      "eval_num_tokens": 562836.0,
+      "eval_runtime": 87.9239,
+      "eval_samples_per_second": 16.15,
+      "eval_steps_per_second": 2.024,
+      "step": 240
+    },
+    {
+      "entropy": 0.7800783462822437,
+      "epoch": 0.627640313820157,
+      "grad_norm": 0.8645269274711609,
+      "learning_rate": 0.00023483919164823825,
+      "loss": 0.6954500675201416,
+      "mean_token_accuracy": 0.8031484372913837,
+      "num_tokens": 606612.0,
+      "step": 260
+    },
+    {
+      "epoch": 0.627640313820157,
+      "eval_entropy": 0.7498679090751691,
+      "eval_loss": 0.7046149373054504,
+      "eval_mean_token_accuracy": 0.802388215667746,
+      "eval_num_tokens": 606612.0,
+      "eval_runtime": 88.2468,
+      "eval_samples_per_second": 16.091,
+      "eval_steps_per_second": 2.017,
+      "step": 260
+    },
+    {
+      "entropy": 0.7586533665657044,
+      "epoch": 0.6759203379601689,
+      "grad_norm": 0.741596519947052,
+      "learning_rate": 0.0002529734921616157,
+      "loss": 0.690476369857788,
+      "mean_token_accuracy": 0.8046923100948333,
+      "num_tokens": 655920.0,
+      "step": 280
+    },
+    {
+      "epoch": 0.6759203379601689,
+      "eval_entropy": 0.7714274066887544,
+      "eval_loss": 0.6992160677909851,
+      "eval_mean_token_accuracy": 0.8042646482419432,
+      "eval_num_tokens": 655920.0,
+      "eval_runtime": 88.1754,
+      "eval_samples_per_second": 16.104,
+      "eval_steps_per_second": 2.019,
+      "step": 280
+    },
+    {
+      "entropy": 0.7585848286747933,
+      "epoch": 0.724200362100181,
+      "grad_norm": 1.0778124332427979,
+      "learning_rate": 0.00027110779267499317,
+      "loss": 0.6790409564971924,
+      "mean_token_accuracy": 0.8060351841151714,
+      "num_tokens": 702638.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.724200362100181,
+      "eval_entropy": 0.7380402811457601,
+      "eval_loss": 0.7054820656776428,
+      "eval_mean_token_accuracy": 0.8003534738267406,
+      "eval_num_tokens": 702638.0,
+      "eval_runtime": 88.238,
+      "eval_samples_per_second": 16.093,
+      "eval_steps_per_second": 2.017,
+      "step": 300
+    },
+    {
+      "entropy": 0.7696981988847256,
+      "epoch": 0.7724803862401931,
+      "grad_norm": 0.8408658504486084,
+      "learning_rate": 0.00028924209318837064,
+      "loss": 0.7070387363433838,
+      "mean_token_accuracy": 0.8035947173833847,
+      "num_tokens": 749377.0,
+      "step": 320
+    },
+    {
+      "epoch": 0.7724803862401931,
+      "eval_entropy": 0.7897409023193831,
+      "eval_loss": 0.7015668749809265,
+      "eval_mean_token_accuracy": 0.805306687113944,
+      "eval_num_tokens": 749377.0,
+      "eval_runtime": 87.9395,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 320
+    },
+    {
+      "entropy": 0.762337189912796,
+      "epoch": 0.8207604103802052,
+      "grad_norm": 0.9843310117721558,
+      "learning_rate": 0.0003073763937017481,
+      "loss": 0.7021088123321533,
+      "mean_token_accuracy": 0.8029616877436638,
+      "num_tokens": 798874.0,
+      "step": 340
+    },
+    {
+      "epoch": 0.8207604103802052,
+      "eval_entropy": 0.7210077121016685,
+      "eval_loss": 0.7082746624946594,
+      "eval_mean_token_accuracy": 0.8049283824609906,
+      "eval_num_tokens": 798874.0,
+      "eval_runtime": 88.1706,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 340
+    },
+    {
+      "entropy": 0.7879025422036647,
+      "epoch": 0.8690404345202173,
+      "grad_norm": 0.9713814854621887,
+      "learning_rate": 0.0003255106942151256,
+      "loss": 0.714624547958374,
+      "mean_token_accuracy": 0.79987031519413,
+      "num_tokens": 840825.0,
+      "step": 360
+    },
+    {
+      "epoch": 0.8690404345202173,
+      "eval_entropy": 0.7866930489459735,
+      "eval_loss": 0.7145519852638245,
+      "eval_mean_token_accuracy": 0.8015657988157165,
+      "eval_num_tokens": 840825.0,
+      "eval_runtime": 88.3746,
+      "eval_samples_per_second": 16.068,
+      "eval_steps_per_second": 2.014,
+      "step": 360
+    },
+    {
+      "entropy": 0.7844726584851742,
+      "epoch": 0.9173204586602294,
+      "grad_norm": 1.2028311491012573,
+      "learning_rate": 0.00034364499472850306,
+      "loss": 0.7040786266326904,
+      "mean_token_accuracy": 0.7981885217130185,
+      "num_tokens": 882885.0,
+      "step": 380
+    },
+    {
+      "epoch": 0.9173204586602294,
+      "eval_entropy": 0.7767668339643585,
+      "eval_loss": 0.7073574066162109,
+      "eval_mean_token_accuracy": 0.8040851833445303,
+      "eval_num_tokens": 882885.0,
+      "eval_runtime": 88.3828,
+      "eval_samples_per_second": 16.066,
+      "eval_steps_per_second": 2.014,
+      "step": 380
+    },
+    {
+      "entropy": 0.7699790760874748,
+      "epoch": 0.9656004828002414,
+      "grad_norm": 0.9397398233413696,
+      "learning_rate": 0.00036177929524188054,
+      "loss": 0.7040313720703125,
+      "mean_token_accuracy": 0.8034727744758129,
+      "num_tokens": 927456.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.9656004828002414,
+      "eval_entropy": 0.7749103545472863,
+      "eval_loss": 0.7093513011932373,
+      "eval_mean_token_accuracy": 0.8019482901926791,
+      "eval_num_tokens": 927456.0,
+      "eval_runtime": 86.2112,
+      "eval_samples_per_second": 16.471,
+      "eval_steps_per_second": 2.065,
+      "step": 400
+    },
+    {
+      "entropy": 0.7672389172888422,
+      "epoch": 1.012070006035003,
+      "grad_norm": 0.99210524559021,
+      "learning_rate": 0.00037628567078152296,
+      "loss": 0.6913118362426758,
+      "mean_token_accuracy": 0.8042210944287189,
+      "num_tokens": 972662.0,
+      "step": 420
+    },
+    {
+      "epoch": 1.012070006035003,
+      "eval_entropy": 0.7294771229283193,
+      "eval_loss": 0.7171520590782166,
+      "eval_mean_token_accuracy": 0.8015128585059991,
+      "eval_num_tokens": 972662.0,
+      "eval_runtime": 88.037,
+      "eval_samples_per_second": 16.13,
+      "eval_steps_per_second": 2.022,
+      "step": 420
+    },
+    {
+      "entropy": 0.7143943183124065,
+      "epoch": 1.060350030175015,
+      "grad_norm": 1.087638258934021,
+      "learning_rate": 0.0003762484015601069,
+      "loss": 0.642704439163208,
+      "mean_token_accuracy": 0.8124966286122799,
+      "num_tokens": 1022525.0,
+      "step": 440
+    },
+    {
+      "epoch": 1.060350030175015,
+      "eval_entropy": 0.743511886744017,
+      "eval_loss": 0.7116662859916687,
+      "eval_mean_token_accuracy": 0.8037210672758939,
+      "eval_num_tokens": 1022525.0,
+      "eval_runtime": 87.9931,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 440
+    },
+    {
+      "entropy": 0.7237346574664116,
+      "epoch": 1.1086300543150271,
+      "grad_norm": 0.9365553259849548,
+      "learning_rate": 0.0003761579008392123,
+      "loss": 0.652644681930542,
+      "mean_token_accuracy": 0.8107496216893196,
+      "num_tokens": 1069194.0,
+      "step": 460
+    },
+    {
+      "epoch": 1.1086300543150271,
+      "eval_entropy": 0.7372823839776972,
+      "eval_loss": 0.7238303422927856,
+      "eval_mean_token_accuracy": 0.8026230124275336,
+      "eval_num_tokens": 1069194.0,
+      "eval_runtime": 87.6778,
+      "eval_samples_per_second": 16.196,
+      "eval_steps_per_second": 2.03,
+      "step": 460
+    },
+    {
+      "entropy": 0.7202137842774391,
+      "epoch": 1.1569100784550392,
+      "grad_norm": 0.7998443841934204,
+      "learning_rate": 0.0003760141942294726,
+      "loss": 0.6485219478607178,
+      "mean_token_accuracy": 0.8092714451253414,
+      "num_tokens": 1117570.0,
+      "step": 480
+    },
+    {
+      "epoch": 1.1569100784550392,
+      "eval_entropy": 0.71547647942318,
+      "eval_loss": 0.7124277949333191,
+      "eval_mean_token_accuracy": 0.8013174219077892,
+      "eval_num_tokens": 1117570.0,
+      "eval_runtime": 87.4387,
+      "eval_samples_per_second": 16.24,
+      "eval_steps_per_second": 2.036,
+      "step": 480
+    },
+    {
+      "entropy": 0.7270819500088692,
+      "epoch": 1.2051901025950513,
+      "grad_norm": 0.9470787644386292,
+      "learning_rate": 0.00037581732239815854,
+      "loss": 0.66352219581604,
+      "mean_token_accuracy": 0.8102593503892421,
+      "num_tokens": 1161717.0,
+      "step": 500
+    },
+    {
+      "epoch": 1.2051901025950513,
+      "eval_entropy": 0.7791565897759427,
+      "eval_loss": 0.7179726958274841,
+      "eval_mean_token_accuracy": 0.7948898328154275,
+      "eval_num_tokens": 1161717.0,
+      "eval_runtime": 87.9927,
+      "eval_samples_per_second": 16.138,
+      "eval_steps_per_second": 2.023,
+      "step": 500
+    },
+    {
+      "entropy": 0.7320936664938926,
+      "epoch": 1.2534701267350634,
+      "grad_norm": 1.0220379829406738,
+      "learning_rate": 0.0003755673410576695,
+      "loss": 0.6581549644470215,
+      "mean_token_accuracy": 0.8098709337413311,
+      "num_tokens": 1209390.0,
+      "step": 520
+    },
+    {
+      "epoch": 1.2534701267350634,
+      "eval_entropy": 0.7710678225822662,
+      "eval_loss": 0.708797812461853,
+      "eval_mean_token_accuracy": 0.8039572778042783,
+      "eval_num_tokens": 1209390.0,
+      "eval_runtime": 88.1193,
+      "eval_samples_per_second": 16.115,
+      "eval_steps_per_second": 2.02,
+      "step": 520
+    },
+    {
+      "entropy": 0.7281674664467573,
+      "epoch": 1.3017501508750755,
+      "grad_norm": 0.8775188326835632,
+      "learning_rate": 0.000375264320949768,
+      "loss": 0.6631278991699219,
+      "mean_token_accuracy": 0.8101261422038079,
+      "num_tokens": 1256611.0,
+      "step": 540
+    },
+    {
+      "epoch": 1.3017501508750755,
+      "eval_entropy": 0.7199799800187014,
+      "eval_loss": 0.6981531381607056,
+      "eval_mean_token_accuracy": 0.8071265254127845,
+      "eval_num_tokens": 1256611.0,
+      "eval_runtime": 88.165,
+      "eval_samples_per_second": 16.106,
+      "eval_steps_per_second": 2.019,
+      "step": 540
+    },
+    {
+      "entropy": 0.722860773652792,
+      "epoch": 1.3500301750150876,
+      "grad_norm": 0.8949288129806519,
+      "learning_rate": 0.00037490834782556,
+      "loss": 0.6639598846435547,
+      "mean_token_accuracy": 0.8105649061501026,
+      "num_tokens": 1302325.0,
+      "step": 560
+    },
+    {
+      "epoch": 1.3500301750150876,
+      "eval_entropy": 0.7134760032209118,
+      "eval_loss": 0.7029948234558105,
+      "eval_mean_token_accuracy": 0.8062985880991046,
+      "eval_num_tokens": 1302325.0,
+      "eval_runtime": 88.1286,
+      "eval_samples_per_second": 16.113,
+      "eval_steps_per_second": 2.02,
+      "step": 560
+    },
+    {
+      "entropy": 0.7299126699566841,
+      "epoch": 1.3983101991550995,
+      "grad_norm": 1.0323039293289185,
+      "learning_rate": 0.0003744995224212291,
+      "loss": 0.6535763263702392,
+      "mean_token_accuracy": 0.8120224848389626,
+      "num_tokens": 1343844.0,
+      "step": 580
+    },
+    {
+      "epoch": 1.3983101991550995,
+      "eval_entropy": 0.7169701182440426,
+      "eval_loss": 0.7067857980728149,
+      "eval_mean_token_accuracy": 0.8040432936689826,
+      "eval_num_tokens": 1343844.0,
+      "eval_runtime": 88.2252,
+      "eval_samples_per_second": 16.095,
+      "eval_steps_per_second": 2.018,
+      "step": 580
+    },
+    {
+      "entropy": 0.7281742256134749,
+      "epoch": 1.4465902232951118,
+      "grad_norm": 1.3370424509048462,
+      "learning_rate": 0.00037403796042952863,
+      "loss": 0.6637272357940673,
+      "mean_token_accuracy": 0.8118261776864528,
+      "num_tokens": 1389471.0,
+      "step": 600
+    },
+    {
+      "epoch": 1.4465902232951118,
+      "eval_entropy": 0.7378843212395572,
+      "eval_loss": 0.7038390040397644,
+      "eval_mean_token_accuracy": 0.8059229371922739,
+      "eval_num_tokens": 1389471.0,
+      "eval_runtime": 87.9301,
+      "eval_samples_per_second": 16.149,
+      "eval_steps_per_second": 2.024,
+      "step": 600
+    },
+    {
+      "entropy": 0.7388614989817143,
+      "epoch": 1.4948702474351236,
+      "grad_norm": 0.8464171290397644,
+      "learning_rate": 0.00037352379246704257,
+      "loss": 0.6604740142822265,
+      "mean_token_accuracy": 0.8102974124252796,
+      "num_tokens": 1435975.0,
+      "step": 620
+    },
+    {
+      "epoch": 1.4948702474351236,
+      "eval_entropy": 0.7126847046814607,
+      "eval_loss": 0.6950703263282776,
+      "eval_mean_token_accuracy": 0.8085885392815879,
+      "eval_num_tokens": 1435975.0,
+      "eval_runtime": 88.2732,
+      "eval_samples_per_second": 16.086,
+      "eval_steps_per_second": 2.016,
+      "step": 620
+    },
+    {
+      "entropy": 0.7258916139602661,
+      "epoch": 1.5431502715751357,
+      "grad_norm": 1.1780085563659668,
+      "learning_rate": 0.0003729571640372223,
+      "loss": 0.6529891014099121,
+      "mean_token_accuracy": 0.8085981778800487,
+      "num_tokens": 1482035.0,
+      "step": 640
+    },
+    {
+      "epoch": 1.5431502715751357,
+      "eval_entropy": 0.7096509638797032,
+      "eval_loss": 0.690051257610321,
+      "eval_mean_token_accuracy": 0.8100520241796301,
+      "eval_num_tokens": 1482035.0,
+      "eval_runtime": 88.3153,
+      "eval_samples_per_second": 16.079,
+      "eval_steps_per_second": 2.016,
+      "step": 640
+    },
+    {
+      "entropy": 0.7151173129677773,
+      "epoch": 1.5914302957151478,
+      "grad_norm": 1.0357667207717896,
+      "learning_rate": 0.0003723382354892108,
+      "loss": 0.6451474189758301,
+      "mean_token_accuracy": 0.8136902332305909,
+      "num_tokens": 1528119.0,
+      "step": 660
+    },
+    {
+      "epoch": 1.5914302957151478,
+      "eval_entropy": 0.7060252652409371,
+      "eval_loss": 0.6899483799934387,
+      "eval_mean_token_accuracy": 0.810834194502134,
+      "eval_num_tokens": 1528119.0,
+      "eval_runtime": 87.8848,
+      "eval_samples_per_second": 16.158,
+      "eval_steps_per_second": 2.025,
+      "step": 660
+    },
+    {
+      "entropy": 0.7141377851366997,
+      "epoch": 1.63971031985516,
+      "grad_norm": 0.8391667008399963,
+      "learning_rate": 0.000371667181972466,
+      "loss": 0.6481579780578614,
+      "mean_token_accuracy": 0.8159015104174614,
+      "num_tokens": 1573661.0,
+      "step": 680
+    },
+    {
+      "epoch": 1.63971031985516,
+      "eval_entropy": 0.7598993319473909,
+      "eval_loss": 0.681204080581665,
+      "eval_mean_token_accuracy": 0.8100194586126992,
+      "eval_num_tokens": 1573661.0,
+      "eval_runtime": 88.2415,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 680
+    },
+    {
+      "entropy": 0.7235980287194252,
+      "epoch": 1.687990343995172,
+      "grad_norm": 0.8197196125984192,
+      "learning_rate": 0.00037094419338719537,
+      "loss": 0.6582849025726318,
+      "mean_token_accuracy": 0.8114834539592266,
+      "num_tokens": 1621714.0,
+      "step": 700
+    },
+    {
+      "epoch": 1.687990343995172,
+      "eval_entropy": 0.7166854568411795,
+      "eval_loss": 0.679177463054657,
+      "eval_mean_token_accuracy": 0.8119039877077167,
+      "eval_num_tokens": 1621714.0,
+      "eval_runtime": 88.0957,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 2.021,
+      "step": 700
+    },
+    {
+      "entropy": 0.7278190158307553,
+      "epoch": 1.736270368135184,
+      "grad_norm": 0.7820602655410767,
+      "learning_rate": 0.0003701694743306164,
+      "loss": 0.6633153438568116,
+      "mean_token_accuracy": 0.8123249113559723,
+      "num_tokens": 1672118.0,
+      "step": 720
+    },
+    {
+      "epoch": 1.736270368135184,
+      "eval_entropy": 0.6980850314826108,
+      "eval_loss": 0.6775653958320618,
+      "eval_mean_token_accuracy": 0.8116872906684875,
+      "eval_num_tokens": 1672118.0,
+      "eval_runtime": 87.9795,
+      "eval_samples_per_second": 16.14,
+      "eval_steps_per_second": 2.023,
+      "step": 720
+    },
+    {
+      "entropy": 0.7149631556123495,
+      "epoch": 1.7845503922751962,
+      "grad_norm": 1.0139760971069336,
+      "learning_rate": 0.00036934324403905795,
+      "loss": 0.6468976020812989,
+      "mean_token_accuracy": 0.8117561548948288,
+      "num_tokens": 1720465.0,
+      "step": 740
+    },
+    {
+      "epoch": 1.7845503922751962,
+      "eval_entropy": 0.69009849094273,
+      "eval_loss": 0.6731093525886536,
+      "eval_mean_token_accuracy": 0.81281964196248,
+      "eval_num_tokens": 1720465.0,
+      "eval_runtime": 87.7412,
+      "eval_samples_per_second": 16.184,
+      "eval_steps_per_second": 2.029,
+      "step": 740
+    },
+    {
+      "entropy": 0.7226049453020096,
+      "epoch": 1.832830416415208,
+      "grad_norm": 1.5040546655654907,
+      "learning_rate": 0.0003684657363259193,
+      "loss": 0.6385328769683838,
+      "mean_token_accuracy": 0.8132639616727829,
+      "num_tokens": 1764440.0,
+      "step": 760
+    },
+    {
+      "epoch": 1.832830416415208,
+      "eval_entropy": 0.7125169618076153,
+      "eval_loss": 0.6864680647850037,
+      "eval_mean_token_accuracy": 0.8049418089095126,
+      "eval_num_tokens": 1764440.0,
+      "eval_runtime": 87.9429,
+      "eval_samples_per_second": 16.147,
+      "eval_steps_per_second": 2.024,
+      "step": 760
+    },
+    {
+      "entropy": 0.707821810990572,
+      "epoch": 1.8811104405552204,
+      "grad_norm": 0.8604223132133484,
+      "learning_rate": 0.00036753719951550327,
+      "loss": 0.6397049427032471,
+      "mean_token_accuracy": 0.8137877315282822,
+      "num_tokens": 1810828.0,
+      "step": 780
+    },
+    {
+      "epoch": 1.8811104405552204,
+      "eval_entropy": 0.6492102551326323,
+      "eval_loss": 0.6743726134300232,
+      "eval_mean_token_accuracy": 0.8130356572317273,
+      "eval_num_tokens": 1810828.0,
+      "eval_runtime": 87.8684,
+      "eval_samples_per_second": 16.161,
+      "eval_steps_per_second": 2.026,
+      "step": 780
+    },
+    {
+      "entropy": 0.6800355084240437,
+      "epoch": 1.9293904646952322,
+      "grad_norm": 0.7270982265472412,
+      "learning_rate": 0.00036655789637274377,
+      "loss": 0.6267755031585693,
+      "mean_token_accuracy": 0.8180945307016373,
+      "num_tokens": 1857600.0,
+      "step": 800
+    },
+    {
+      "epoch": 1.9293904646952322,
+      "eval_entropy": 0.6727538651294922,
+      "eval_loss": 0.6696028709411621,
+      "eval_mean_token_accuracy": 0.8116443090224534,
+      "eval_num_tokens": 1857600.0,
+      "eval_runtime": 88.2449,
+      "eval_samples_per_second": 16.092,
+      "eval_steps_per_second": 2.017,
+      "step": 800
+    },
+    {
+      "entropy": 0.6746691606938839,
+      "epoch": 1.9776704888352445,
+      "grad_norm": 0.9771540760993958,
+      "learning_rate": 0.0003655281040288461,
+      "loss": 0.6326992988586426,
+      "mean_token_accuracy": 0.8179976396262646,
+      "num_tokens": 1901998.0,
+      "step": 820
+    },
+    {
+      "epoch": 1.9776704888352445,
+      "eval_entropy": 0.6636319716324967,
+      "eval_loss": 0.6714363098144531,
+      "eval_mean_token_accuracy": 0.8126635943235976,
+      "eval_num_tokens": 1901998.0,
+      "eval_runtime": 87.7666,
+      "eval_samples_per_second": 16.179,
+      "eval_steps_per_second": 2.028,
+      "step": 820
+    },
+    {
+      "entropy": 0.6319125778489298,
+      "epoch": 2.024140012070006,
+      "grad_norm": 0.8380444645881653,
+      "learning_rate": 0.0003644481139028622,
+      "loss": 0.5780903816223144,
+      "mean_token_accuracy": 0.8250188401767186,
+      "num_tokens": 1947553.0,
+      "step": 840
+    },
+    {
+      "epoch": 2.024140012070006,
+      "eval_entropy": 0.6627151989535,
+      "eval_loss": 0.6774935722351074,
+      "eval_mean_token_accuracy": 0.8140510819601209,
+      "eval_num_tokens": 1947553.0,
+      "eval_runtime": 87.8287,
+      "eval_samples_per_second": 16.168,
+      "eval_steps_per_second": 2.027,
+      "step": 840
+    },
+    {
+      "entropy": 0.6059147048741579,
+      "epoch": 2.0724200362100182,
+      "grad_norm": 0.9745152592658997,
+      "learning_rate": 0.0003633182316192229,
+      "loss": 0.546702527999878,
+      "mean_token_accuracy": 0.8358186542987823,
+      "num_tokens": 1995234.0,
+      "step": 860
+    },
+    {
+      "epoch": 2.0724200362100182,
+      "eval_entropy": 0.6620710384979677,
+      "eval_loss": 0.6769291162490845,
+      "eval_mean_token_accuracy": 0.8135082835561773,
+      "eval_num_tokens": 1995234.0,
+      "eval_runtime": 88.0008,
+      "eval_samples_per_second": 16.136,
+      "eval_steps_per_second": 2.023,
+      "step": 860
+    },
+    {
+      "entropy": 0.6076117843389511,
+      "epoch": 2.12070006035003,
+      "grad_norm": 0.6914380192756653,
+      "learning_rate": 0.0003621387769212491,
+      "loss": 0.5324535846710206,
+      "mean_token_accuracy": 0.8366568259894848,
+      "num_tokens": 2040639.0,
+      "step": 880
+    },
+    {
+      "epoch": 2.12070006035003,
+      "eval_entropy": 0.6592224213514435,
+      "eval_loss": 0.670383632183075,
+      "eval_mean_token_accuracy": 0.8162150088320957,
+      "eval_num_tokens": 2040639.0,
+      "eval_runtime": 88.324,
+      "eval_samples_per_second": 16.077,
+      "eval_steps_per_second": 2.015,
+      "step": 880
+    },
+    {
+      "entropy": 0.6158072650432587,
+      "epoch": 2.1689800844900424,
+      "grad_norm": 0.9649198055267334,
+      "learning_rate": 0.0003609100835806688,
+      "loss": 0.5564568996429443,
+      "mean_token_accuracy": 0.8329654954373836,
+      "num_tokens": 2090604.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.1689800844900424,
+      "eval_entropy": 0.608588819758276,
+      "eval_loss": 0.6745102405548096,
+      "eval_mean_token_accuracy": 0.8155080542135774,
+      "eval_num_tokens": 2090604.0,
+      "eval_runtime": 87.52,
+      "eval_samples_per_second": 16.225,
+      "eval_steps_per_second": 2.034,
+      "step": 900
+    },
+    {
+      "entropy": 0.5938734326511621,
+      "epoch": 2.2172601086300543,
+      "grad_norm": 0.6915557384490967,
+      "learning_rate": 0.0003596324993031632,
+      "loss": 0.5351875782012939,
+      "mean_token_accuracy": 0.8385631121695042,
+      "num_tokens": 2142246.0,
+      "step": 920
+    },
+    {
+      "epoch": 2.2172601086300543,
+      "eval_entropy": 0.6251631205001574,
+      "eval_loss": 0.6664518117904663,
+      "eval_mean_token_accuracy": 0.81711940417129,
+      "eval_num_tokens": 2142246.0,
+      "eval_runtime": 87.1792,
+      "eval_samples_per_second": 16.288,
+      "eval_steps_per_second": 2.042,
+      "step": 920
+    },
+    {
+      "entropy": 0.623173613846302,
+      "epoch": 2.2655401327700666,
+      "grad_norm": 0.8366610407829285,
+      "learning_rate": 0.00035830638562997063,
+      "loss": 0.5488213539123535,
+      "mean_token_accuracy": 0.8321483485400677,
+      "num_tokens": 2183144.0,
+      "step": 940
+    },
+    {
+      "epoch": 2.2655401327700666,
+      "eval_entropy": 0.6237085649136747,
+      "eval_loss": 0.68075031042099,
+      "eval_mean_token_accuracy": 0.814671738429016,
+      "eval_num_tokens": 2183144.0,
+      "eval_runtime": 88.0652,
+      "eval_samples_per_second": 16.124,
+      "eval_steps_per_second": 2.021,
+      "step": 940
+    },
+    {
+      "entropy": 0.6236911740154027,
+      "epoch": 2.3138201569100785,
+      "grad_norm": 0.7812452912330627,
+      "learning_rate": 0.00035693211783557416,
+      "loss": 0.5696200847625732,
+      "mean_token_accuracy": 0.82991396561265,
+      "num_tokens": 2228098.0,
+      "step": 960
+    },
+    {
+      "epoch": 2.3138201569100785,
+      "eval_entropy": 0.6327033426319615,
+      "eval_loss": 0.6742984652519226,
+      "eval_mean_token_accuracy": 0.8148086091105857,
+      "eval_num_tokens": 2228098.0,
+      "eval_runtime": 88.1711,
+      "eval_samples_per_second": 16.105,
+      "eval_steps_per_second": 2.019,
+      "step": 960
+    },
+    {
+      "entropy": 0.6218426413834095,
+      "epoch": 2.3621001810500903,
+      "grad_norm": 0.8719790577888489,
+      "learning_rate": 0.0003555100848215035,
+      "loss": 0.5582651615142822,
+      "mean_token_accuracy": 0.8323961742222309,
+      "num_tokens": 2273062.0,
+      "step": 980
+    },
+    {
+      "epoch": 2.3621001810500903,
+      "eval_entropy": 0.6338012375858393,
+      "eval_loss": 0.6685822606086731,
+      "eval_mean_token_accuracy": 0.8152180300669724,
+      "eval_num_tokens": 2273062.0,
+      "eval_runtime": 88.101,
+      "eval_samples_per_second": 16.118,
+      "eval_steps_per_second": 2.02,
+      "step": 980
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4150,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 9.613105656982118e+16,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}